3D Assembly and Actuation of Nanopatterned Membranes using

Nanomagnets
by

Anthony John Nichol

B.S. Mechanical Engineering, The University of Wisconsin, Madison (2004)

Submitted to the Department of Mechanical Engineering

in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

February 2011

© Massachusetts Institute of Technology 2011. All rights reserved.

Author

Department of Mechanical Engineering

January 20, 2011

Certified by

George Barbastathis
Professor of Mechanical Engineering
Thesis Supervisor

Accepted by

David Hardt
Graduate Officer, Professor of Mechanical Engineering
 3D Assembly and Actuation of Nanopatterned Membranes using
Nanomagnets
By
Anthony J. Nichol
Submitted to the Department of Mechanical Engineering
on January 20, 2011, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy

Abstract
A new method for aligning and actuating membranes for 3D nano-assembly based on the
interactions of nanomagnets has been developed. Arrays of nanopatterned magnetic material are
integrated onto thin-film membranes. It is shown theoretically and experimentally that separate arrays of
nanomagnets attract and self-align when brought into close proximity. This provides a useful mechanism
for 3D nanofabrication because it allows higher positional tolerances as the nanomagnets provide the fine
alignment. The features on the membrane can first be patterned using traditional wafer and thin-film
processes with features such as nanophotonics as well as the nanomagnets. The membranes are then
folded or stacked bringing the magnets into coarse alignment. The magnets pull-in, align and bond the
membrane segments to create a highly aligned structure in 3D space. A theoretical framework was
developed for the magnetic and mechanical design elements of the system. A set of experiments based on
folding the membrane to bring the arrays into coarse alignment was designed requiring new mechanical
hinge elements at the micro-scale. Nanomagnets were shown to provide the folding actuation, as well as
alignment. It was shown that the nanomagnet arrays can provide better than 30nm self-alignment over a
100 micron membrane segment. It was also shown that the nanomagnets can provide reconfigurability, in
that the systems can controllably be shifted between more than one alignment position. Further steps
were taken to improve the alignment, increase the membrane size, integrate other nanofeatures and
improve the membrane processing.

Thesis Supervisor:
George Barbastathis, Professor, Department of Mechanical Engineering

Thesis Committee
Henry I. Smith, Professor, Department of Electrical Engineering
Caroline Ross, Professor, Department of Material Science and Engineering
Sang-Gook Kim, Professor, Department of Mechanical Engineering
3
Table of Contents

Acknow ledgm ents ......................................................................................................................... 6

List of acronyms............................................................................................................................ 8
Chapter 1: Introduction ............................................................................................................... 9
1.1 Thesis Outline ...................................................................................................................... 10
1.2 Applications of 3D nanotechnology ................................................................................. 13
1.3 Com parison of 3D nanofabrication techniques................................................................ 16
1.3.1 Lattice-based ............................................................................................................... 18
1.3.2 D irect laser write...................................................................................................... 20
1.3.3 Layer-by-layer............................................................................................................. 22
1.4 Mem branes for 3D nanofabrication.................................................................................. 26
1.4.1 M embrane processing ............................................................................................. 26
1.4.2 N anostructured Origam i actuation m echanism s ...................................................... 28
1.4.3 Required innovation and im provem ents ................................................................. 30
1.5 Survey of mem brane alignm ent m ethods ........................................................................ 32
1.5.1 Mechanical.................................................................................................................. 33
1.5.2 Surface tension............................................................................................................ 36
1.5.3 Magnetic or electrostatic......................................................................................... 36
Chapter 2: Theory and design of nanomagnet alignment.................................................... 38
2.1 M agnetic forces.................................................................................................................... 38
2.1.1 Force from external m agnetic field............................................................................. 41
2.1.2 Force betw een nanom agnets.................................................................................... 43
2.1.3 N anom agnet array design ........................................................................................ 53
2.2 M embrane mechanical forces ........................................................................................... 66
2.2.1 M embrane distortion................................................................................................ 67
2.2.2 M embrane ................................................................................................................... 70
2.2.3 Mem brane actuation and displacem ent.................................................................... 77
Chapter 3: Nanomagnet actuation and alignment experiments......................................... 81
3.1 Process design ...................................................................................................................... 81
3.1.1 M aterials ..................................................................................................................... 81
3.1.2 Fabrication steps ...................................................................................................... 88
3.2 M agnetic folding and hinge experim ents......................................................................... 98
3.2.1 M agnet array and hinge design............................................................................... 98
3.2.2 Experim ental results and analysis............................................................................. 102
3.3 M embrane alignm ent using nanom agnet arrays ................................................................ 104
3.3.1 Experim ental process ................................................................................................ 104
3.3.2 Results....................................................................................................................... 110
3.3.3 Analysis..................................................................................................................... 113
3.4 M embrane reconfiguration using nanom agnet arrays........................................................ 115
Chapter 4: Novel membrane processing for 3D nanofabrication ........................................ 120
4
 4.1 Dual-side m embrane patterning......................................................................................... 120
4.1.1 Resist spinning .......................................................................................................... 122
4.1.2 D ual-side electron beam lithography ........................................................................ 128
4.2 Nanomagnet embedding in ultra-compliant nanomembrane ............................................. 132
4.2.1 D esign ....................................................................................................................... 135
4.2.2 Process and results .................................................................................................... 137
Chapter 5: Future opportunities ............................................................................................. 140
5.1 Improvem ents to overall alignm ent ................................................................................... 141
5.1.1 Process control.......................................................................................................... 141
5.1.2 M agnet pattern ...................................................................................................... 145
5.2 Integration with stressed folding........................................................................................ 148
5.3 Stack-n-Snap 3D nanofabrication approach ...................................................................... 150
5.4 D ielectric m agnets or electrets........................................................................................... 153
C hapter 6: Conclusion.............................................................................................................. 161
Acknowledgements

There are a great number of people and organizations that I must thank for their guidance,

inspiration, assistance or funding of this research. First, the 3D Optical Systems Group under

Professor George Barbastathis has played a key roll in everything from concept generation to

fabrication. I originally joined the group because of the unique combination of my research
interests: optics and nanotechnology. I have not been disappointed. Although my research has

focused on nanofabrication, the optical side of the group has been a frequent partner in the drive

towards 3D optical systems at the nano-scale. During my tenure, the NanoStructured Origami
team of Shaar, Arora, In and Deterre were instrumental in brainstorming, fabrication experience,

and analysis. In particular, I must thank Will Arora for his groundwork on much of the

processes that were extended into nanomagnet alignment in this work.

The NanoStructures Laboratory at MIT has a unique approach to nano-fabrication that

enabled this research. Professor Henry I. Smith set up a lab that gives experimental flexibility

beyond that of typical microtechnology fabs. While still ensuring cleanliness, safety and proper

equipment, the atmosphere is one of "well, see what happens" and allowed me the kind of
freedom to do a lot of failed experiments, but also a few good ones. The staff of Jim Daley and

Mark Mondol was unbelievably helpful. I came to Jim dozens of times needing something in
short notice, and after a little faked irritation he would always help out. The NSL also has a

wealth of experienced researchers that helped in every step of the fabrication processing.

I have to give thanks to the organizations that funded this research. Part of the reason I

originally came to MIT was the offer of the Martin Fellowship from the Department of
Mechanical Engineering. This allowed me to pursue a research group and project that I was

passionate about. The research was further funded by the National Science Foundation Graduate

Research Fellowship. I must thank the MIT Institute for Soldier Nanotechnology for their

support of my work and much of the NanoStructured Origami research.

 Before coming to MIT, I co-founded a small technology start-up with two friends, Mike
Casper and Shawn Pucylowski. Although you always hear to never get into business with

friends, these two individuals have been incredible. They have understood my limitations, while
they slowly developed the company into something "real." In 2009-2010 I took a year off from

my graduate research to work with the company in Chicago. Although the business was hectic

when I came back to MIT, I must thank them and the investors for understanding the importance
of me finishing this thesis work.

While at the University of Wisconsin, Professor Vadim Shapiro, played an instrumental

role in putting me on the path to Ph.D. Before he took me under his wing, I had very little idea
of this world of engineering research. Graduate school was definitely not on the radar. He

helped me see that I had a potential to do something truly impactful and there are forms of

payment more valuable than the money.

I must thank my friends and family for all of their support. There is no way this could

have happened without the love and guidance of my parents. My father has always shown me an

incredible work ethic and drive. He made sacrifices for the family that I will forever appreciate.

My mother is definitely my biggest cheerleader and has always nurtured my creative side. They
have together built a secure foundation, which allowed me to take chances and fully develop my

abilities. Finally, I would like to dedicate this work to my grandma, "Grannie Annie". She is an

intelligent and curious woman and (although she probably does not really get it) always asks

about my research, affectionately calling me "The Nano Man." I believe that in a different time
and place she would have been a wonderful professor.
List of Acronyms

2D Two-dimensional
3D Three-dimensional
COMET Compliant Mechanism Tool
DI De-ionized
DLW Direct laser write
FEMM Finite Element Magnetic Modeling
MEMS Micro-electromechanical systems
NEMS Nano-electromechanical systems
OOMMF Object Oriented Micro Magnetics Framework
RIE Reactive ion etch
SEBL Scanning electron beam lithography
SEM Scanning electron microscopy
SPLEBL Scanning phase locked electron beam lithography
Chapter 1

Introduction

In a three-dimensional world, microtechnology fabrication has traditionally been limited

to two-dimensional processes. The core steps of patterning, material addition and material

removal all have been optimized for processing a flat wafer surface. The 2D approach has

become standard partly because of its early history in integrated circuits. With that came

substantial industrial resources for tool and process development. However, beyond the history,

there are fundamental reasons for the 2D approach. There are techniques to write arbitrary 3D

nanostructures into a polymer, but the means to transfer that into a semiconductor, metal, high

index dielectric, bio-material, or combination thereof is anything but trivial. For these and other

reasons, most 3D nanofabrication will likely follow a layer-by-layer approach.

The question is how to get to dozens, hundreds, or even thousands of layers without

sacrificing alignment, yield and throughput? The current solution of planarize, pattern, material

process and repeat is limited by alignment and material/process compatibility. For example, if a

top layer requires high temperature deposition, it could damage more temperature sensitive

materials in the bottom layers. Nanomembranes, or suspended sub-micron thin films, offer an

elegant solution to these problems. The technology is largely based on early efforts for thin X-

ray lithography masks, where it was shown that these membranes have remarkable tolerance for

nano-processing. By simply processing using the standard 2D tools and then stacking

nanomembranes, there is no need for planarization and non-compatible materials can be

9
processed separately. The nanomembrane approach does introduce other problems that must be

overcome, which was the driving force of this research. Although remarkably robust,

nanomembranes introduce distortion and require changes to standard wafer processes based on

sensitivity to stresses and fracture. Furthermore, the alignment of the stacking process is a

critical necessity. A solution to the alignment problem has been developed and proven

experimentally in this work. The disadvantage of high membrane compliance has been utilized

to create a self-aligning strategy based patterning powerful nanomagnets. When the membranes

are brought into close proximity, the nanomangets attract and align the layered membranes.

Membrane stacking with self-alignment to better than 1Onm will be a leading contender for 3D

nanofabrication of next generation nanophotonics, quantum devices, microelectronics, 3D

biotech, meta-materials and other future technologies.

1.1 Thesis Outline

This document is organized as follows. Chapter 1 presents an overview of the motivation of

this research and provides a survey of the state of the art in this field. The numerous applications

of 3D nanotechnology ranging from photonic crystals to 3D biotech are given in Section 1.2.

This is followed by an overview in Section 1.3 of the current methods used for 3D

nanofabrication of lattice-based (1.3.2), direct laser write (1.3.3) and layer-by-layer (1.3.4)

approaches. Section 1.4 presents the membrane approach for layer-by-layer assembly in more

detail, with an overview of membrane processing in 1.4.1, a summary of the NanoStructured

Origami folding approach in 1.4.2 and the improvements over prior work that are required in

Section 1.4.3. Finally, an overview of passive alignment methods for membrane stacking is

shown in Section 1.5. These can be separated into mechanical (1.5.1), surface tension (1.5.2)

10
and the magnetic or electrostatic approach developed in this research. The approaches are

contrasted showing the necessity for the nanomagnet alignment strategy.

Chapter 2 shows the design, modeling and simulation that was necessary for realizing the

nanomagnet alignment strategy. The magnetic forces involved in the process are presented in

Section 2.1 and can be separated into force and torque from an external field (2.1.1) and the

forces between nanomagnets (2.1.2). This is continued to analyze the interaction and design of

nanomagnet arrays in (2.1.3). The mechanical elements for the membrane folding, stacking and

alignment are shown in Section 2.2. These can be separated into the distortions from the

magnets and membrane stress 2.2.1, the modeling of the folding about a micro-fabricated hinge

2.2.2 and the displacement compliance in Section 2.2.3. Folding was the most critical design

step and a new type of torsional folding hinge developed specifically for this research is shown

in 2.2.2.

The core of the experimental work is presented in Chapter 3. Section 3.1 overviews the

process design, including a number of new membrane fabrication improvements. The material

selection is given in 3.1.1 and the membrane fabrication, including a new dry release strategy for

fragile membrane structures, are given in 3.1.2. The magnetic folding and hinge experiments are

presented in 3.2. The core experimental design taking into account the material/process

limitations is presented in 3.2.1, with particular focus on membrane distortion and folding.

Section 3.2.3 gives the experimental results and analysis of the new nanomagnet folding strategy

using the ultra-compliant hinge design. After proving the folding method, Section 3.3 gives the

core proof of concept results of the magnetic alignment strategy. The experimental procedure is

given in 3.3.1 for the two rounds of experiments, which including a proof of concept and a

secondary experiment to explore alignment improvements. The results of each of these

11
experiments are given in Section 3.3.2 including better than 30nm self-alignment between optical

features. This is followed by an analysis of the data in 3.3.3. Finally, it is shown experimentally

in Section 3.4 that the nanomagnet approach can be used for reconfigurable membrane systems,

since the alignment of the layers can be shifted by the magnets' internal states.

Chapter 4 introduces two additional membrane processes invented to improve the

nanomagnet alignment method. The first method (4.1) is dual-side patterning of suspended

membranes, which enables magnets or other features to be patterned on both sides of the

membrane. The experimental work on spinning PMMA resist onto the backside of the

membrane is given in 4.1.1 and the modeling and experimental results of the scanning electron

beam scattering are shown in 4.1.2. Contrary to the rigid membranes used in the proof of

concept experiment, Section 4.2 presents a new strategy based on an ultra-compliant membrane

that is distorted from the alignment forces. The design and basic modeling is shown in 4.2.1 and

the experimental results of showing nanomagnets embedded in an 100nm thick PMMA

membrane is shown in 4.2.2.

Chapter 5 suggests future work to further develop the nanomagnet alignment method into a

viable bulk nanofabrication approach. Improvements to the alignment based on the alignment

models and experimental results are shown in 5.1. Methods to better utilize the external

magnetic field and increase the lithographic accuracy are given in 5.1.1. Processes for the key

improvement of embedding the magnets within the membrane are given in 5.1.2. In order to

realize three dimensional structures, it is suggested that the nanomagnet alignment approach is

combined with stress-based membrane folding (5.2) or membrane stacking (5.3). The "Stack-N-

Snap" approach shown in Section 5.3 combines many of the important properties of the magnetic

alignment for industrial production of 3D nanostructures. Section 5.4 presents preliminary work

12
on replacing the nanomagnets with dielectric "nano-electrets" to provide greater material

selection and remove the requirement of a magnetic material. It is shown that nanomagnet

alignment theory and design primarily holds for the electrets. Initial designs and experimental

suggestions are provided. Chapter 6 concludes the thesis, with a statement on the contributions

of this research project.

1.2 Applications of 3D nanotechnology

There are numerous realized and envisioned applications of 3D nanofabrication. The

applications utilize one or more advantages of moving to the third dimension, including: higher

feature density, electrical/optical/fluidic pathways, improved interfaces, novel material

properties, 3D encapsulation, biomimetry, and a host of other new functionalities. Figure 1-1

shows a range of applications of 3D nanofabrication in the fields of photonics, electronics,

biology, environment, materials and energy. Early work in 3D nanofabrication was based on

dispersed nanoparticles in a matrix for improved material performance. This has moved towards

more deterministic patterning of nanofeatures in 3D space. One of the most promising

applications in 3D nanophotonics is the 3D photonic crystal, which includes the periodic

placement of features at sizes in hundreds of nanometers. One and two dimensional photonic

crystals have been commercialized in the form of thin film optics and photonic-crystal fibers, but

the third dimension offers significant additional functionality. The 3D arrangement of features

near or below the wavelength of light enables improved manipulation of photons for devices in

communication, computing, sensing, nano-fabrication and vision. The biggest challenge in

fabricating these structures include: accurate patterning and placement of nano-scaled features

and a larger selection of high index and photo-dynamic material choices.

13
 Biology Photonics Electronics

Micro-probes Photonic crystal 3D Memory

Materials Energy Environment

oo
Nanotube composite Origami Super capacitor Catalytic Air filter

Figure 1-1: Overview of applications of 3D nanofabrication with examples of Micro-probes',

photonic crystal 2, 3D Memory 3, nanotube composite origami super-capacitor 5 and catalytic
air filter6

Many devices in micro-electronics are already nearly three dimensional in structural form.

From the transistor to the layering of circuitry, the common integrated circuit is essentially 2.5D.

The 3 rd dimension essentially adds more logic, memory and circuit density making the device

faster or higher capacity. The micro-patterning has been an extension of 2D lithographic and

material processes, so there are limitations to moving too extensively into 3D. Other

applications include 3D memory systems 3 and hybridizing systems that combine electronics,

photonics or biological systems.

Biological systems have inherently three-dimensional properties. From the base molecules

and proteins to cell matrices and organ systems, the body operates in the third dimension.
14
Therefore, many 3D micro and nano-technologies must at least partially operate in the third

dimension. A very promising application of microtechnology is in cell matrix growth for tissue

replacement in which a 3D scaffolding of a biocompatible material, such as collagen is first

structured and then seeded with cells, which reproduce into the matrix. Nanostructuring in 3D is

particularly important in sub-cellular applications such as cell-therapy, cancer therapy, drug

development and molecular sensing. Other applications of 3D micro-scaled devices are for

micro-probes, lab on a chip with 3D filtration or cell sorting, and 3D cell capturing systems.

When one thinks about solving our most macro-scaled environmental problems, we do not

normally think to nanotechnology as an option. Climate change, water and air pollution,

deforestation and species decline are, of course, greatly tied to population growth and industrial

causes. However, most of these problems occur near the molecular and particulate level. Three-

dimensional nanotechnology provides a great platform for chemical/particle sensing, filtering,

sorting and signaling. A 3D matrix of particle filters has shown to efficiently separate

problematic particles from others. Whereas photonic, electronic, and biological 3D

nanotechnologies may require precise feature nanometer-scaled feature patterning, the biggest

hurdle to overcome is production in the massive volumes necessary for a true impact on the

environment.

The use of 3D nanofabrication in materials has traditionally been the dispersion of nano-

sized particles or fibers into some binding matrix 6 . Materials can be made with enhanced

strength, conductivity, magnetism or combinations thereof. Further anisotropic properties can be

controlled by the dispersion and orientation of the nano-features. Materials can be enhanced

further with the accurate placement of nano-features by setting up an interconnected matrix or

structure, particularly for anisotropic or hybridized "meta-materials".

15
 Finally, 3D nanofabrication is well positioned for use in energy generation and storage

because of the improvement in surface area and volume for a given footprint. For example,

many MEMS applications such as MEMS-robotics require on-board energy because they cannot

be tied to an outside power source. There are means to power these devices via harvesting

vibrational or electromagnetic energy but typically this energy must be stored. The layered

supercapacitor made by NanoStructured Origami 5 is an example of solution for storing this

energy in a compact form. On the larger scale, the hydrocarbon energy infrastructure will need

to be replaced in the coming years. The prime candidates of hydrogen, wind, nuclear, solar and

combinations thereof require unique innovations in energy transport and storage as well as more

efficient energy conversion as possibly enabled with 3D nanofabrication.

1.3 Comparison of 3D Nanofabrication Techniques

As seen in the range of applications presented, there is a great incentive to develop 3D

nanofabrication tools for both near and long-term applications. Nanofabrication in the third

dimension offers unique problems because it is below the size scale used in macro and micro-

machining processes but larger than most chemical processing of molecules and materials. The

majority of micro- and nano-fabrication tools developed over the last fifty years have focused on

2D lithography for fabrication of 2D and 2.5D assemblies. The prominent nanofabrication

winners have been photon, electron and imprint lithography for pattern creation and a range of

etching, implantation and deposition processes for material manipulation.

 Lattice-based (3D) Serial (1D)
photopolymerizable
~ ~44 ~ resin

glass cell

He-Cd laser

Self-assembly Holographic Direct laser write

Layering (2D)

Lithographic Membrane stacking Membrane origami

Figure 1-2: Overview of methods of 3D nanofabrication. The lattice-based processes

primarily use colloidal self-assembly or holographic patterning . The serial approach
of direct laser writing9 is 1D. The layering approaches of lithographic 0 , membrane
stacking" and membrane origami1

In order to nanofabricate in the third dimension a range of approaches can be used.

These can generally be characterized as ID, 2D and 3D building approaches. Figure 1-2 and

Table 1-1 give an overview of the techniques used in recent research. Each method has its own

benefits and limitations related to resolution, material selection, throughput, pattern accuracy,

pattern flexibility and size range. For example, two-photon methods based on 3D scanning of a

focused laser spot to expose a photo-active polymer has great pattern flexibility and accuracy but

is limited in resolution, material selection and throughput. On the other hand, 3D holographic

techniques based on interfering coherent light give great throughput and pattern accuracy but is

limited by resolution, material selection and pattern flexibility (e.g. only periodic structures).
Sections 1.3.2 to 1.3.5 explain the various approaches and compare them to show that a layering

approach holds great promise for 3D nanofabrication.

1.3.1 Lattice-based

Figure 1-2 shows the most common approaches to lattice-based 3D nanofabrication. In

colloidal assembly, particles typically scaled 1Os to 100s of nanometers are assembled in a lattice

based on their size and shape distribution. The assembly typically occurs as the solution is

evaporated or the sample is pulled from a solution and the surface-tension pulls the particles into

a tight assembly. If the particle size and process is well-controlled, one or more layers of lattice-

based structures can be assembled quickly and with very small lattice sizes. Similar to atomic

arrangements in many materials, there may be crystal boundaries and other defects as the

particles settle. The colloids are typically plastic' 3 or glass' 1,and other materials can be used to

fill in the matrix and then etch away the colloid leaving a pattern reversal". Colloidal assembly

offers high throughput and small feature size but lacks deterministic patterning, although

techniques can be used in very specialized cases as discussed below.

A comparable lattice-based approach is the holographic or interferemetric 3D nano-

patterning of a photoresist slab. Interference lithography for two-dimensional microstructure

arrays is a common technique based on interfering light from two or more coherent beams

leaving standing waves that expose the resist". The same effect using three or more beams can

create standing waves in 3D16 . Instead of separate beams, 2D gratings or holographic masks can

be used to simplify the set-up with similar results". Overall, holographic patterning sees many

of the same advantages and disadvantages of colloidal assembly. The patterning is high

throughput but is limited to lattice-based patterns and a limited material choice. Colloidal self-

18
assembly may be able to provide better resolution, but currently holographic gives more design

freedom (although still lattice-based).

Since the main limitation of the lattice approaches is the requirement of a periodic structure,

there have been some techniques explored to add local defects to a lattice structure. One method

is to combine holographic 3D lithography with multi-photon or photolithography. In a similar

technique, focused ion beams can be used to mill structures into periodic lattices17 . Finally, the

crystal of the colloidal assembly can be affected locally by using mechanical or local fields as is

shown in template self-assembly. Overall, these techniques may work for a few specific

applications such as waveguide channels in photonic crystals, but are still limited by material

selection, resolution, assembly size and potentially throughput.

 Table 1-1: Survey of recent 3D nanofabrication results
I I I-I I I
M inof
Accuracy
Strategy / Area Layers Materials Aperiodic Author
Feature
size

Direct laser writing 400nm lOum 6 CdS-polymer Yes Sun

nanocomposite
Direct laser writing 200nm 80um 16, 40 Photoresist Yes Deubel
Point-based Nd:YAG
Direct laser writing 250nm 35 Transparent Yes Rodenas
Ceramic
Direct laser writing 1000nm 50um 16 SU-8 Yes Seet

Membrane stacking 50nm 25um 20 Semiconductor Yes Aoki

___________________ _______GaAs, InP, Ys Ak
Layered Optical Contact
Layered Lithography 500nm 2000um 4 NR7, SU Yes Yao
Two-photon layer-by-layer 500nm 25um 3 PS, SU8 Yes Kim
E-beam layer by layer with
planarization 5Onm 25um 4 SiO2/polymer Yes Wang
Self-assembly of silica
nanosphe res at liquid 330nm Sum 5 Silica No He
interfaces
4 beam Laser interference
wopl
Volmeticwoodpile
Volumetric 3um 20+ Polymer No Shoji
Grating holographic optical 0.4um 4cm 20 Photoresist No Zhang
element
Multi-beam interference lum 5cm 150 Photoresist No Shoji
2D diffractive optical element 500nm 25+ SU8 No Xu
Colloidal silica or PS opals
followed by atomic layer 233nm cm 25+ Silica, ZnS/TiO2, No Summers
deposition Mn' Nm

Other 2D Patterning with

O Passivation 300nm 5cm 2 Si Yes Venkataraman

1.3.2 Direct Laser Write

Direct laser writing (DLW) utilizes non-linear optical effects to pattern small voxels in 3D

space in a photoresist. Instead of single photon absorption, which is very linear with optical

energy, DLW utilizes a focused laser spot. The spot increases the probability that two or more

photons are absorbed by the atoms allowing the electrons to overcome a higher energy barrier.

This requires a significantly higher optical energy, but allows a roughly diffraction limited or

20
slightly better spot. The method typically starts with a slab of photoresist such as SU-8, which is

scanned in 3D space with focused spot. A typical slab thickness would be 20 microns and up to

100s of microns in size with voxels on the order of 250nm. DLW can be valuable for

prototyping 3D photonic and mechanical structures and has been critical for testing new designs

in photonic crystals 8 .

The primary benefit to direct laser writing is to be able to controllably place nano-scaled

features in 3D space. The technique is not limited to periodic structures, but rather only

structurally secure and open-flow forms that allow development and removal of the photoresist.

In other words, there cannot be positive or negative "holes" in the desired structure or negligible

channels or structural arms in the final form. The thickness of the resist slab is largely limited to

optical artifacts (e.g. scattering or absorption) in the photoresist, but thicknesses up to 100

microns can be achieved. The overall area is limited to the sample scanning system so

theoretically very large areas in centimeters can be achieved.

There are two main drawbacks to direct laser writing. First, it is a sequential process, so

macroscopic volumes can take days to pattern. Secondly, the material is limited to special

polymers, glass or polymers with nanoparticles dispersed in them, which greatly limits the

method for most electronic, biological, and photonic applications. There are refilling strategies

that allows one to fill voids with additional materials, such as metals, but this is clearly limited in

the extent it can be used. There is also appreciable shrinkage in most polymers upon exposure

and development, which further limits material selection and overall feature accuracy. Finally,

although direct laser writing has shown features sizes in the lateral direction down to 30nm in

special situations due to non-linear optical effects, the resolution is still limited by the diffraction

of light- typically in 100s of nanometers.

 1.3.3 Layer-by-layer

In the last fifteen years, the multitude of two-dimensional nanopattering tools developed for

micro and nano-technology have started to be translated to 3D nanomanufacturing. These tools,

originally developed for microelectronics, allow for large area patterning of 2D patterns with

resolutions deep below optical wavelengths. Optical projection lithography of a mask written

with electron lithography has been the industry standard for decades. Other patterning tools

include scanning ion, scanning photon, interference, and nano-imprint lithography. There

additionally is a multitude of material processes that are optimized for 2D processing. The

additive processes include: material evaporation, chemical deposition, electro-plating, spinning,

crystal/nanotube growth, among other material-specific processes. There are a multitude of wet

and dry subtractive processes, which selectively remove materials in isotropic or anisotropic

fashion.

With the decades of research and development, one would think that the 2D lithographic

methods could easily be layered and extended into 3D space. This is true to a limited number of

layers, but as multiple layers are used alignment errors build up. Another problem is most of the

processes require a 2D planar or nearly-planar substrate. This requires a complicated strategy of

planarization sometimes using chemical-mechanical polishing. In the optical projection

systems, the alignment is typically specified to be better than 115th of the lithographic resolution.

Nanoimprint lithography has shown to have alignment over large areas of 1Os of nanometers.

Therefore, even if layer-to-layer alignment accuracy is better than 10 nanometers, there is an

error build up as you build to N layers for a 3D system. Nanoimprint and contact lithography

further suffer from substrate, mask, or e-beam patterning distortions, resulting in local patterning
misalignments. Finally, the build-up of layers creates material and process compatibility

problems since the bottom layers are affected by process parameters of the upper layers. A

simple limitation is the lower layers must be able to withstand high temperature if one or more of

the top layers is deposited in a high temperature process. Therefore, unless there are significant

industrial drivers, the layer-by-layer lithography approach appears to be most useful for

applications with less than 20 layers, planar in form and/or able to withstand a looser alignment

tolerance.

Another method that uses many of the same advantages of two-dimensional nano-patterning,

but may be more apt for fully 3D devices is nanomembrane manipulation. The most simple of

these approaches is membrane stacking as first shown by Aoki' 1 and later by Patel19 in large area

form. The methods involve membrane fabrication, patterning and processing using tools

developed for 2D micro-technology. As a final step, the membranes are aligned, stacked and

bonded one-by-one to create a 3D structure. The membranes range in materials from polymers

to semiconductors and can have a multi-layer structure with multiple materials.

The technique developed by Aoki et al. uses a micro-manipulation technique, which is a

modified scanning electron beam system with multiple nano-probes on precision stages for

moving, aligning and even fracturing small membranes- typically on the order of 25 microns in

size. They have demonstrated up to 20 layers stacked with alignment better than 50nm using the

micron-manipulation and mechanical alignment features 20 . This is a very promising technique,

and they have shown novel photonic crystal arrangements as well as cavity defects. The main

limitations appear to be membrane size and throughput, but this will continue to be an important

tool for prototyping and possibly small volume industrial production. The approach used by

Patel was largely based on research on X-ray lithography using silicon nitride masks. A silicon

23
nitride film in multiple millimeters dimension and multiple hundreds of nanometers thickness

were nanopatterned using interference lithography. After patterning, the membranes are stacked

one-by-one. The difficult elements have proven to be controlled release of the membrane and

particle control. No alignment measurements were taken.

Another form of 3D assembly using membranes has been demonstrated in the

Nanostructured Origamil method. Membranes are first patterned and processed using standard

2D tools. Features such as nanophotonics are patterned on the membrane as well as prescribed

hinges. The hinges are either stressed portions of the membrane that curl and cause folding or

weaker portions that an external force folding the membranes can overcome. Section 1.4.3

details Nanostructured Origami.

Layer-by-layer and membrane folding hold four primary benefits when compared to other 3D

techniques. This is illustrated in Figures 1-3 and 1-4. First, the features can be the highest

resolution possible since the tools are the same as 2D nanopatterning. Features in the Is or 1Os

of nanometers are possible with various methods such as nanoimprint lithography. Secondly, for

the same reason, layer-by-layer has the highest theoretical throughput. This strictly considers the

feature patterning throughput and ignores the layering or stacking delay. The resolution and

throughput are shown in comparison to state-of-the-art patterning in Figure 1-3 based on the best

case performance found in the survey. Figure 1-4 shows the other two advantages. Unlike

lattice-based processes, non-periodic structures can be easily patterned within the volume. The

layer-by-layer approach shares this benefit with direct laser writing. However, unlike all other

3D processes, there is significantly more freedom for material choices.

 0L.n- -I

rII% I
I I I
OSM%
| |
E
C
%MWO Direct
0 Holo-
0-
Ln SLaser graphic
Writing
*J
M I I II
0
IJ.
Ln
rq
Layering
Colloidal
I I I I

102 103 104 10s 106 10 108 109 1010 1011 112

Patterning speed (voxels/sec)

Figure 1-3: Comparison of the various 3D)nanofabrication tools for the estimated
throughput and feature resolution.
 Material choice
Polymer and simple Semiconductor,
material fill ceramic, metal, etc.

i
I
Holo-
M
I graphic

2--
.

0. Direct
0
Laser Layering
U

Writing

Figure 1-4: Comparison of the various 3D nanofabrication tools for material

choice and the ability to pattern non-periodic structures

1.4 Membranes for 3D technology

This section overviews some of the processing and required improvements to make

nanomembrane stacking or folding a mainstream 3D nanoprocessing approach.

1.4.1 Membrane processing

Nanomembranes are based on thin films that are processed to be free-standing. They

typically start from a silicon wafer and one or more thin films from 25 nanometers to 2 microns

thick are deposited on the surface. There is then two different approaches to processing the thin
film. The first, more standard approach involves the film being processed while still coated on

the silicone substrate. The underlying sacrificial material is removed as a final step to create the

freestanding membrane. The primary advantage to this approach is the membrane stays flat and

secure against a background material while lithography and wet or dry processes are used.

Membranes are particularly susceptible to fracture when they have been nanopatterned and

during wet processes. The approach works best if the final release is done as a dry process or

supercritical drying is used to reduce surface tension and stiction damage.

A much less common approach is to remove the underlying sacrificial material as a

preliminary step and do most of the processing on a free-standing membrane. The majority of

the research presented in this thesis was based on this approach as it presents a few advantages.

The main advantage is the underlying sacrificial material removal, which is typically a wet

process, is done before nanopatterning, when the membrane has more structural integrity due to

less stress concetration points (e.g. holes). As shown in Section 3.1, I utilized a process that was

completely dry after nanopatterning a free-standing membrane. There are additional lithography

benefits to processing on free-standing membranes due to reduced proximity effects and dual-

sided patterning of the membranes (Section 4.1). The main disadvantages of releasing the

membrane as a preliminary step relate to difficulties in handling and processing without

breakage, distortion and contamination. Most wafer cleaning steps, wet or dry, cannot be done

with a free-standing nanopatterned membrane without breakage. Additionally, two sides of the

membrane are exposed, with the bottom side particularly susceptible to particles from the

underlying sample holders during processing.

Effectively any material that can be made into a thin film on a wafer could be a membrane.

Some materials are more appropriate as the membrane because of structural integrity. Materials

27
that are low in stress and give a smooth surface, reducing stress concentrations, are ideal. Silicon

nitride is a particularly useful membrane material because of these characteristics and its

compatibility with the silicon substrate. Additive processes can also be used to create a single

free-standing membrane layer with more than one material. This is shown throughout my

research with the addition of magnetic material on or embedded within a semiconductor or

polymer membrane.

A final issue that is significant for membrane stacking and folding, is distortion in the

membrane from intrinsic stresses, temperature gradients and from the stresses introduced from

material addition. Significant research has gone into creating low-stress membranes22 and efforts

have been made in MEMS research to add materials with reduced or balanced stress 23 . Section

2.2.1 goes over in more detail the stresses within silicon nitride membranes and those from metal

evaporation onto the substrate.

1.4.2 Nanostructured Origami actuation mechanisms

In my thesis work, I developed a new form of membrane folding based on nanomagnets for

the Nanostructured Origami method. Figure 1-5 shows some of the other methods used for

origami folding at the micro scale. The stressed bilayer occurs from one of two strategies. The

first is based on depositing a stressed material on a typically unstressed material. This was

achieved to great control by depositing stressed chromium onto silicon nitride membranes24 .

Figure 3-4 shows examples of structures that I built using the stressed-bilayer approach. The

advantage of the bilayer approach is that the patterning and folding actuation can be done in

parallel. The main disadvantage is the creases have a large radius of curvature relative to the size

of the origami structures. This makes it difficult to fold structures such that the surfaces align

28
face-to-face. Additionally, all the bilayer folding has been with the deposited material being

placed in tension, which makes the folding only one-directional. Kubota et a125 present folding

with a lattice-mismatch, which can created bi-directional folding but is limited by needing crystal

materials and the folding radius are even larger.

Stressed bilayer Focused Ion Beam

maged laver Curlingup

Surface Tension Lorentz force

Figure 1-5: Overview of the methods to fold origami structures at the microscale. The stressed
bilayer approach is based on a prestressed material or lattice mismatch between two materials. The
focused ion beam is based on injecting the membrane with ions to create typically compressive
stresses. Surface tension uses the line forces to fold and Lorentz force uses is based on a loop
current in an external magnetic field.

An improvement on the bilayer approach was made utilizing ion implantation 26 . In this

method, helium or gallium ions are implanted at hinge areas. By controlling the ion dose and

bias voltage, silicon nitride membranes were folded to prescribed angles. Unlike the metal
bilayer, the folds can be made in both directions and the effective radius of curvature is roughly

lOX smaller allowing face-to-face contact. The main disadvantage of this approach is the ion
implantation of the necessary doses can only be done with a focused ion beam system, which is a

serial process, making it slow and expensive. This is a promising direction of research and can

be used in conjunction with the results of this thesis work as outlined in Section 5.2.

Instead of patterning or implanting stress into hinge areas, the other approach is to make

structurally compliant hinges and actuate from an external force. The first MEMS folding2 7

developed at Berkeley was simply based on an external mechanical probe folding over micro-

plates connected by compliant micro-patterned hinges. From there, folding strategies using an

external magnetic field and a circuit to create a Lorentz force was developed5 . This is

schematically shown in Figure 1-5. The advantage of this method is complex, sequential folding

can theoretically be achieved, but it is greatly limited in that live circuits must be created

throughout the origami structure. High currents are necessary that have been shown to heat and

therefore warp the MEMS structures. This limitation was a motivation for developing the

nanomagnet folding approach presented in Section 2-1.

1.4.3 Required innovation and improvements

The following is a review of the required improvements to make membrane stacking or

folding a prominent 3D nanomanufacturing approach. These include: improvements to

membrane processing, reduction of stress and thermal-induced distortions, controlled release of

the membrane, and layer-to-layer alignment. The necessary improvements to membrane

processing are in particular improved cleanliness, patterning, material choices, and reduced

fracture for higher yield. My work on patterning and materials is presented in Chapters 3 and 4.

30
Since the membranes are more susceptible to distortion, there must be improvements in handling

stresses and temperature gradients. My work on this is presented in section 2.2 and 4.2. The

third required improvement- controlled release of the membranes after they are stacked- is likely

the most straightforward. Aoki' uses an electrostatic probe to hold the small membrane

segments. A mechanical fracture release mechanism, as shown in Section 5.3, or a sacrificial

frame are also possible solutions. Origami does not need a release step since there is no frame

holding the individual layers. On the other hand, origami has a fundamental problem with yield.

Whereas membrane stacking is a serial processes such that each layer can be inspected for

defects before stacking, the origami approach requires each segment of the origami to be of the

necessary quality. For example a 10 layer structure made with origami where each layer has

90% yield would have only a 0.9A10 or 35% overall yield. Therefore, it is of the author's

opinion that origami holds great advantages for structures with only a few layers or for structures

with volumetric 3D properties, such as the corner-cube retro-reflector. Origami achieves coarse

alignment with folding about a hinge and stacking typically utilizes a stage or other actuator for

coarse alignment. Both can be easily achieve on the order of microns and potentially better with

feedback systems. However, for both approaches, fine alignment on the order of 1Onm or better

is required for many of the envisioned 3D assemblies. This becomes even more difficult when

considering the distortion within a membrane. The focus of this thesis is a new membrane

alignment method that can utilize the typical coarse alignments while achieving the necessary

fine alignment. The following section overviews approaches used for passive alignment at the

membrane, chip, or wafer level.

1.5 Survey of membrane alignment methods

 Membrane alignment can be separated into the categories of active and passive. The active

approach is similar to standard mask and wafer alignment systems seen in nanolithography. For

mask alignment, these typically use matching alignment indicators (e.g. gratings or cross-hairs)

on the mask and wafer. A sensor, feed-back system and actuated stage ensure that the wafer and

mask can get to better than 1Onm alignment. For membrane stacking, the approach includes

putting alignment features on the membrane and matching features on the substrate on which the

stacking occurs. The main difference is the membrane cannot be assumed to be a single rigid

structure, so therefore a "global" alignment achieved with the active approach does not necessary

indicate "local" alignment to that precision. For this reason, active alignment may be best for the

coarse positioning, while some sort of passive alignment method gives the fine alignment on the

local scale. Table 1-2 shows an overview of the achieved or theoretically achievable membrane

stacking or folding passive alignment methods. These are outlined in the following sections.
 Method Schematic Coarse Fine Dry Simple to Non- Pressure Is
alignment alignment alignment fabricate absorbing not
>1lpm <25nm process material needed

Layer-to-
layer
A A2
mechanical O0000 @0e
Base
mechanical
alignment
guide 00 @0 @0
Mechanical
alignment

hinges @0
Surface
tension
0 0 00
Electro-
static
*@00 0
Magnetic I S

@ 0 0 0 0 0
*Yes O No
Table 1-2: Comparison of passive alignment methods for membranes and wafers. The empty
boxes for electro-static are unknown

1.5.1 Mechanical

The most straight forward approach for passive alignment of nanomembranes involves using

mechanical features that constrain movement within a small tolerance. In general, a female

feature on one layer matches with a male feature on another layer. This was first seen in

microtechnology for wafer stacking using prismatic shaped elements that fit to matching holes.

There can be a benefit of "elastic averaging", which serves to reduce alignment error when the

alignment features are distributed with a positional noise. In effect, the errors of the individual

33
features average out to give an alignment position that is statistically more likely to be near the

ideal position than the individual features. Section 2.3 shows that this same benefit of elastic

averaging may be achieved with a particular orientation of nanomagnet alignment.

Figure 1-6: Techniques used by Aoki for alignment of small membrane

segments. (a) colloidal spheres that fit into matching circular voids and (b)
pillars patterned onto the substrate, which contact the segments at 3
points to constrain alignment.

Mechanical alignment features were used by Aoki"' 2 0 for accurately stacking small

nanopatterned membranes within a modified scanning electron microscope. There are two

primary approaches. The first approach shown in Figure 1-6(a) is analogous to stacking Legos.

Approximately one micron diameter microspheres were placed into circular voids patterned into

the membranes. A nanoprobe with electrostatic holding force accurately places the

microsphere. This demonstrated that the alignment accuracy of the probe system was on the

order of 500nm or better. After placing the microspheres in the holes, an additional membrane
34
layer is placed on to the stack and forced into alignment. The method showed overall alignment

accuracy of 50nm for the small 20um membrane plates. The method has some key advantages.

First, it is simple both schematically and for design and modeling. Second, the fine alignment of

50nm is impressive, although it has been demonstrated over a small overall membrane size. The

main disadvantage of the approach is the coarse alignment must be close, requiring a specialized

set-up such as used by Aoki. Second, any approach of putting a peg-like feature onto a

membrane (e.g. nanosphere, nano-pillar, nano-pyramid), requires either the local addition of

nanofeatures after patterning such as shown by Aoki or conversely the difficult processing of

high aspect ratio and/or thick features onto the membrane. Finally, an external element such as a

probe or field must force the membranes into contact.

The second mechanical alignment strategy used by Aoki is based on large alignment pins

placed on the substrate matching with alignment notches on each of the membrane. This is

shown in Figure 1-6(b). The method was used to reduce the particle contamination of the

polystyrene nanospheres and presumably to eliminate the tedious task of placing nanospheres

into each of the holes in the membrane. The columns over 5 microns tall were patterned into a

GaAs wafer. These match with three notches on each of the membrane segments effectively

giving a kinematic coupling and alignment on the order of 50nm. Although a very promising

approach for unique applications of small membrane segment stacking, there are three primary

disadvantages. First, like the Lego approach, the coarse alignment must be precices- typically

less than one micron. However, the coarse alignment may be improved by patterning the pins to

have guiding features such as pyramid tops. Secondly, the method needs a force to push the

membranes together, which is difficult over larger area membranes of a scale in millimeters
instead of tens of microns shown by Aoki. Finally, the method is very-much tied to the non-

trivial ability to accurately pattern nano-pillars over 10 microns tall over large areas.

1.5.2 Surface tension

At the micro and nano-scale, surface tension forces are comparatively much more

dominant than they are at the macro-scale. This favorable scaling has been utilized for alignment

of microtechnology components. Leung et a12 9 showed alignment of 50um x 50um x 5um silicon

die. Binding sites with various "wettabilities", or affinity for the liquid, are patterned on the die

and substrate. The use of surface-tension forces for die alignment is effective because the chips

are typically structurally sound enough to resist damage during the critical liquid evaporation

step. However, for membrane alignment, the requirement of an evaporating liquid is prohibitive

because of the fragility of a nanopatterned membrane. Like the mechanical alignment, this

method holds promise in very select non-fragile assemblies, particularly since the technique

provides an alignment force without needing an external pushing mechanism.

1.5.3 Magnetic or electrostatic

Electrostatic actuation is very common at the MEMS scale. Due to the ease of electrical

control and reasonable force scaling, the capacitive actuator has become a key element of both

simple and complex mechanical systems. The use of capacitive attraction between two

membranes for alignment would be a poor approach because of the need for a live, complicated

circuit to supply the voltage bias. Magnetic actuation is not as common due to non-ideal scaling

of electromagnets. Cugat et a130 suggested in 2003 that although electro-magnets do not scale

well to the micro-scale, the force on permanent magnets favorably scale. Their main interest was
applying force from a micro electro-magnet to a micro permanent magnet. Although no

alignment applications were presented, they also showed that the most favorable micro-scaling

occurs between two permanent magnets.

Nanomagnets have become an important research area, primarily due to their promising use

in computer memory systems. More than in macro-scaled bulk materials, the shape and size of a

nanomagnet directly affects its magnetic properties. This is due to the magnet being on the scale

of the magnetic domains of the material and thus manifests itself in shape anisotropy. This effect

can be utilized to create strongly preferred magnetic poling. For example, long, slim

nanomagnets made of otherwise isotropic magnetic material preferably magnetize along their

axis. These effects are understood well and a variety of shapes, sizes and materials have been

explored. Nanomagnets supply many of the benefits of mechanical alignment with two key

advantages. First, the stray field from the magnets interacts at distances beyond their size, so the

coarse alignment tolerance can be much higher. Secondly, the magnets provide a local pull-in

force, which can be important over large areas or for otherwise distorted membranes. Third,

nanomagnets can be patterned to high precision as they can be a simple metal layer in lOs of

nanometers thick. The core of this research was to model, understand and prove experimentally

the use of nanomagnets for membrane alignment. To the author's knowledge, the use of

nanomagnets for micro-and nano-technology alignment had not been explored before the

research in this thesis.

Chapter 2

Theory and design of nanomagnet alignment

2.1. Magnetic forces

The forces generated by macro-scaled magnets are well understood and are used in devices

from compasses to electric motors. The traditional model of the magnet being a dipole is valid

for specific geometries and magnetizations. In particular, if the magnet has a high aspect ratio

and is magnetized along its length, the force calculations derive from considering the attraction

and repulsion on the two magnetic poles. In more complicated geometries and magnetizations,

the internal state and shape must be taken into consideration. This makes the calculations more

complicated, but most can be modeled easily with assuming distributed poles, volume integration

or numerical modeling such as FEMM31 .

For the nanoscale, many of the same equations and models can be utilized. However,

additional considerations must be made since the magnets are near the size of the magnetic

domains and thus have substantial anisotropy from patterned shape. The hysteresis curve and in

particular the coercive field, determines the magnets "hardness." A magnetic material that

would otherwise be a soft, easily switched material at the macro-scale may be made to be

effectively harder by the shape anisotropy. For example, Kartopu32 showed that pure cobalt

patterned into nano-wires can exhibit a coercive field of 900 Oe compared to its macro-scale

value of 10 Oe. The strongest macro-scaled magnets exhibit coercive H-fields over 40,000 Oe.

38
For that reason, when two strong permanent magnets are brought into proximity, they usually do

not appreciably switch the other's internal magnetization. This, for example, allows them to

place a repulsive force on each-other without one of them switching magnetization direction.

The coercive field of nanomagnets that can be easily deposited with e-beam metal evaporation is

not nearly as high as the more complex alloys used to get over 10,000 Oe. Therefore, simply

treating the nanomagnets as permanent magnets with set magnetization is often incorrect.

Figure 2-1 shows Object Oriented MicroMagnetic Framework3 3 modeling that was

used to understand the internal magnetizations and force interaction between

nanomagnets. Figures 2-1(a-c) show the internal magnetization states for two different

aspect ratios of 100nm wide and 50nm thick cobalt nanomagnets in various B-fields. They are

placed in an external field at 45 degree angle as shown. As can be seen, the longer aspect ratio

magnet exhibits more shape anisotropy by resisting magnetizing in the short direction. This is a

critical effect for applying torque on a soft nanomagnet using an external field as discussed in the

next section. Even more interesting and novel to this research is looking at the internal

magnetizations of soft nanomagnets that are nearby each other.

 0. 5T

-A A 0.2T
* i
I I

d- 0.5T

-
-0-9 -W -V 0- 0-44 0 .6 . -44 *-W .3-W

Figure 2-1: OOMMF Micromagnetics modeling showing the effects of magnet

shape and external B-field. All magnets are 50nm thick cobalt. (a-c) show 100nm
wide nanomagnets with lengths 150 and 500nm. The B-field is angled at 450
initially at (a) 0.5T, then (b) 0.2T and then (c) reduced to no field. (d-e) show two
100nm by 400nm magnets in close proximity (d)with an external B-field and (e)
without after being being magnetized along axis to show coupling between the
adjacent nanomagnets.

Figure 2-1 (d-e) show the effect of magnetizing nanomagnets along their length and then

gradually removing the external field. As can be seen, the local field created by the magnets can

cause magnetization shifts within the neighboring magnet. Therefore, the present research used

the novel approach of aligning the magnets while still within an external field and removing the

field after alignment is reached. In this approach, the nanomagnets can be more accurately
modeled as a single magnetization along their length. This is exhibited in Figure 2-1(d) which

shows a 1T external B-field dominating the coupling B-field between nanomagnet. In order to

utilize this approach, the external field must place a negligible force on the magnets, which is

presented in the following section. Section 2.1.1 presents the torque and forces from the external

field. Section 2.1.2 presents the force between nanomagnets in an external field and within close

proximity. Section 2.1.3 goes into the design of the arrays given the force scaling and geometry

constraints.

2.1.1. Force from external magnetic field

When a magnet is placed within an external magnetic field, there is both a torque and force

applied on the magnet. For the slender magnet geometries used in this research, the dipole

approximation is accurate for calculating the force and torque from external fields. The pole

strengths are given by:

P = ipo-_ MA (2.1)

where M is the internal magnetization in Tesla and A is the cross-sectional area of the magnet,

7 Vs. Assuming that the field is directed

and po is the permeability of free-space, 4r x 0 A~m

along the length, L, of the magnets, the force from the external B-field is given by:

Fe = Be,1P1 - Be,2 P2 = Yo~1MAL(VBe) (2.2)

Fortunately, the external field can easily be made to be quite uniform relative to the size scale

of the nanomagnets. Therefore, the value of L(VBe) can be negligible such that the external field

can be used to solely magnetize the nanomagnets. For example, a B-field gradient of IT/cm

easily made with an electromagnet setup would place a force of only 0.05pN on a 500nm long
41
by 100nm diameter Co nanomagnet. In sections 2.1.2 and 2.2.3, it is shown that this is a

negligible force compared to the inter-magnet and mechanical forces, which are typically over

InN. The torque from the external B-field is not negligible and is given by:

Te = BePLsin(a) (2.3)

where a is the angle between the length of the magnet and the external field. It should be noted

that this requires a relative angle between the magnet and external field so the shape anisotropy

becomes very important so that the magnets can be magnetized along their length even if the

field is not perfectly aligned with that axis. Therefore, a is typically less than 45 degrees, but

this can easily be modeled using the OOMMF method presented here. If internal magnetization

can be controlled via shape or material anisotropy, Equation 2.3 becomes

Te = o 1 BeMVsin(a) (2.4)

for an arbitrary volume, V, of magnetic material such as a patterned array of nanomagnets. The

majority of the work of this research has been on I 0 0 pm membrane segments and there is a

limitation to the thickness of nanomagnets by membrane distortions presented in Section 2.2.1.

Additionally, the external B-field is likely less than 2T, unless a substantially large and

expensive electromagnet is used. Therefore, the torque that can be applied from the field is

limited and drives the compliant hinge designs presented in Section 2.2.2. Figure 2-3 shows the

torque on 100pm membranes for various magnet thicknesses and 5 percent coverage and 1T

external B-field.
 45

40

35

E 30
Z25

0 20
E
015 4n
0

10
2Onm

0
0 10 20 30 40 50
Angle between external field and magnet axis

Figure 2-2: Moment applied on an array of cobalt nanomagnets

covering 5%of 100pm x 100pm membrane in a 1T external B-field.
The relative angle and thicknesses of the films are shown.

2.1.2. Force between nanomagnets

The force between magnets is benefited by a more favorable scaling to the nano-scale

compared to the force and torque from external an external B-field. The force between magnetic

poles separated by a distance d is approximated by

Fm = 4i0d
41rd2
2
(2.5)

Where P1 and P2 are the magnetic pole strengths P, = po-MAn, n = 1,2. When the magnets

are of the same magnetization Min Tesla and cross-sectional area A this becomes:

f, (d) = m2,
41ryo dl
(2.6)
 ! - w -

Figure 2-3: Schematic of in-plane nanomagnets coming into alignment. The top two represent magnets
on one membrane while the bottom ison the other membrane with a resultant force, F.

As all dimensions scale down by some factor k, the force between magnet poles scales down

by only k2 . Therefore, the force per volume increases as dimensions are reduced. In other

words, at the nanoscale the force between nanomagnets scales advantageously. The magnets

are, of course, dipoles so the force on each of the poles must be considered to get the entire force

on the magnets. Figure 2-3 shows the dimensions considered for modeling the force between

magnets. The force between magnets, F, is defined as the combination of the four forces

calculated from the distance between poles.

di 'di ~ di ~ di
s1= dj ,S4 = dj +L, S3 = +2L 4 = d;+ L
,s (2.7)
d dA d[ dk
Using the force between poles from Equation 2.7 and assuming di >> wi, the net force is given

by

F = Zn= 1(-1)snf(|s,||) (2.8)

 For the work in this thesis, it is most interesting to separate the force vector into its

components because only two of them act as alignment while the other acts as the pull-in force.

The orientation of the nanomagnets on the membrane determines which directions are the

alignment and which is the pull-in force. If the magnets are patterned to be magnetized in-plane

as shown in Figure 2-3, the alignment components of the force would be in the i andj directions

while the k direction is the pull-in between membranes. If the nanomagnets are patterned to be

pillars, the alignment components are in the i and k direction, while the pull-in happens in the j

direction. Since the magnetic force goes with the inverse of distance squared at larger distances,

it is first important to understand the pull-in force for both the in-plane and pillar orientations of

nanomagnets.

In the remainder of this section, both in-plane bar magnets and out-of-plane nano-pillar

magnets pull-in force is calculated. Table 2-1 compares the options of in-plane and pillar

nanomagnet arrays. The out-of-plane nano-pillars show greater pull-in force at close distances

and should result in more precise alignment than the in-plane set-up. However, it is more

difficult to fabricate high aspect-ratio nano-pillars than in-plane bar magnets and the nano-pillars

result in a significant gap between the components that are being aligned. The in-plane bar

magnets are very easy to fabricate with e-beam or optical lithography, metal evaporation and lift-

off. Nanomagnet pillars can exhibit a greater pull-in force at large gaps between magnets (e.g.

multiple microns), but the pillars provide greater force when the membranes are in closer

proximity.
 TOP VIEW
aa TOP VIEW
D
Attractive force between nanomagnets
L 0 0
1.E-06

t SIDE VIEW 1.E-07

21.E-08
7-----------
51.E-09
U-

1.E-10

I.E-11
0~ 0.E+00 1.E-06 2.E-06 3.E-06 4.E-06
Z77
Space between arrays s (m)

Figure 2-4: (a)Diagram showing the dimensions s, t, Land Dnecessary for modeling the nano-magnet
pillar pull-in force. (b)Force on individual pillar nanomagnets at the diameters, height, and gap size
given.

Nanopillars of magnetic material such as Ni, Co or Fe can have a preferred magnetization

out-of-plane. Figure 2-4(a) shows the dimensions that are varied by the pattern and material

deposition. Most importantly, for a preferred out-of-plane magnetization, the aspect ratio, t/D

should be greater than 1 depending on the material. For example, Co nanomagnets should be

patterned to at least t/D=1.25 for out of plane preference. If t/D is much smaller than 0.5, the

disk-shaped magnets have strong in-plane magnetization preference, which can be problematic

for this alignment technique. If 0.5<t/D<1.25, the magnetization preference is dependent on a

number of factors, including the external field strength and orientation. The force between the

magnets is given in Equation 2.9 where Mo is the magnetization in Teslas and pto is the

permeability of space 47E x 10-7 Tesla-m/A.

 MOD 2 (t2+D 2
) 1 1 - 2 (2.9)
F pn, ,-
pillar 2
16poL _s2 (s+ 2t) 2 (s+ t)2 (

Figure 2-4(c) shows the results for the attractive force between Co nano-pillar magnets

separated by a distance of 0 to 3um. Two different magnet diameters were modeled- 0.5um and

0.25um, while varying the magnet height- 0.25um, 0.5um and lum. This assumes that the

magnets are saturated out-of-plane, which may be difficult for the smaller aspect ratio magnets.

Overall, the forces ranged from 0.0 1nN to 100nN. Assuming a coarse alignment (initial gap

size) of lum, the force between two magnets can easily be greater than 1OnN with the 0.5um

diameter magnets, increasing to 1OOnN in contact.

The force between nanomagnet arrays in contact can be estimated as a pressure depending on

contact area. The pressures are approximated as M2at/2po. For Ni, Co, and Fe pillar magnets in

saturation, we get the pressures to be PNi 00 Ni - and PNi = 1.5 ,

These numbers can be used to approximate a bond force between components if van der Waals is

not significant. For example, if 10% of a 1 mm2 area is covered with Co nanomagnet pillars and

brought into contact with a matching array, the bonding force between them is 0. 1N if brought

into perfect contact. If nanomagnet pillars are too difficult to integrate into a component or a

large gap between the components is problematic, in-plane bar nanomagnets can be used. In

this case, the force is given in Equation 2.10 with the dimensions given in Figure 2-5(a-b). The

force is valid for t<s<L.

= 4Asat2 t
4Min-plane 2 +s 2 -s (2.10)
pull-inin-plan
 a L TOP VIEW b Force between 50nm thick in-plane magnets
1.E-06
L MsaI

Mat ^ 1.E-07

SIDE VIEW
t 1.E-08
_ E

-- 1.E-09 w=1 pm
w=0.5 pm
0
1.E-10 -w=0.25 pm

1.E-11
O.E+00 5.E-07 1.E-06 2.E-06 2.E-06 3.E-06 3.E-06
z
Spacing between magnet arrays, s (m)

Figure 2-4: (a) Diagram showing the dimensions s, t, L necessary for modeling the in-plane nanomagnets
pull-in force. (b) Force on individual 50nm thick in-plane magnets widths shown and length much larger
than gap size.
Figure 2-5(b) shows the force between 50nm thick nanomagnets at different widths-

1pm, 0.5 pm, 0.25 pm for s<L. As shown, the forces are similar to those shown in the vertical

pillars. However, the pillars can be placed much more densely on the component. For example,

compare the 5Onm thick by lum wide in-plane nanomagnets with the 0.5um diameter

nanopillars. The force at lum spacing is give by lOnN and 30nN respectively. The nanopillars

can be placed at a spacing of lum (1 magnet/pm 2 ) while the in-plane are at a spacing of

2umx4um (0.13 magnets/pim 2).

Once pull-in is achieved, the more important components of the magnetic force are for

alignment. For a range of the alignment process, the dipole approximation remains valid,

Wk <Wi <Wj < 6- . Unlike the pull-in force, the alignment forces are most interesting when

the magnets are in close proximity, 8 ~ Wk < wi < wj, so they generally cannot be treated as

dipole forcing. Referring again to Figure 2-3, since the magnets are magnetized in the j
48
direction, the magnets can only be treated as a dipole to estimate the alignment force in the j

direction. In the other directions, the alignment force is dependent on overlapping magnet areas.

The alignment force in the j direction is estimated as:

fi(d, dk) = A2 di 3 (2.11)

47r1oLd +dk

If dk >> di, the above equation can be approximated as a spring force dependent on the

distance of misalignment, with spring constant.

= m (2.12)

Although this requires a gap between the magnets, this can be easily achieved by a

mechanical spacer. This is an important fact because the force scales with the misalignment in

the form dj as opposed to 1/dj2 as is typically expected as the force between magnetic poles.

The main benefit of having a linear spring-like force scaling with misalignment is a phenomenon

of "elastic averaging" 28. In elastic averaging, the error of alignment between arrays of alignment

2
features is reduced due to averaging of the individual feature errors. Conversely, a 1/d1 force

scaling would risk outlying alignment features dominating the alignment process and causing

worse overall alignment.

 1.4e-7
1.4e-7 1
3- 1.2e-7
OL t-06e
L>
- A
E 8.0-8
E
r 6.10"- 0.33-W

20.08 w+1 w=0.1

ao.e+O _i I
0 02 0.4 0.6 0.8 1 1.2 14 1.6 w
Relative overlap area (di/w) w
Figure 2-6: Alignment force depending on the overlap and aspect ratio (wj/wi) for 1 micron wide
nanomagnet (wj=1pm).
For the i direction in the in-plane orientation and i and k for nanomagnet pillars, the overlap

of the magnets must be considered. The dipole model breaks down and the force calculation in

the form of A f f f f ((X(Xl-X-X

1
2)dXldX
2 2 dy 2dy32 2
2 ) +((Yl-Y 2 ) ) /
has proven too difficult to solve analytically, even

when considering simplifying assumptions. Therefore, numerical simulation was used to

understand and model the alignment force when the surfaces of the magnets are overlapping.

Referring to Figure 2-6, consider two overlapping magnets of dimensions wi, wI and Wk

wi, w; with misalignment, di, the alignment force is shown in Figure 2-6 for various aspect ratios

w;/wi and percentage of magnet overlap. The y-axis shows the force per micron of magnet

cross-sectional area for a cobalt magnet in magnetic saturation. As can be seen in the plot, there

is a region where di « wi in which the magnetic alignment force sharply drops with decreased

misalignment distance. Regardless of aspect ratio, there is a region roughly defined as 0.2wi <

di < wi where the alignment force stays relatively even with a maximum force occurring at
50
roughly 0.6w. Finally, when the magnets are no longer overlapping and the misalignment

distance, di > wi, the plot starts to approach the more recognizable di-2 magnetic force scaling.

For overlapping magnets, this poses an interesting problem when considering alignment

equilibrium with multiple features. Instead of a linear scaling of alignment force with

misalignment distance, the overlapping magnets with small di see a force F(di) oc "a , n > 1.

This force scaling is not exactly the same as needed for true elastic averaging where F(di) oc di.

However, unlike the d- 2 scaling of magnetic dipoles, the overlapping magnets still give the

important necessary properties of Fj (0) = 0 and a monotonically increasingforce at small

displacements. This effect is illustrated in Figure 2-7 for the simple one-dimensional case of two

alignment points with mating alignment features. Figure 2-7(a) first shows the alignment given

perfect positional tolerances for force scaling with d, d~2 , and Vd. In a perfect system (a) the

expected final position is perfectly aligned regardless of the force scaling. Ignoring all other

forces, the array should find the minimum energy position given that it is roughly in coarse

alignment. Now consider four positional errors ei,1 , ei,2 , e 2,1 and e 2,2 for each of the four

alignment features. The final alignment error is the solution to F1 + F2 = 0 where F1 and F2 are

the resultant alignment forces for each of the feature sets. For the straight elastic averaging case

and el = e1 ,2 - e1 ,1 and e 2 = e 2,2 - e2 ,1 this is given as

F1 + F2 = k - (Sf - e1 ) + k - (6f - e 2 ) = 0 ~*6f > Sf,N=Lj (2.13)

 (a)
....Si,
+F ocd-2
+ +
+ Focd, "

(b)
0 +0

e2 ,1 F c d- 2
0 + 0
el,,

o + 0
0 + 0 (c)
1,,2 e2,2 -+0
1 0 + 0
FocdV"

f 1 +e,2 +
,(e +2,2

Figure 2-7: Comparison of the effect of the alignment feature force scaling for the simple ID case of two
alignment features on membranes starting with a coarse alignment error S. (a)The features are
assumed to be perfectly patterned (en,n=O) resulting in perfect final alignment (6f = 0). A patterning
error isintroduced for (b-c). The resultant position for force scaling F oc d- 2 isshown in (b)showing
unwanted pull-in. The resultant position for F oc d and %I,n > 1isshown in (c)with a final position
being the sum of patterning errors in the alignment features.

Expanding this to more than two alignment points, the expected value of the final alignment

of N features equals zero with a variance -1K times the variance of the positional error of the

alignment features. When considering the d- 2 force scaling, the alignment occurs much

differently. In this case, the initial position must be considered. The equilibrium position results

in an error defined by:

{5f = el f6i - e < (i -e 2

I
(2.14)
5 = e2 otherwise

Instead of averaging the positional errors, one of the errors results as the final position.

Clearly, this is non-ideal because an alignment feature with a large positional error could

dominate the final alignment. It should be noted that there is a theoretical equilibrium with
52
Ef = but it is an unstable equilibrium as any disturbance would result in pull-in. Now
Ef 2

considering the force scaling of the overlapping magnet faces by Wd-, it has the first benefit that

the final alignment is not dependent on the initial alignment. There is no pull-in effect so one

can consider the force balance, as long as the features are within the initial alignment

aforementioned. The force balance thus gives:

k - (6f - e,), = k - (45 - e2)1/ -> Ef = (2.15)

This solution works for the case of two features. With additional features, it does not scale as

clearly as in the linear force case. However, considering the alignment system as sets of two, it

can be considered a form of quasi-elastic averaging as there is benefit with each additional

alignment feature. In the case of overlapping magnet faces, these benefits are achieved in the i

direction as long as the distribution error in the alignment features is less than the dmax where

F't(dmax) < 0. As shown in the plot, this distance, dmax, can be safely approximated as 50%

of the magnet width wi for the aspect ratios shown. If elastic averaging is completely necessary,

the alignment force from overlapping magnet faces also becomes more linearly scaled when a

spacer is introduced. The above analysis was based on slender in-plane nanomagnets. For

typical nano-pillar magnets the force from overlapping faces is similar, but one ignores the

interaction between the other poles.

2.1.3. Nanomagnet array design

The previous section focused on the forces between individual magnets and alluded to the

fact that there are gains from the use of multiple magnets. Beyond averaging of positional errors,

multiple nanomagnets are needed to supply enough force and overcome the mechanical elements
presented in Section 2.2. Therefore, it becomes critical to consider not only the individual

magnet interactions, but the interaction of arrays of nanomagnets. There are a number of

considerations when designing magnetic arrays for membrane alignment including: magnet

orientation (in-plane or pillar), magnet size, magnet shape (rectilinear or rounded), magnet angle

(rectilinear or angled), input course alignment, goal fine alignment, mechanical forces, external

field, magnetic material, location on or in membrane, material/process compatibilities,

processing accuracy and so-forth. Chapter 3 walks through multiple processes and designs that

considered all of these topics for specific applications. This section looks at the theory and

design considerations behind the magnet array orientation, shape size, angle and location on or in

the membrane.

The orientation of the magnets in or out-of-plane is one of the most significant considerations

for the array design. The previous section detailed the scaling of pull-in and alignment forces on

individual magnets and showed that nanomagnet pillars exhibit more proper alignment force

scaling when the magnets are at or near contact. When considering interactions between arrays,

one must look at the net interaction when they are at a far distance. At coarse alignment the

major considerations are: (1) Whether enough force can be provided for pull-in, (2) whether the

arrays pull in at the correct relative position. Both of these questions depend greatly on the

initial alignment error and the gap between membranes at coarse alignment. First, if a

nanomagnet array has a set period, the spacing obviously cannot be less than the initial

misalignment error at coarse alignment. This would result in accurate magnet-to-magnet

alignment but the entire system would risk being one or more periods off in the alignment.

When looking at the force at coarse alignment, it is best to consider both the force between

individual magnets as in Section 2.2 as well as considering the effects of neighboring magnets.

54
Figure 2-8 shows 2D FEMM models of the local H-field strengths from vertical and horizontal

magnet orientations at various lengths. For nanomagnet pillars, the use of additional magnets

actually provides a greater pull-in magnetic field everywhere. However, the magnets are

typically shorter in length compared to the in-plane options so their individual interactions are a

shorter distance. A typical size of a nanomagnet pillar may be 500nm tall, whereas in-plane

magnets can be over a few microns long. As can be seen in Figure 2-8 the longer the magnet, the

farther away a given field strength travels. For example, comparing the 4x1 and 2x1 magnetic

pillars, the H-field strength is sustained more than two times as far for the 4x1 magnets for both

the individual magnets and magnetic arrays. For in-plane magnets, the positive and negative

poles in close proximity reduce the effective field, particularly at distances greater than the

length of the magnets. This can be seen very clearly in Figure 2-8(b) in that the field falls off

very quickly at a distance roughly the length of the magnet.

 2.458e+005 : >2.587e+C
a 2.329e+005 : 2.458e+00
2.199e+005 2.329e+00
2.070e+005 2.199e+00
1.941e+005 2.070e+00
1.811e+005 1.941e+00
1.682e+005 1.811e+00
1.552e+005: 1.682e+00
1.423e+005 : 1.552e+00
4X1 in-plane 4X1 in-plane, 8 spacing 1.294e+005 1.423e+00
1.164e+005 : 1.294e+00
- 1.035e+005 1.164e+00
9.056e+004 1.035e+00
7.762e+004 9.056e+00
6.469e+004 : 7.762e+00.
4X1 pillar 4X1 pillar, 2 spacing 4X1 pillar, spacing 5.175e+004 : 6.469e+00.
3.881e+004 5.175e+00
2.587e+004 3.881e+O0
1.294e+004 : 2.587e+00.
<0.000e+000 : 1.294e+O

2X1 pillar 2X1 pillar, 2 spacing 2X1 pillar, 4 spacing Density Plot: IH1, A/m

Figure 2-8: FEMM modeling of magnets showing the stray H-field depending on nanomagnet
orientation, size and spacing. (a-b) Show in-plane magnets and (c-d) show magnet pillars.

Table 2-1 shows the trade-off between in-plane and out of plane nanomagnets for the

membrane alignment system. In general, the in-plane orientation wins on fabrication, long

interaction, and large coarse alignment. The pillars allow the highest bonding force, and work

well for multiple layering, which will be discussed later in this section. Pillars are also more

forgiving for the lithography dose. This occurs because the in-plane magnets fit together at an

exact duty cycle in one of the directions. If the magnets are too long, they will not fit together

physically. If the magnets are too small, the result is two possible alignment positions with an

expected alignment error of y/2 where y is distance that the magnets are smaller than their

intended length. Since the error from dose can easily be controlled to less than 20nm for optical

or e-beam lithography, one would not expect dose to account for more than a 1Onm error. For
pillars, a larger or smaller dose will only change the pillar diameter and not position, resulting in

a greater tolerance for dose errors.

Long Low aspect . Multi- Large Applytorque

interaction ratio High bonding layering lithography with
force adds force dose range external
distance features

In-plane
nanomagnets

Pillar
Nanomagnets

0 0 0

Table 2-1: Comparison of the advantages and disadvantages of in-plane and pillar nanomagnets for
membrane alignment.

The final consideration when looking at the comparison of in-plane and pillar nanomagnets is

the degree of shape anisotropy that is possible. In order to apply a torque as discussed in Section

2.1, the magnets need a high aspect ratio to bias the magnetization along their axis. This is

theoretically achievable with nanomagnet pillars, but getting high aspect ratio features in pillar

form is a difficult process and takes up an appreciable amount of space between layers. For this

reason, more than any other, in-plane nanomagnets were used for the majority of the

experimentation done in this thesis work.

 Y

x a. b. C. d. e. f

L._

AK~J\ --->
-f

Figure 2-9: The various shapes that the nanomagnets may take along with their
failure modes. (a) rectangular (b) rounded (c)tapered (d)angled (e) arrow and (f)
rough.

The shape of the nanomagnets also plays a role in the expected accuracy of the alignment

approach. There are other options to consider besides the traditional bar magnet rectilinear

shape. Figure 2-9 presents a number of in-plane magnet shapes that have been considered and

modeled in this research. The rectangle (a) presented in the previous chapter has the one major

drawback of needing the magnets to fit perfectly inter-digitally. The most obvious approach to

58
reduce this requirement is to taper or round the ends of the magnets (b-c). The rounding (b) is

generally a poor choice as the magnetization is difficult to control along a rounded surface. The

tapering (c) is more promising as a theoretically smaller magnetic pole can be achieved. Further,

if the magnets are patterned too long in the y-direction from a dose error it just becomes a

translational error in the x-direction. The greatest problem with the tapering is controlling the

magnetization to stay in the y direction. Further, if the magnets are patterned too short, there is

the same problem as the bar magnet of a bi-positional alignment. In order to overcome this

problem, the approaches of d and e are suggested. Instead of the point coming into contact, the

magnets are in contact along angled faces. If the magnets are patterned too long or short, the

resultant error in the y direction is zero and the resultant error in the x direction is equal to

Ex = tan (6)Ay where Ay is the error in length positive or negative. Another option of tapering

is the arrow form shown in (e). However, there is the same limitation of the rectangular magnet

of needing a perfectly controlled length so this suggested against. Finally, the tapering can

actually occur out of plane, resulting in only an expected error Ez from an error in length Ay.

The final important consideration of magnet shape is surface roughness at the line of contact.

Even if this is not purposely patterned in the lithography step, this is a valid assumption for

nanomagnets as the metal material is deposited as a granular structure and perfect edges are not

achieved. If a y surface roughness is assumed, this must be modeled into the expected magnetic

attraction and mechanical model. Figure 2-10(a) shows a simplified model separating the

magnet into N poles with s= y spacing and 0 to y change in length for each sub-magnet with a

cumulative distribution function of 1. The force in the y-direction is defined as:

F = C El Y1,+Y2, (2.16)
J=
JJ((x2,+xloj)2+(ygi+y2,) ) /
 Figure 2-10(b) shows the simulation results of applying various noise levels to a 500nm wide

magnetic pole. Magnetic poles of 500nm width and 50nm height were simulated at various gap

lengths. For each gap length and surface roughness 100 simulations were run with the average

force and variance shown in the plots. As expected, the magnets still scaled 1 1<n< 2

where d is the gap between the magnets. However, with additional roughness, the force with the

higher surface roughness is considerably lower with a greater variance. The reduced force

during magnetic alignment may reduce alignment pull-in moderately, but other techniques are

required. Overall, the addition of surface roughening does not appear to add any tangible

benefits and was discarded. If the nanomagnets poles are overlapping but misaligned by some

distance d (Figure 2-11(a)), the resultant alignment force is defined in Equation 2.17. The same

simulation as described above was repeated for the geometry in Figure 2-11(a). The results are

shown in Figure 2-11(b) with similar effect of the previous simulation. In essence, a rougher

surface reduces the force and adds additional variance.

 10 20 W0 60
W 70
M M too

50
25
10nm

S0
10 20 30 40 50 60 70 60 90
Gap (nm)

Figure 2-10: (a)schematic of a rough nanomagnet surface being split into smaller
subsections. (b)The average alignment and variance depending on the gap and
simplified roughness (10, 25 or 50nm) for 50nm thick nanomagnets.

b. T.

10nm
L6
25
3
soQ

;1~
0 10 150 200

Id -25,

10nm

Misalignment-d (nm)
Figure 2-11: (a)schematic of a rough nanomagnet surface being split into smaller
subsections with a gap and translational misalignment d (b)The average alignment
force and variance depending on the misalignment and simplified roughness (10,
25 or 50nm) for 50nm thick nanomagnets at a separation gap of 25nm.
61
 Another magnet array consideration is to use a range of magnet size, shape and orientation to

get the benefits at each stage of alignment. For example, an array with perpendicular magnets

shown in Figure 2-12(a) was considered. The black shows one magnet array, which would be on

the first membrane, and a second mating array in gray for the other membrane. In this case, the

magnets must have strong shape anisotropy such that they magnetize along their length even as it

is non-parallel with the external field. Secondly, many of the same problems are still

encountered from the physical constraints of the magnets needing to fit inter-digitally. A method

to overcome this problem is presented in (b), as only one end of the magnets is in contact. This

reduces the requirement of the magnets to be patterned to proper length since the magnet set

perpendicular provide a proper force regardless of length. However, the 1/d 2 scaling of the

improper alignment force will likely overcome the F(di) oc Id, n > 1 scaling in the other

direction (see Section 2.2). For this reason, the benefits of changing the angles of the magnets

does not outweigh the problems of shape anisotropy that it introduces.

(a) (b (C) I

tBext fBext IBext

i0 Pi P2e
Figure 2-12: Orientations of nanomagnets that can further improve alignment. (a)angling nanomagnets
to get benefits of both alignment direction, (b)same without interlocking magnets and (c)cascading

62
 magnet size to get benefit of long and short nanomagnets.

Changing the magnets to have multiple lengths in the array has theoretical benefits. As

shown previously, longer nanomagnets have a longer interaction range and a larger tolerance for

coarse alignment. On the other hand, smaller nanomagnets allow the patterning of more

magnetic poles on the surface, which give a larger final alignment force and give more beneficial

averaging of patterning errors. Further, it is easier to control the lithography and patterning of

smaller magnets to a controlled error in length. Therefore, the approach of cascading the

magnets such that effective coarse alignment at far distances, L is maximized while also

maximizing the effective number of magnetic poles, neff at short distance. The below give the

weighted approximations for the various magnet lengths, forces, pole strength, coarse alignment

and fine alignment: Ln, Fn, Pn, dmax and d min.

Lcoarse LF(dm)+ L2 F2(dm)-++LF(dm) F = PF, a

a F(dma)+
Fe = F2 (dmx) + -- -+ Fn (dma) " '""(2.18)

n n F(d)+n 2F2 (dn)+---+nF,(dn) n A=n

"
F (dmi)+ F2(dm) +--+ F(dn-) " EL2 (2.19)

The final consideration for array design is the location on the cross-section of the membrane

that it is on. Referring to Figure 2-13, patterning magnets onto only the top or bottom surfaces of

the membranes can result in alignment problems. As shown in (a), if magnets are patterned onto

the top surface for stacking, the magnets create a gap between the membranes and the distance

between respective magnets becomes large, because of the membrane spacer. This results in an

advantageous linear alignment force approximation (section 2.2) and there is not inter-digit

mechanical problems discussed above. However, this also creates a significantly weaker

63
alignment force. For folding (b) the problem is even more pronounced. Every other set of

magnets come into alignment, which creates a significantly weaker attraction between the 2 nd

and 3 rd layers as shown in the picture. This is even more pronounced as the magnets from layers

1 and 2 match up poles and give a very negligible local external field.

a. b.

ext

C. d. Too small

->
Too lare
"LDBext BL--..
_
~ I -...
~ ...
z

e. f.

ext
Bext

Figure 2-13: (a)Stacking with nanomagnets on one side. (b)Folding with nanomagnets on one side (c)
stacking or folding with nanomagnets on two sides. (d)alignment errors from in-plane nanomagnets
being too short or long and solutions based on (e)embedded in-plane nanomagnets and (f) embedded
nanomagnet pillars

A technique to overcome this problem is shown in (c) with patterning magnets onto both

sides. This creates a higher alignment force for both stacking and folding. Additionally, the

intrinsic stresses in the magnets can be balanced across the membrane. A method for achieving
this double-sided patterning was developed in this research and is presented in Section 4.1. The

main problem, besides the additional processing requirements, is the magnets still must fit

together inter-digitally when in alignment. This results in the spacing problems shown in (d) and

discussed previously.

A promising approach, which overcomes nearly every problem presented thus far in the

theory is to embed the magnets within the membrane. This holds four primary advantages.

First, the magnets do not need to fit inter-digitally. Second, although the pull-in force is scales

by 1/d 2 , the alignment force scales with d. This combined with the first advantage, allows a

very forgiving processing tolerance and gives the full benefits of elastic averaging. The third

advantage is the membranes can now be in intimate contact. The final advantage is the distortion

from the magnets becomes smaller due to the change in moment arm in the cross-section force

balance. The primary disadvantage is an extra patterning step and a weaker membrane. The

other disadvantage is the field from the magnets becomes reduced as they match up with the

membrane below them, although not to the extent of inter-digitally fitting magnets.

Nnanomagnet pillars (f) can be used in select membrane assemblies in which the membrane is

thicker to reduce this problem. In this case, the local field increases with each additional layer.

A method for embedding nanomagnet pillars in polymer membranes was developed for this

research and is presented in section 4-2.

2.2. Membrane mechanical forces

The previous section detailed the forces on nanomagnet arrays from an external field and

from a mating array in close proximity. The importance of the forces and torques are directly

related to the mechanical forces involved in the alignment process. There are three primary

65
mechanical considerations when designing the nanomagnet alignment approach. The first is the

distortion from the stressed metal material deposited on the membrane presented in Section

2.2.1. Next is the compliance of the hinge for folding presented in Section 2.2.2. Finally, the

translation compliance is considered in 2.2.3. The translation compliance from a thin, low

modulus membrane or patterned flexures allows the nanomagnet forces to dominate the

alignment step.

2.2.1. Membrane distortion

Evaporated metal typically is deposited with a strong tensile stress due grain formation. This

is a well understood problem in MEMS that can be used purposely for applications such as

membrane folding24. Cobalt, Nickel and Iron nanomagnets can be easily electron beam

evaporated and patterned using standard lift-off. For example, Riesenberg3 4 showed that cobalt

is deposited at a roughly 2 GPa tensile stress when deposited at 40nm at a rate of 2nm/s. Jergel35

showed stresses up to 10 GPa were possible with annealing. As a comparison, the chromium

used to fold silicon nitride membranes in Nanostructured Origami was deposited at 1 to 2 GPa.

Therefore, it becomes critical to consider the stress and distortions of the magnetic materials

upon deposition. The first option is to act to minimize the stress. Jergel showed that annealing

steps can be used to reduce the residual stress in the cobalt and even put it into a compressive

residual stress mode. However, this is not a well-controlled process and is dependent on a

number of factors including: deposition rate, annealing time and temperature, material thickness,

membrane material and alloy formation at boundaries. A method to balance the stresses in situ

may be possible using feedback during annealing, but thermal cycling steps are necessary,
making the process less reliable. Further, the author found no equipment is readily accessible for

that exact purpose at the membrane sizes being used.

In this research, a more direct approach was used minimize the distortion. The membrane

plate modulus Em, thickness tm and width wm as well as the nanomagnet residual tensile stress

a-, and plate modulus En' were used to determine the acceptable magnet thickness and percent

areal coverage. The radius of curvature assuming 100% coverage of the magnetic material is

given in Equation 2.20 from Arora 24 . For the case of silicon nitride membranes used in this

research, the plate modulus is given as EmSiNx = 326GPAand the plate modulus of evaporated

cobalt is En,co = 232GPA.

= Ektir+E t+2EmnEntmtn-(2tn+2tn+2tmtn)
6ErEntmtn(tm+tn)-( )

Assuming that the nanomagnet thickness is much smaller than the membrane thickness and plate

modulus are relatively similar, Equation 2.2 simplifies to:

pPn<tm Emtm (2.21)

stnan
 V

Schematic:
Membrane distortion dependency on magnet thickness
- -- (100um $LNMmembrane, 5% and 10% cobalt magnets)
- m
- --
- -
- m m 0.2 5m- -
- -
0.5 pm -
1pVm

04 / tm=2 pm
Lb

1.de-0
0 10 20 30 '40 50 60 70 80 90 100 eOj
tlm mmll
O 20 40 60 so 1O0120 W T 20D
o80 s
Nanomagnet thickness, t,,(nm) fora given e,y
tm, E'I

Figure 2-13: (a)Schematic of nanomagnet coverage and bending from nanomagnet stress. The
nanomagnets cover a percentage ex and Ey in the respective directions resulting in a radius of
curvature p. (b) The resulting bending distortion for 100 pm membrane at (0.25, 0.5, 1 and
2pm) with 5%or 10% coverage and the
corresponding nanomagnet thickness on the x-axis.

Referring to Figure 2-13/14(a), the magnets do not cover 100% of the membrane, but rather a

fraction eX and ey in the x and y directions. It can be shown that the radius of curvature for the

patterned membrane is well approximated by substituting in Un,eff = eX Eyon assuming

uniformly distributed magnets with length l, «L «p The resultant out-of-plane distortion is

given in Equation 2.22. This is plotted out for the case of silicon nitride membranes and cobalt

nanomagnets in Figure 2-13/14(b). Given two different percentages of magnet coverage 5% and

10% it shows the out of plane distortion for a given magnet and membrane thickness

combination. For example, for a 1pm thick membrane, and 5% magnet coverage, the magnet

thickness should be less than 1OOnm for a distortion out-of-plane less than 200nm.
 (4Emtm

(2.22)
Sz= p(l - cos(L/2p)) 3t Ex
U
Z
EyG

The effective change of membrane length from bending is given by:

Lb sin (fl)p Y22 (2.23)

-2pLL = 1- = 1 - 3tn (223
(pL = L L L/2 24p
2 2U2
2E,2 t4 2.3

2 2 3
3tAE6X2 A
ALb -
2 4
E t (2.24)

The change in length from bending ALbend is negligible. For example, take the same lim

thick 100pm long membrane with 5% magnet coverage and 100nm thick magnets. The bending

from the bilayer stress results in a reduction of membrane length of only ALbend = 1.4nm. The

other form of in-plane distortion is from the residual stresses in the magnets and membrane given

in Equation 2.25 where am is any residual stress in the membrane. Again, taking the same

membrane example with a stress-free silicon nitride membrane, the in-plane distortion from the

magnets ALres = 3nm. Conversely, if the nitride membranes have residual tensile stress of

25mPa, the resultant change in membrane length combining magnet and membrane stresses is

ALres = 10nm.

ALres = ALr,mag + ALr,mem - L (tnex Eyo-n + tmom) (2.25)

Emtm

Overall, it is critical to consider the residual stresses in the membrane and nanomagnets when

using the magnetic alignment approach. Any gains made in alignment could quickly be lost in

membrane distortions if the above equations are not considered. If the magnets are considerably

thinner than a low stress, high modulus material, there likely is a solution with minimal

distortion. If not, the suggested approach would be the embedded magnets presented in Section

2.3. In that arrangement, the moment arm from the tensile stress of the magnets is near the mid-
69
plane of the membrane resulting in a considerable reduction in distortion, even with thicker

magnets.

2.2.2. Membrane folding

The approach of using nanomagnets and an external field to fold the membrane requires a

compliant hinge region to allow folding along prescribed lines. There are effectively three

primary hinge options for folding at the MEMS or NEMS scale shown in Figure 2-14. The first

option is similar to a door-hinge, in which a pin rotates in a mating feature. The second is beam

bending in a weakened area of the membrane. The third approach is a torsional hinge, which

was used in this research for a variety of reasons. The main considerations, when comparing the

various hinge options are: 1.) Ease of patterning 2.) Compliance 3.) Robustness 4.) Folding

predictability 5.) Susceptibility to fracture.

 a Joint hinge

b Beam Bending hinge

C Torsional hinge

Figure 2-14: Overview of hinge types used in microtechnology. (a)Based on a patterned joint, (b)hinge
based on beam or plate bending, and (c)based on one or more torsion bars. Examples are given for
7
(a)27, (b)6 and (c)1

The use of a joint-hinge at the microscale has been common since many of the first MEMS

structures 3 8. Multi-layer lithography steps and sacrificial material removal can create a pin

connected to one membrane that fits into a mating feature connected to the edge of another

membrane. This is particularly easy with micro-plates that are 1Os of microns thick and multiple

micro tolerances since roughly eight lithography steps are used to make the hinge. The benefit

of the joint-hinge is it has theoretically the highest compliance since the only folding resistance

would arise from friction. This high compliance can also be problematic. For example, for
 origami folding, it is favorable that the structure has some structural rigidity so it stays nearly

paper-like before the folding sequence. Further, the membrane sizes used in the present research

are typically 50-1OOOnm thick, and no joint-hinges have been shown at anything near this

thickness scale. However, if the processing and structural rigidity problems are addressed, the

joint-hinge has favorable properties for the nanomagnet alignment system.

The beam-bending approach overcomes the structural rigidity problem and uses simple

processing steps that scale well to thin membranes. A weakened hinge area is created using one

or more of the following approaches: thinning the membrane material, removing material along

the width of the hinge, or using a lower modulus material in the hinge region. All of these

methods were used in the Nanostructured Origami approach when Lorentz forces were used to

fold thick SU8 substrates with thin gold hinges. Although more complex plastic deformation

may occur, the hinges can be estimated in the elastic regime to require a folding moment

Mb given in Equation where wh is the hinge width, th is the hinge thickness, R is the radius of

curvature, which can be estimated as the membrane thickness tm.

Mb = R EWhtt
l 2 tm
(2.26)

The moment to fold 1 pum silicon nitride membranes separated by a bending hinge is given in

Figure 2-15(a) for various hinge thicknesses and widths. Referring again to 100 pm wide

membrane, the moment required to fold hinges of 100%, 20%, 10%, and 5% of the hinge length

are shown. Comparing this to the 10 to 30 pm -yN moment shown in Figure 2-2 for the

magnetic moment, the hinge thickness must be well under 50nm for most of the hinge widths.

The processing in order to make a 10 to 40nm thick hinge area in a 1pm thick membrane is not a

trivial process. In particular, as shown in Figure 2-15(b), the hinge should be located in the
72
middle plane of the thicker membrane segments. This would require controlled processing on

both the top and back of the membrane. Assuming this is possible, the bending hinge provides

very limited translational compliance. This is not a problem if the hinges are perfectly located

relative to features on the membranes. However, in the scheme of the nanomagnet alignment, a

compliance is necessary as shown in Section 2.2.3. In order to achieve this compliance, the

bending strategy of folding would require additional flexures to be patterned into the membranes.

a.) 300 'b),

250 -
E
200 -

150 . w 100pm
E
20pm t
100 - 10pm
5pm
0 0 -

01
0 10 20 30 40 50 60 70 80 90 100

Hinge thickness (nm)

Figure 2-15: (a) Simulation of a bending-style silicon nitride hinge at various thicknesses and widths
showing the moment needed to folding 180 degrees and (b) illustration why a bending-style hinge
should connect at the center plane of the membrane to allow face-to-face contact.

Compared to the bending and joint hinges, the torsional hinge is much more common at

MEMS and NEMS scale due to favorable compliance, straightforward patterning and the ease at

which the folding is constrained to a single axis. It is commonly seen in rotating MEMS mirrors,

such as in the Texas Instrument DLP 39 . Referring to Figure 2-16, the moment to rotate the hinge

1800 is given in Equation 2.27, where G is the shear modulus, s is the arm width, t is the arm

thickness, and L is the arm length.

 = Gfst(w 2 2 (2.27)
Mrorsion
M~orson - 12L +t )

Figure 2-16(b) shows the bending moment for the 1pum thick silicon nitride membrane and

hinge at various hinge widths and lengths. As shown, the bending moment is higher than

magnetic torques presented in Figure 2-2 for the 100pm long torsion bar for widths larger than

30nm. Since the lithography and processing limit the width to over 200nm, the length of the

torsion arm must be considerably longer than the 100m bar. There are strategies to effectively

lengthen the arm with multiple connected arms as shown in Figure 2-16(c). For the various

geometries an effective arm length Leff is given. The two most simple forms are shown in (i)

and(iii), with a single bar. Compared to the geometry in (i), the geometry in (iii) is symmetric,

but the resultant moment necessary to fold is four times larger. In each case, multiple torsion

bars are added in series to create an n times longer Leff for an n times more compliant hinge.

This comes at the major drawback that the folding no longer takes place along a single line of

action but rather there is an effectively expanded hinge region. This results in a spacing between

the membranes at the preferred fold location using the traditional multi-segment torsion bars.
 d
Schematic of torsional hinge b.)
Magnetic torque required forma 1800~fold for
250
-+ +- s Folding

"To nZ L=100
"Torsion arm" 10
/2
Schematic Leff Er
0

4,. 0
L

L1/4

-V 100 200 300 400 500

I_4 _Hinge arm width, w(nm)

Figure 2-16: (a)Schematic of 1-bar torsion hinge (b)Model results of 1800 twisting of a lm thick bar at
the widths and lengths shown. (c)comparison of standard approaches of adding compliance to
torsional hinges with added bars and their effective length.

The problem of expanded hinge region was addressed in this research to create a new style of

torsional hinge, which I named the "star" torsional hinge for the shape it makes. Figure 2-17

shows multi-segment torsion bar sets with the traditional style and the new star hinge developed

for this research. At three segments, there is no advantage to using the star hinge as both result

in the same fold gap. However, when moving to four or more hinges the gap can be

significantly reduced. Equations 2.28 show the approximate gap sizes for the old torsional hinge

(2.28.2, 2.28.3, 2.28.4) and the star torsional hinge (2.28.2, 2.28.5, 2.28.6). The first subscript

represents the number of bars with the second indicating the style. Referring to Figure 2-17 for

3, 5 and 7 segments there is always a substantial gap for the traditional multi-bar format. The

star hinge with n>4, on the other hand, can be used to minimize the gap by minimizing Equation

2.28.5 or 2.28.6. For example, for the 5 segment hinge, the gap is zero when b ~ 1.6a. That
length set is a difficult geometry to realize, with the closest at b=2a resulting in a final gap of

0.22a for the star hinge compared to 2.1 a for the traditional multi-segment hinge. Further efforts

can be made to reduce the amount of twisting a given bar does, such as make it thicker or shorter

to create a zero gap assembly for the 4 or more segment star hinges. Therefore, an effectively

zero gap is possible with using 5 or more bars. Chapter 3 presents the processing results for

realizing the star hinge.

g1 = 0 (2.28.1)

93,0 = 93,S= 2a -sin (60*) (2.28.2)

9s, = 2a - (sin(36*) + sin(72*)) (2.28.3)

97,0= 2a - (sin(25*) + sin(50*) + sin(75*)) (2.28.4)

95,s = 2(a - sin(36*) - b - sin(72*)) (2.28.5)

97,s= 2(a - sin(250) - b - sin(50*) + c - sin(75*)) (2.28.6)

 Regular
n=3
Star a

Regular I

Star la b

omY

RegularI

n=7

Sta r 77 Ma lb Ic

Figure 2-17: Schematic of regular torsion bars in series and the "star" hinge arrangement
for 3, 5 and 7 bar assemblies. On the right, the resultant side view of the hinge after
folding is shown showing dramatic reduction in gap size for the 5 and 7 bar hinges.

2.2.3. Membrane actuation and displacement

The last mechanical consideration for this research is the compliance of the membrane

segments for the magnet array alignment. Section 2.1.2 presented the force between magnets

that were typically in the nano-Newton range. Given an array with 10-1 OOOnN magnetic force,

the membrane mechanics must be compliant enough to allow the initial magnetic pull-in and for

the alignment force to dominate at close proximity. Therefore, the mechanical requirements are

77
highly dependent on the magnetic force, the coarse alignment and the required fine alignment.

After these factors are determined, the compliance can be designed into the system using a low-

modulus, thin membrane or if that is not compliant enough, patterned flexures.

In the case of membrane stretching, first consider a rigid membrane with compliant regions

at each bounding edge. To get the first order approximation of the compliance, we first look at a

square membrane constrained on two sides and a force F distributed along its length. Referring

to Figure 2-18(a), the approximate compliance of a membrane secured on two sides is given in

Equation 2.29 for a deflection smaller than the membrane thickness. The results for various

membranes are shown in Figure 2-18(b) assuming EPDMS = lMPa,EPMMA = 2GPa,and

ESiNx = 300GPa,w = 100pm and L = 10pm. As is evident from the plot, using a low

modulus material PDMS can achieve compliances over 1 IN. However, the use of a medium

or high modulus material PMMA or SiNX results in compliance less than 3e-3 m/N, which is

insufficient for both pull-in and alignment compliance.

2L(1-v) (.9
Csq,2 = F wtE (229)

If a higher modulus membrane is used, there needs to be additional mechanisms to increase

structural compliance. This is a very common problem in MEMS design, and flexures based on

multiple connected bending bars have been developed for this purpose. MEMS-style flexures

range from simple single bar designs to dozens of bars. Further, the design is well understood

and there are great tools, such as the CoMeT 40 software developed at MIT for modeling flexure

systems to have specific compliance and degrees of freedom.

 a.) w L. 10+1
oe~o - PDMS
E 1.00-1
1 2 - ---
.Lo
co__PMM A

EE
top-?
1-0e41 1,00+2 1.Oe+3

Membrane thickness (nm)

w=10 pm and L=100 pm
Figure 2-17: (a) Schematic and model of a compliant membrane based on low modulus and/or thin
regions at the perimeter with dimensions shown. (b) The compliance of PDMS, PMMA and SiN,
membranes for w=10pm and L = 100im.

For multi-bar flexure systems, the compliance is not simply additive. There is typically a

complex interaction between flexure bars, which limit the compliance and degrees of freedom.

CoMeT and Solidworks were used to visualize and model these multi-segment flexures. One of

the most basic flexure geometries that gives high compliance and full translational degrees of

freedom is shown in Figure 4.11 as modeled using the CoMeT software. Comparing this to the

compliance in Figure 2-18(b) it is evident that the addition of flexures is a significant benefit for

compliance, particularly if constrained to a high modulus material such as silicon nitride.

Now returning to the hinges designed in Section 2.2.2 and Figures 2-16 and 2-17, they also

clearly can serve as translational flexures when in the folded position. The compliance of the 3

and 5 bar star hinges were inspected using the CoMeT software for lpm thick silicon nitride

flexures for 100pm bars and widths ranging from 0.1 to 1 pm showing compliance orders of

magnitude above the necessary compliance. Therefore, for the case of magnets serving as the

folding actuator and alignment features, the more critical compliance design is for folding, since
the translational compliance will be sufficient. This is due to more favorable scaling of inter-

magnet forces compared to torque from an external magnetic field.

The membrane compliance results presented above are for small membrane segments in the

100pm range with compliance at the outer perimeter through either low modulus material or

flexures. This assumes that features are aligned everywhere within the small membrane. In the

case of larger stacked membranes with dimensions in millimeters, there will be distortion or

patterning errors over the whole membrane. Instead of assuming a rigid membrane with

compliant perimeter, the entire membrane must be made compliant by either a low modulus, thin

membrane or an array of flexures. This is a novel approach to membrane stacking or compliant

mask alignment and is presented in more detail in Section 4.2.

Chapter 3

Nanomagnet actuation and alignment experiments

3.1. Process design

Processing on nano-membranes introduces a number of problems that are not

encountered in regular wafer level processing. The membranes distort, fracture and are

contaminated substantially more than single-surface processing of wafers. A significant amount

of the work in this research was in improving membrane processes to make quality test samples.

The NanoStructures Laboratory at MIT has built a strong understanding of silicon nitride

membranes mainly for use in X-ray lithography masks and more recently for the Nanostructured

Origami process. I utilized much of this expertise to develop new silicon nitride membrane

processes. Additionally, the material used for the nanomagnets needed to be compatible with the

membranes and equipment readily available. Section 3.1.1 walks through the magnet and

membrane material choices used for the experiments. Section 3.1.2 presents the old and new

membrane processes developed for testing the magnetic actuation and alignment.

3.1.1. Materials

In typical permanent nanomagnet applications, the magnet coercivity and remanance are the

critical parameters. The coercive field dictates the amount of external H-field that is necessary to

change or switch the internal magnetization. This is critical for applications such as a permanent
magnet in a motor. If the coercive field is low, the internal magnetization would switch with the

field instead of creating a torque or force on the magnet. Remanance, or the amount of

magnetization when the magnet is not in an external field, is critical because it dictates the force

that the magnet can apply to a nearby magnet. Both of these properties are well understood and

are important parameters in nanomagnet systems- particularly coercive field as this dictates the

field required to switch a bit and its effective stability. Much research has been done to control

these properties at the nano-scale . Shape anisotropy helps to increase the remanence and

coercive field and material deposition processes have been developed for more complex alloys42

Another important magnet parameter is the magnetic saturation, which is the highest induced

magnetization state reached by the magnet in an applied H-field. The ratio between magnetic

remanence to saturation known as "squareness" can range from 0 to nearly 1. For example

Farhoud et a143 showed 80nm tall and 80nm diameter Cobalt nanomagnet to have a squareness of

roughly 0.3 with identical nickel magnets having a squareness of nearly 1. However, Ni shows a

saturation roughly one-third of the Co. Their coercive fields are both less than 750 Oe. These

properties are important for the designs outlined in Section 3.1.1.

The alignment method requires the patterning of an array of nanomagnets on each of the

membrane faces at some point during the membrane fabrication process. For easiest processing,

nanomagnets can be defined with e-beam or optical lithography followed e-beam evaporation of

a magnetic material and resist lift-off. The magnetic material usually consistes of one or more

layers of iron (Fe), cobalt(Co), nickel (Ni), and their alloys. An additional adhesion layer (in this

case 3-5nm Cr or Ti) and capping layer (3-5nm Au) to protect from oxidation may also be

necessary. The nanomagnet alignment scheme may take one of two forms", as shown in Figure

3-1. In form A, the magnetic material is patterned onto the membranes and an external field is

82
applied before folding (i). The external field initially aligns the magnetization of the hard

nanomagnets in the same direction. The field is then removed and the magnets hold some

magnetic remanance (ii). Upon folding, the poles are oppositely aligned and brought into close

proximity, thus inducing a magnetic attractive force between membranes (iii). The difficulty of

this method is that creating a nanomagnet at this scale with high remanence and coercive field

can take additional difficult processing. Further, a high coercive field may not be able to reduce

the local effect of the nanomagnets in close proximity effecting each other's internal

magnetization orientation. Figure 2-1(e) showed the coupling between Co nanomagnets in close

proximity. This occurs because the local field produced by the neighboring magnet can shift

internal magnetization, which thus reduces the strong necessity to align the magnets perfectly

along their axis.

 Form A

Bex

Form B

B i

Figure 3-1: Two forms of the alignment method. In form A, hard

magnets are magnetized in the same direction before folding. In
from B, soft magnets are magnetized to saturation during folding.

The second alignment strategy considered to overcome local coupling and provide a higher

alignment force is shown as form B in Figure 3-1. I this case, a magnetic material with high

magnetic saturation (e.g., pure iron) is patterned on the membranes (i). After the membranes

have been coarsely aligned via folding (ii), an external magnetic field is applied so that the

magnets are magnetized in the same direction(iii). In this method the magnets work at

saturation, providing forces stronger by as much as ten times compared to form A. Additionally,

there is more freedom in nanomagnet material and geometry, since high remanence and coercive
field are not required. Finally, the external field can be used for the folding as it applies a torque

on the membrane that is free to rotate.

For the experimental design, the strength of the external field affected the choice of

nanomagnet material, size and array. Figure 3-2(a-c) shows the final experimental alignment

process based on folding with a rotating external field that both magnetizes and applies a torque

on the nanomagnet array. A number of options for a rotating external field were considered.

The first was a two or three axis electromagnet. By controlling the field from orthogonal

electromagnets, an effective field can be produced. The main problem with this approach is the

set-up is expensive and bulky. The electromagnets in close proximity will have a strong

attraction. Further, the fields created are typically limited by the core material (e.g. iron), unless

a higher field (>1T) electromagnet, such as used in the MIT Magnetic Materials and Devices

Group is used. In the case of the higher field electromagnet, the field can effectively be rotated

by rotating the sample. However, it was found in this research that a single strong permanent

magnet was preferred for the folding and alignment for two reasons. First, the compact size and

inexpensive materials enabled this to be done in the Nanostructures Laboratory, close to the

SEM for multiple folding iterations. Second, the folding could easily be viewed under the

microscope in a modified setup. Multiple sizes of the N52 grade Neodynium rare earth magnets

(Mr=1.3T) from K&J Magnetics were used for various applications. In the case of viewing the

folding under the microscope a small 0.5" cube was rotated around and under a modified

microscope stage to not hinder the view. The equipment setup for the folding experiments is

shown in Figure 3-2(d) and discussed in Section 3.2.2. Using the permanent magnet approach,

the B-field applied to the magnets was in a range of 0.1 to 0.3T controlled to +/- 0.02T.
 -'~d)
Roa ~2- 1. Folding
a) Nanomagnets

Plastic1-axis
Hinge Membrane segment rotating stage 25 X25 X 12mm
16 sample ch p SmCoPermanent
b) c) o X 8mm Magnet

Folding Attraction' -- : -

/2. Measurement 3. Ufldng

Rotating external SEM M2mm

SmCoPermanent
Magnetic FM+ --- t-i- Magnet

Figure 3-2: The experimental approach to membrane folding and alignment. (a)Magnets are patterned
onto two segments of membrane separated by acompliant hinge. (b)They are placed in an external
field which magnetizes the nanomagnets in-plane and then rotates to fold the free membrane segment.
Upon folding nearly 180 degrees, the nanomagnets attract to the other array and self-align. (d) The
experimental process for folding based on a rotating stage near a permanent magnet followed by
measurement in the SEM and unfolding with a magnet in close proximity.

When choosing between Fe, Ni, and Co e-beam evaporated nanomagnets a few key

characteristics were required. The 0.1 to 0.3T B-field from the N52 magnet is sufficient to

overcome the coercive field of all of the materials. Therefore, the magnetic saturation and ease

of processing become the most important characteristics. To show that the external magnetic

field could apply a torque an array of nanomagnets, Ni nanomagnets were fabricated. Nickel

was used because it is the easiest of the materials to e-beam evaporate and it is more resistant to

oxidization. It was shown that a 200nm thick silicon nitride cantilever patterned with various

sized 60nm thick Ni nanomagnets could be actuated with an external field. However, since the

Ni magnets had a low saturation Ms,Ni = 0.6T, the actuation of the membrane was only slight.

In order to increase the deflection distance, Fe was next material that was experimented with.

The Fe nanomagnets showed significant increase in deflection, although the results were

unpredictable over time. It was later determined that the Fe nanomagnets were oxidizing over
86
time and losing their net magnetization, even with Au capping layer. Cobalt nanomagnets were

chosen as a compromise between Ni and Fe. They show relatively high magnetic saturation

Msco = 1.8T but are not as susceptible to oxidation as Fe. However, it was shown that an Au

Capping layer was necessary to reduce oxidation.

A cantilever deflection experiment was used to confirm the magnetic properties of the

materials before stepping into other experimental design. The magnetic pole strength of a

1 x 5pm x 60nm nanomagnet was derived by actuating a cantilever patterned with an array of

200 nanomagnets. Figure 3-3 shows a side view of the cantilever (a) without an external field

and (b) with an external field at 45*. When the field is applied, the cantilever deflects by a

distance 6. The moment required to bend this distance is given by Tbend = 2EIL-2 , where L is

the distance from the cantilever base to the magnet array, E is the elastic modulus, and I is the

are moment of inertia. The torque on the magnet array is given by Tmag = NBextPlsin(a),

where N is the number of magnets. Setting the bending moment equal to the moment on the

array, we calculated the magnetic saturations in-plane for the nanomagnets as Ms,Fe,IP = 1.6 +

0.2T and Ms,Co,Ip = 1.4 + 0.2T. These are somewhat lower than the literature values of

Ms,pe = 2.2T and Ms,Co,Ip = 1.6T. The small discrepancy was likely due to magnetization being

slightly out of plane and/or possible oxidization in the Fe magnets.

 (a)

Figure 3-3: Side view of 400nm thick SiNx cantilever patterned with
nanomagnets without an external magnetic field (a) and with an
external field at 45 degrees (b).

The membrane material choices fall within two categories: rigid low-stress materials and

high compliance but stressed polymers. For the initial magnetic folding and alignment

experiments, silicon nitride was used because of its low residual stress and high rigidity.

Membranes with thicknesses of 50nm to 1000nm and 0.1 to 10mm width were used within the

course of my research. Fracture becomes the biggest concern, although all of these sizes could

withstand the final process shown in the next section. The nitride can easily be made into

membrane form by selectively etching away the underlying silicon with a KOH etch.

Furthermore, the silicon nitride provides adequate process freedom since it is very different than

the other material sets used in the experiments: chromium, photoresists, conductive gold layers

and the magnetic stack. In contrast, a polymer membrane would likely run into some material

process compatibility with some photoresists. Finally, silicon nitride membranes can be used for
optical elements and have similar enough characteristics to silicone, poly-silicone, and other

semi-conductors as long as the stresses are accounted for. Therefore, the results of this research

can be applied to other material sets.

3.1.2. Fabrication steps

The first full fabrication process that I did was designed by Will Arora to show a

NanoStructured Origami proof of concept24 . Figure 3-4 shows the process and some of the

results of the fabrication. A 200nm low-stress LPCVD silicon nitride layer was patterned into

various cantilever shapes via photolithography and a CF4 reactive ion etch (i). The gratings

shown in the figure were patterned via interference lithography and aligned photolithography

(ii). An 80nm CF4 reactive ion etch was used to transfer the pattern into the grating, followed by

a piranha clean (3:1, H2 SO4 : H2 0 2 ) and organic clean (5:1:1, H2 0: H2 0 2 : NH4 0H) to remove the

photoresist and other organics. Positive photoresist was patterned to define the stressed-hinge

areas. Next, 120nm of Cr was deposited via e beam evaporation and the photoresist was lifted

off (iii). The sample was placed in a 70*C KOH bath for 4 hours (iv). This was followed by a

15min isopropanol rinse to reduce surface tension. After removal from the isopropanol, the

sample folded due to the prescribed hinge stress. Although the more robust features had high

yield, the features more susceptible to surface tension had less than 5% yield. This is an

important limitation to this wet release process, although there are strategies such as super-

critical drying, which reduce the effects of surface tension. The bridge structure in Figure 3-4

shows some of the benefits and limitations of this approach. The nanopatterned membranes

indeed controllably folded, but the folding is uni-directional and the radius is rather large in

comparison to the thickness of the membrane. This was later address in Arora's work on ion

89
implantation and also with the magnetic folding with compliant hinge approach in this thesis

work.

SiNx (b)
(0) si
SiN

0
Cr

Figure 3-4: (a) Process for creating SiN origami structures folded with a stressed Cr hinge. (b) Resultant
structures with nanopatterned gratings.

The initial magnet experiments were an extension of the membrane bilayer folding work.

The membranes were replaced with 400nm thickness LPCVD silicone nitride. There was no

nano-patterning of the membranes (ii) and the chromium deposition step was replaced with

photo-lithography definition of an array of nanomagnets. After defining the array, 5nm

Au/60nm Ni/3nm Cr was evaporated onto the membrane and the photoresist was lifted off. After

90
KOH release and drying, the membranes showed minimal deflection in a magnetic field below

the necessary measurement. It was evident from these steps that a significantly more compliant

membrane structure and a non-wet release would be needed for the magnetic folding approach to

test the alignment.

In order to create a dry release and ensure that the Co or Fe magnets are not attacked in the

long KOH etch, the process shown in Figure 3-5 was developed. First, photoresist was spun on

the top surface to protect the wafer while processing the backside. Next, 30nm of Ni was e-beam

evaporated on the back surface of the wafer. The array of membranes was patterned in

photoresist using various arrays as shown in Figure 3-6. The membranes were patterned with

additional width to account for the 54.74 degree etch angle of the KOH step. Further, it was

found over time that splitting the sample in sets of N x N arrays where N=2,3,4 was the most

convenient for the handling necessary in subsequent steps. The sets were made easier to separate

by the addition of a spacer channel patterned roughly 100pm wide that provides a clean cleaving

line when separating the samples. It was found that the lines should not be continuous but rather

should end in the corners as shown to keep the KOH etch from attacking the corners of the

samples and making an overall weak structure for handling in subsequent steps. A gap of 200

microns at the corners of the samples was sufficient to allow a strong enough wafer for handling

but allow easy cleaving when that step is necessary. After backside patterning, the membrane

pattern was transferred into the Ni layer, followed by a CF4 reactive ion etch through the

backside nitride (i). It was later discovered that for thin nitride layer (50-200nm) the nitride

layer was not necessary and the photo-resist had enough etch selectivity in the CF 4 etch. Further,

it is not problematic if the backside nitride is etched somewhat as long as there is at least roughly

25nm of backside nitride remaining before the long KOH etch. After the nitride etch, the top and
91
bottom photoresist are removed and the samples are placed in a 7 hr 70'C KOH bath to produce

silicon nitride membranes (ii). Multiple DI water rinses were done to fully dilute any KOH

remanence. For critical processes, the KOH etch was followed by an organic clean (5:1:1,

H2 0: H2 02: NH 4 0H).

(a)
SiN, (b)

Si
NWNSiN 1

@1 mOm

i- L

Q
Co
ia E 100 pm

Backside

0
1 ---- 7 "-L-

0
URN

Figure 3-5: (a)Process for fabricating suspended SiNx membranes with patterned flexures and
nanomagnets based on afinal release step (v) using a backside RIE etch. (b)Optical micrographs of
devices made with this process. The first two show shattering from stresses. The third is before the
final release step and the fourth and fifth show the bottom and top of a released but stressed
membrane with flexures.

After creating the set of membrane arrays, the proper handling of the samples becomes very

important since many processing steps must be made. The first issue is spinning resist onto the
membranes since a vacuum chuck cannot be used and the membrane can be prone to breakage or

distortion. I was given advice from other NSL researchers to tape the wafer with membranes

onto a support wafer that can be attaches to the vacuum chuck. Although this process works for

large samples, it introduces what appears to be pressure sensitive adhesive contamination, creates

thick points in the resist near the tape and requires significant sample handling with tweezers

before and after spinning. This was particularly problematic when smaller samples needed to be

spun. To eliminate these problems I developed a strategy that has become commonplace in the

NSL for membrane spinning. The initial attempt was to use double-sided tape. However, the

adhesive transfers to the sample, and the tape does not have enough compliance to easily add and

remove the sample. On the other hand, the gel pack sample holders have been developed with a

low tack silicone pressure sensitive adhesive slab that does not transfer to the sample. Samples

can be easily placed and removed from the silicone with minimal contamination. Slabs of the

silicone were cut from the cases and placed on roughly the center of dummy spin wafers.

Initially, two slabs were used and the samples was bridged across to eliminated any pressure

effects on the membrane. However, it was later found that this bridging was only necessary

when spinning resist onto the backside of the membrane. Therefore, silicone slabs of roughly 1

to 4 cm width were placed on the spin wafer and were reused up to 50 times. Instead of aligning

the wafer to be centered on the spinner, the sample should be centered or else centrifugal forces

can cause it to dissattach from the silicone pad.

 p ---- 4-
C

Figure 3-6: Schematic of the mask patterned used to create the membranes. In this
case, each sample is a width W (typically 8-10mm) and has a 3X3 array of
membranes. The membranes are patterned at width p resulting in a final size m
because of the angled KOH etch. The samples are separated by a channel c=100pm
wide to allow cleaving along prescribed edges.

The spinning can occur on one array of membranes or on a set of membranes. For many

processes, the complete set of membrane arrays were spun as an initial layer before cleaving into

the smaller samples. This reduces spin time and contamination from cleaving particulates since

the top surface stays protected. If the backside surface is critical to keep clean, which was not

the case for the folding and alignment experiments, a backside spin should be done to protect the

back of the membranes. Separating the arrays into roughly 8- 10mm samples proved to be most

94
advantages as the tweezers could be used to grab the outside perimeter instead of pinching the

top and bottom surfaces adding more contamination and making it harder to remove from gel

packs. It was noticed that even if the back of the membranes are not in contact with stages (e.g.

in the photolithography setup) the non-protected backs of the membranes become contaminated

during the air drying step which passes contaminated on the back of the wafer onto the otherwise

clean back of the membranes. This is one significant disadvantage to the membrane-first

process.

In the case of the proof-of-concept experiments, the two patterns that must be produced are

the membrane mechanical segments and the nanomagnet arrays. These can be done in either

order. In Figure 3-5, the membrane segments were made before the nanomagnets (iii). A

photolithography contact mask was aligned on the wafer level with a set of membrane arrays.

The pattern was transferred into the resist, and then transferred into the silicon nitride with a CF4

reactive ion etch. The etching step only went partially through the membrane so that it would

stay structurally sound enough for additional processing steps. A typical depth would be 200nm

into a 400nm membrane. The nanomagnets were then defined using aligned optical lithography,

followed by magnetic material e-beam evaporation and lift-off (iv). Finally, the sample was

placed in a timed CF4 reactive ion etch on the backside of the sample to etch up to the point of

the original membrane segment etch.

After showing successful membrane actuation, both lithography steps were replaced by

scanning electron beam lithography (SEBL) for greater pattern flexibility, accuracy and

resolution. This required steps (iii) and (iv) to be switched so that the conductive metal from the

magnets could serve as the alignment features for the subsequent SEBL step. Further, it was

found that the release strategy of partially etching through the membrane and then releasing from
95
the back was problematic for multiple reasons. First, it created significant stress concentrations

in the membrane, which were sensitive to shatter during handling and the wet processes.

Secondly, the depths of the etches could not be perfectly controlled, so the final thicknesses of

the membrane was unknown. Finally, any contamination on the backside of the membrane

would be transferred into the backside etch step. For all of these reasons, the process shown in

Figure 3-7 was developed. Starting with the same membrane fabrication step (i) as the previous

process, a thin layer of chromium was deposited onto the membrane (ii). Next, the magnet

arrays are patterned onto the membrane (iii). Within the e-beam system, the membrane locations

can be seen with high enough voltage electrons, so that the initial magnet arrays can be roughly

aligned over the unreleased membranes. SEBL writes the pattern into typically 200nm PMMA

resist and no additional conductive layer is needed for the SEBL process, since the Cr can be

grounded within the system. Typical nanomagnet writes would involve one 8-10mm sample,

which contains an array of 9 or 16 individual membranes ranging in dimension from 150 to

500tm in widths. Each membrane in the array is patterned except for the two that are used to

align the SEBL system, since they become exposed during that process. Each membrane gets

two arrays of matching nanomagnets and at least one set of alignment marks for aligning the next

SEBL pattern. Magnetic material is deposited with a 3nm Ti or Cr adhesive layer, magnetic

layer and capping layer. In the experiments, the magnetic layer was 30 to 100nm Ni, Co or Fe.

The capping layer for this process needed to be thicker than the previous, since some of it is

removed in the final RIE release step. Step (iv) is an aligned SEBL write in PMMA. In the case

of critical alignment experiments, each membrane is aligned individually before its respective

write. Again, no additional conductive layer is required. The perimeters of membrane segments

were written to define the features. Any blank area with dimensions larger than 2 pm was "cut"

96
only at the perimeter as opposed to exposing the entire region. In some cases, this allowed

released sacrificial segments to fall. Stress relief cuts were used in later iterations to control and

direct any fracturing to occur away from the critical features during the final release. The PMMA

is developed and then a chromium etch took place next. This was a significant point of

problems, as the etch rates ranged by orders of magnitudes in seemingly similar setups on etch-

controlling samples. It was later proposed that the chromium was oxidizing in some samples

making a considerably slower etch rate. Once the oxide was etch through, the etch accelerates,

making the entire process quite uncontrolled. The solution for this research was to immediately

spin resists onto the chromium and to ensure that the etch-timing sample goes through the same

process history as the true sample. After etching, the PMMA resist was cleaned from the sample.
 (a) SiNx (b
Si

~
oW aWC
Cr

Cross-hairs

Nanomagnets Dot arrays Gratings

Figure 3-7: (a)Process used to create free-standing SiNx membranes with nanomagnets and sensitive
flexure hinges. The process isbased on athin top layer of Cr, which isetched away during the final
topside RIE etch.

The final step of the process was a dry topside CF4 reactive ion etch (v). The chromium

layer added in step (ii) and again in (iii) act as hard masks for the silicon nitride etch. However,

even a small thickness of Cr (e.g. 20nm) on a thick membrane (e.g. 1ym) causes too large of a

distortion on the membrane. Therefore, the Cr must be removed. Unfortunately, any wet

process to remove the Cr cannot be done after the full release of the membrane. Therefore, a set

of experiments varying etching voltage, power and time was done to leave the minimum amount

of Cr at the point of release. The rate of Cr etching was determined by determining the point of

surface reflectivity/color significantly changing when roughly 3-5nm is left. In the case of

98
1pm thick nitride membranes the original Cr layer was 15nm thick. A thicker Cr layer was

possible since the nitride layer can be over etched. However, if there is an appreciable amount of

Cr (e.g 1Onm) on the membrane at the point of etching through, the membrane becomes very

susceptible to fracture before the Cr is removed. On the other hand, if less Cr is used than is

necessary (e.g. 1Onm), this only creates a pitting problem on the membrane surface (as long as

the magnets are fully protected). In most of the experiments, I erred on the side of fully

removing the Cr to ensure no unwanted stresses are introduced. This improved and efficient

process enabled successful fabrication of not only the cantilevers and robust flexures of the

previous process, but also highly sensitive hinge features as shown in Figure 3-8 and discussed in

the following section.

3.2. Magnetic folding and hinge experiments

As discussed in Section 2.2.1, a magnet array that does not distort the membrane can only

provide a slight torque. Using the models in Section 2 and the process of Section 3.1, a

successful proof of concept of both the magnetic torque and hinge designs were shown. The

design is given in 3.2.1 with the experimental results in 3.2.2.

3.2.1. Magnet array and hinge design

After confirming magnetic deflection of relatively stiff cantilevers, various flexures were

fabricated as shown in Figure 3-5. The flexures were patterned using a contact optical

lithography mask aligned to the underlying wafer using the membranes as rough alignment

indicators. The process shown in Figure 3-5 was used to make 150nm to 300nm silicon nitride

membranes with various arrays of 3nm Ti/60nm Co/5nm Au nanomagnets. The arm width of the
flexures ranged from 5 to 25 microns and 1 to 5 flexure bars were per side. There were three

major observations. First, the magnet to membrane thickness ratio caused a level of distortion

that was beyond the tolerance for the alignment system. This led to the stress model shown in

Section 2.2.1 for designing future stacks with negligible distortion. The second observation was

that a number of the weaker flexures were shattering as shown in the Figure 3-5. This occurred

because of a combination of the stress from the metal material and the weakened membrane from

the partial etch step. This was alleviated with the process shown in Figure 3-7. The final

observation was that rotation up to 45 degrees for the most compliant samples could be achieved

within a magnetic field of 0.1 T. However, because of the need to greatly thicken the membrane

and/or reduce the magnet thickness, it was clear that the flexures were still too stiff for use in 180

degree folding. This was further confirmed by developing the models of Section 2 and CoMeT

flexure modeling.

(a))

Figure 3-8: (a)Free-standing star hinge based on 5 arms at less than 200nm thickness and 10 0pm width.
Close-ups show the detail. (b)After folding showing the form that the hinge takes with a close-up
showing elements of the star pattern.

100
 With the move to the process shown in Figure 3-7, it became easier to control the membrane

thickness and also utilize thicker (>600nm) membranes. The patterning, deposition and

properties of 50-75nm Co nanomagnets were well understood at this point so this thickness range

was chosen. Further, it was clear that at least 1Onm of Cr would be necessary on the top surface

to act as an etch stop for a continuous coating that could be locally etched away in a controlled

manner. At this point 1pm thick silicon nitride was used for the membranes for a combination

of three reasons. First, the thickness allowed the overall expected distortion from the

nanomagnets (Section 2.1.1) to be less than 20nm over the 100pm membrane. With sparser

coverage, as in the second alignment experiment, this was brought down to roughly 5nm with

70nm magnets covering only 4% of the membrane. Secondly, a thicker membrane was

necessary to fully remove the top Cr layer at the final CF4 etch step without etching a high

percentage of the underlying membrane. Third, the thicker membranes are more robust and rigid

for the spinning steps. This thickness of membrane is readily available as it is a common

thickness of LPCVD silicon nitride thin film.

After the thickness of the membrane and magnets were determined, the nanomagnet array,

membrane segment size, and hinges were designed. The process starts with 200im by 200im

membranes fabricated with a backside KOH etch. The released membrane segment size was

chosen to be a 100 micron square since the magnet array could be written in a single write field

in the Raith 150 SEBL system. In the initial experiment, nanomagnets were not patterned on the

membranes for first check for any fracture or collapse in the membrane or flexure arms. Twelve

different hinge designs were first inspected for structural integrity during the final release step.

Three samples were made for each hinge design. The hinges had 98ptm long torsion bars spaced

by 1 micron in an arrangement similar to Figure 2-17. The hinges were patterned to have
101
nominal widths of 250, 500, 750 and 1000nm and segments of 1, 3 and 5 bars were produced.

Surprisingly, every hinge type had at least one successful freestanding sample. Each sample was

confirmed using a probe on a micro-scope stage to confirm that the segment was truly released.

The experiment was repeated with adding two arrays of matching nanomagnets on each

membrane window before creating the hinge and membrane segment. The magnets were 3nmTi/

75nm Co/ 5nm Au with 1 x 5pm size spaced at a period of 1Opm parallel with the magnet and

5pm perpendicular. This results in an overall coverage of 10% magnetic material on the

membrane segment. Next, 12 different hinge designs were made using the "star" hinge pattern

presented in Section 2.2.2. The arms of the hinge were patterned to be 100, 200, 300, and

400nm nominal bar width. To look at various different star geometries, 3, 5 and 9 segment

hinges were created. Samples of the 200, 300 and 400nm bar-width at all 3, 5 and 9 segments

were successfully fabricated.

The samples were placed in the modified microscope stage presented in Section 3.1.1.

External N52 Neodynium magnets were used to actuate the membranes initially by hand

allowing 3 axis of rotational freedom. An experimental setup was made to measure the

rotational compliance of the hinges. The large permanent magnet shown in Figure 3-2b was

moved from 15mm down to 3mm away from the membrane in 1mm increments. The B-field

from the magnet was measured at these locations using a magnetometer and confirmed with

FEM magnetic modeling. The 5 segment star hinges with 200, 300 and 400nm bars were then

placed in the setup. The stage was rotated and if the sample did not fold 180 degrees, the magnet

was brought 1mm closer to create a higher field and effectively higher torque on the nanomagnet

array.

102
3.2.2. Experimental results and analysis

There were multiple important observations in the hinge fabrication and testing. The first is

the hinges were quite robust to the dry release process and the necessary handling. It was very

apparent that the membrane segments would actuate when moving the sample through the air,

such as is necessary to move the sample between the plasma etcher and microscope stage.

However, with cautious handling they would only open like a door and allow the flow of air

without breakage. A second interesting observation is the membranes sometimes would start

dislodged similar to the orientation shown in Figure 3-7(b). The membrane would be

mechanically stuck, but this could easily be fixed with using the external field to manipulate the

membrane back into the correct location. The hinges exhibit compliance not only in torsion, but

the membrane segment could be rotated in nearly all directions do to the addition flexure

bending. The external field did not have a strong enough gradient to see an appreciable force on

the membrane compare to the hinge compliance. However, it was observed in a fractured

segment that was not attached to the substrate that the external field could pull the segment off of

the substrate. That same sample could be "walked" around the substrate by placing the magnet

under the substrate and changing the angle of rotation.

The results of the compliance test are shown in Figure 3-9(a) for the 200nm, 300nm and

400nm star hinges using 5 bar segments. The results of the various samples are compared with

the calculated torque derived from Equation 2.4. As can be seen, the results generally follow the

expected curve but are shown to be 20% to 50% lower depending on width. The lower torque

result is likely due measurement errors in the widths of the torsion bars since their widest point

were taken. Secondly, the angle of the external field to the magnets internal magnetization

103
direction at the point of folding was assumed to be 30 degrees. This could have had a +/- 10

degree error resulting in a +/- 30% error in the measured compliance. A third potential effect is

the matching magnet array on the substrate may have caused pull-in before the 180 degrees was

fully folded. Finally, the B-field was only controlled to 0.01T. If future testing was ever

necessary, I would suggest repeating this experiment in an electromagnet with a more controlled

field and angle. I would not have a matching array of magnets on the substrate but rather some

sort of adhesion mechanism that keeps the membrane in contact after folding 180 degrees.

Processing results for hinge folding compliance

(a) 100
90-
E 80-
Calculated
2 70-
- 60-
60
50

- 40-

30 -

20 . Measured
z x

50 jim
200 300
Width of torsion arm (nm)

Figure 3-9: (a)Calculated vs. experimental results for flexure folding at various widths of the flexure
arms. (b)Example of the 9-bar star hinge being stretched to roughly 300im showing its elasticity and
compliance without fracture.

Observations were also made on the hinge translational compliance and elasticity. Figure 3-

9(b) shows an example of a hinge that was manipulated to be over 200 microns away from its

starting location and at an angle. Amazingly, the hinges not only did not fracture at such

104
displacements, but they showed complete elastic recovery when allowed to return to their initial

arrangement. The translational compliance was not directly measured since the external field

could not create a sufficient force. However, the modeling showed that the design was more

than an order of magnitude more compliant than what was necessary for the magnets to dominate

the alignment. Further, the results of aligned membranes be more than one period off (Section

3.3.3) showed that the hinges played a negligible role in the force balance of the alignment.

3.3. Membrane alignment using nanomagnet arrays

The progress on processing, hinge design and folding enabled an effective method to

bring two arrays of nanomagnets into close proximity". The folding provided a repeatable

means to coarsely align the arrays to within 2 microns and the hinge theoretically provided

adequate compliance given the expected forces of the nanomagnet array. Two full experiments

were run on the alignment. The first used the nanomagnets from the folding and hinge

experiment to show there is indeed attractive force between the arrays. This was a successful

proof of concept of the method with clear evidence of self-alignment. The second experiment

added in a number of improvements to lithography steps, nanomagnet design and introduced

simple optical features. This resulted in highly accurate alignment between the optical features

and a number of important observations for future improvements.

3.3.1. Experimental procedure

Membranes were created using the process shown in Figure 3-7(a). In the proof-of-concept

experiment, the same membrane (100pm x 100pm x 1tm SIN) and magnet arrays (5pm x

1.5pm x 75nm Co) presented in 3.2.1 were fabricated. Folding was first viewed and recorded

105
under a modified optical microscope stage with a manually rotated magnet used to actuate the

membranes. After the membrane segment folds roughly 160+/- 10 degrees, there is a clear

moment of pull-in. In fact, the moment that the array is in the required coarse alignment, the

alignment happens so quickly that I frequently describing it as "snapping" into alignment. The

membrane stayed magneto-statically latched once they were in alignment. Therefore, no

additional adhesion mechanism was necessary. However, if a large B-field over 0.3 T was

applied in the vertical direction, the magnets could be separated and the alignment procedure

could be repeated over 50 times without any plastic deformation to the hinge or visible changes

to the membrane/array. Being able to pull the membrane out of alignment and repeat the folding

was an important ability for the second alignment experiment.

After qualitatively confirming that the magnetic pull-in occurs under the microscope, a set of

samples were made to measure the alignment. A set of sixteen 200 jim membranes on a single

chip were patterned using the process described in Figure 3-7(a) with the thicknesses/geometries

of 3.2.1. Of the 16 membranes, 2 were used as sacrificial alignment features for aligning the first

SEBL step. The other 14 were fabricated into samples for the alignment test. Each had the 5-

segment star hinge with 180+/-20nm bar widths. Of the 14, there was a yield of 11 samples.

The 3 other segments fractured at points on the hinge during the final reactive ion etch step.

After fabrication, the sample was placed under a microscope on a rotating stage. A large SmCo

permanent magnet was placed within close proximity to the sample with a magnetometer

measuring the field strength. The B-field could be controlled to 0.4+/-0.03T. The nanomagnets

were magnetized to saturation and the sample was rotated relative to the external field. After

folding, the sample was placed within the SEM. Under high voltage (>20kV) both magnet

arrays could be roughly seen through the top membrane. Figure 3-10(a) shows an example of a

106
folded membrane magneto-statically held and viewed at 20kV in the SEM. As can be seen, the

SEM viewing is not ideal because the top membrane creates a low contrast image. In order to

improve the contrast, the sample was placed back into the reactive ion etcher and the top

membrane CF4 dry etched. As can be seen in Figure 3-10(b), there was a significant

improvement in viewing contrast. Regular viewing conditions (e.g. 5kV) were used in the SEM.

Alignment measurements were taken at 9 locations on each membrane. The alignment error

between magnets in both directions was measured. The results are presented in the next section.

(a) (c)

10pm

Figure 3-10: (a)SEM of a folded-over segment in alignment. (b)SEM of the sample after a topside RIE to
remove the upper membrane segment exposing the nanomagnets.

After the proof of concept experiment, a number of improvements to the experiment were

made to try to push the alignment down to that necessary for layered optical devices. Since the

hinges showed adequate compliance for both folding and actuation, the areal coverage and

thickness of the magnets could be reduced. In order to do this, the 5 bar star hinge was increased

to a 9 bar star hinge, still with 200nm bar width. The decrease in magnet thickness reduces the

distortion in the membrane and further constrains the internal magnetization of the nanomagnets
107
to be in-plane. The reduction in areal coverage of magnets reduces bilayer-stress distortion and

allows for space for the addition of other features. In this case, the extra space was used for

optical features and cross-hair marks to assist in the alignment measurement. The features were

patterned during the same lithography step as defining the membrane segment and hinge.

Therefore, the features were etched through the entire membrane. The optical features were

arrays of 500nm and 1000nm period gratings, as well as arrays of 500nm and 1000nm circular

voids as the precursor to a layered photonic crystal. Four cross-hairs were patterned onto the

membrane and near-by substrate. The cross-hairs on the substrate had arm widths of 800 +/- 30

nm and the cross-hairs on the membrane were 1000 +/- 30nm. The cross-hairs matched after

folding and alignment and the substrate's marks could be viewed through the marks on the

folded membrane. Alignment was measured between the optical features and not the magnet

arrays as in the previous experiment. Further, the top membrane does not need to be removed

for measuring the alignment, which could introduce some error by allowing individual magnets

to move. Finally, measuring between the actual optical features is more in-line in what would be

expected in a fabrication process utilizing this alignment strategy.

The most significant improvement to the prior experiment was controlling the lithography to

higher precision. The magnets again were designed to have a length 50% of the period such that

they fit inter-digitally together. In the previous experiment, only one e-beam dose matrix was

done for the magnets and was only done on the substrate. However, because of e-beam

proximity effect, the lengths of the magnets on the membrane are actually different than on the

substrate. Further, the magnets that are close to the edge of the membrane receive an overall e-

beam dose between that of the magnets on the membrane and substrate. Finally, the magnets at

the perimeter of the substrate receive reduced dose because of they are flanked by non-exposed

108
areas. Furthermore, the dose matrix only considered the pattern in the PMMA resist and

assumed that the pattern was perfectly transferred in the metal evaporation and lift-off steps. An

improved series of dose matrix iterations accounting for all of the regions was done. The matrix

measurement was done after evaporation and lift-off to more accurately indicate the final magnet

length. This combined with tighter resist process control allowed for approximately 15nm

uniformity of the lengths of 1 and 3 micron magnets over the membrane and substrate.

In order to further filter out lithographic errors in the samples, an additional step was added

to the experimental process. After fabrication, the samples were first folded and inspected for

contact. This was easily viewed under the microscope due to the interference effect between

membranes. Only the membranes that came within 100nm of surface-to-surface contact were

used in the experiment. The folding and unfolding process shown in Figure 3-2(b) was used to

measure repeatability of the magnetic alignment. The sample was placed on a rotating stage,

which had markings to ensure the sample was placed within +/- 3 degrees accuracy. This

ensured the external magnetic field was aligned within 3 degrees of the axis of the magnets.

After magnetization along the axis, the sample was rotated 225 +/-10 degrees on the stage. The

magnet was removed as to assure the sample would not be magnetized improperly at the point of

removal. The sample was removed from the stage and inspected within the SEM. The

alignment between membranes was measured using two sets of overlaying cross-hairs described

above. The alignment error in the x and y directions were averaged separately and combined as

6t = 6x,avg + 6
y,avg2 to give the overall alignment error.

Two magnet sizes were used for the experiment to explore if magnet size appreciably affects

the alignment. Two sizes of magnets, 1pm x 0.3pm and 3pm x 1Mm, were in arrays with 2L

109
spacing along their length and L spacing orthogonally, where L is the length of the magnet. Five

array sets were created with relative area covered by two magnet sizes in a ratio of roughly 1:0,

3:1, 1:1, 1:3 and 0:1. Samples were first folded to ensure that the magnet arrays could

mechanically fit together. After viewing the folding under the optical microscope, four samples

were chosen for the full alignment repeatability experiment. The folding, SEM and unfolding

process shown in Figure 3-2(b) was used. Starting from the folded samples, the permanent

magnet was brought into 1 mm proximity with the surface of the sample. Spacers were used to

ensure non-contact near the membranes. After this point, the sample was placed onto a rotating

stage. The stage had a gel PSA layer to hold and easily remove sample. An N52 magnet was

brought into close proximity to apply a B-field ranging from 0.1 to 0.4 T over the sample. 46
The

sample was magnetized using the external magnet, which was aligned to +/- 3 degrees of the axis

of the nanomagnets. The sample was rotated between 215 and 230 degrees resulting in contact

as shown in Figure 3-11. The bar magnet was carefully removed such that no unwanted

magnetic field affects the samples, while they are being removed from the stage. After folding,

each of the four samples was placed in the SEM and the alignment was measured as previously

described. The samples were taken out of the SEM and the unfolding and folding steps were

repeated. The time between unfolding and folding was less than 1 minute to reduce the potential

for particle contamination between measurements. This process was repeated 10 times for each

of the samples. The results are shown in the following section.

110
 N /
Optical Features Nanomagnets

Figure 3-11: Unfolded and folded origami showing circular void optical features coming into alignment
using the nanomagnet approach.

3.3.2. Results

Figure 3-12(a) shows the average alignment accuracy between nanomagnet arrays as

measured by the 9 points on the membrane for the initial proof of concept experiment. The

measured mis-alignment was consistent throughout the array with a standard deviation averaging

38nm parallel to the magnets' length (x-direction) and 29nm perpendicular. Therefore,

experimental factors such as rotational misalignment, membrane distortion, patterning

imprecision and measurement errors caused only moderate variance in the alignment. For the

ten samples, the average alignment error was 276+/- 101 nm in the x-direction and 125+/-41nm

in the y-direction. The reasons for the overall alignment error and anisotropy of alignment error

are presented in the following analysis section.

111
(a)Layer-to-layer alignment accuracy for 10 samples
o-250 -----------

E
c
200 Average

150 Aligned
Magnets / 4

.100

c 50
0

0 100 200 300 400 500

Misalignment in x-dlrection (nm)
(b Fol ding Hinge Substrate

Nanomagnets Folded membrane

112
 Figure 3-12: (a) Alignment results of the initial proof of concept
experiment for ten different origami samples. (b) Images of the
alignment before and after removal of the top membrane using RIE.

After the improvements for the second experiment, considerably better alignment results

were measured. The results are shown in Figure 3-13(a) for each of the four samples. The

points at 2000nm+ represent points that aligned to 1 or more periods off for the smaller length

magnet array. This only occurred for the samples A and B as the other samples presumably had

enough of the longer magnets to overcome the pull-in effects of the smaller magnets when the

period was off. These outliers were omitted when calculating the average and variance for

samples A and B. Sample A with only lpm x 0.3pm magnets had an average alignment of

41nm with a standard deviation of 15nm. Sample B with 75% 1pm x 0.3pm and 25% 3pm X

im magnets had an average alignment of 44.5nm with a standard deviation of 9.9nm. Sample

C with 25% 1pm x 0.3pm and 75% 3pm x 1pm magnets had an average alignment of 33.4nm

with a standard deviation of 9.3nm. Sample D had an average alignment of 27.6nm with a

variance of 9.2nm. Figure 3-11 shows the before and after SEM images of folding, with an

example of the circular voids aligning between layers. Figure 3-13(b) shows the same images

after etching away the top membrane in a CF4 reactive ion etch. As can clearly be seen for both

the magnet sizes, the arrays arrange inter-digitally and appear as a straight line. Further, the

alignment of the various optical features can be seen as etched, with an interesting feature based

on the aligned circular voids shown.

113
 (a) (b)
2000+

80

60 I features
CD60

C
40

0 Samples A Samples B Samples C Samples D

100% 0.3x1 pm 75% 0.3x1 pm 25% 0.3x1 pm 100% 1x3pm
25%1x3pm 75%1x3pm

Figure 3-13: (a)Alignment results for four different samples each folded and measure ten times. (b)SEM
images of the nanomagnets in alignment after removal of the top membrane using RIE.

3.3.3. Analysis

A number of observations could be concluded from the experimental results. The first is that

the in-plane magnet geometry results in highly anisotropic alignment accuracy. Referring to

Figure 3-12(a) the alignment in the direction of the magnet's length was accurate to 290nm

compared to 125nm in the other direction. This is due to the requirement of the magnets fitting

together mechanically. If there is a lithography error in the magnet length, the magnets will bias

to one direction resulting in a larger misalignment as shown in Figure 3-12(a). This was

improved with greater attention to electron beam doses across the arrays and with first testing the

arrays for mechanical interference as discussed in the previous section. A second reason for

increased alignment error along the length is the internal magnetization can be pulled out of

114
plane during folding by the external field and the other magnet array. Since the folding approach

requires a magnetic torque the external field is at an angle with the plane of the membrane at the

point of folding. This allows the magnets to align in a more stacked position, particularly if they

approach as such or are too long. This was clearly seen in the SEM images of the first

experiment after the top membrane was removed. In the second experiment, the ability for the

magnets to magnetize out of plane was reduced with two methods. The first was thinner

magnets (60nm vs. 75nm) were used. Secondly, a more compliant hinge was used so that the

external field was more in-plane at the point of coarse alignment. In other words the angle

between the external field and magnet axis did not need to be as great to create the necessary

torque. These improvements, along with greater lithography greatly reduced the anisotropy of

the alignment.

The same effect of the magnets not being magnetized perfectly along their axis appeared to

affect the final alignment perpendicular to their axis. In the proof of concept experiment, the

external field was not tightly controlled to be aligned with the magnet axis. This combined with

curved ends of the magnets likely allowed the internal magnetization of the nanomagnets to shift

near the poles. This effect was reduced in the second experiment with tighter control of the

external field (±30) and by squaring the ends of the magnet bars in the larger magnets. The

rounding of the magnet shape may also have caused lithographic errors in the magnet length to

be translated into alignment errors in the perpendicular direction (see Section 2.1.3).

The second experiment introduced cascading the magnet lengths to get the supposed benefits

of the larger and smaller nanomagnets. The proposed benefits were that the larger magnets

would give greater coarse alignment tolerance and the smaller magnets would give more poles

for improved quasi-elastic averaging (Section 2.1.3). The first benefit was shown

115
experimentally. Referring to Figure 3-13(a) both sample A with 100% smaller magnets and

sample B with 75% coverage of smaller magnets had alignment iterations that were a period or

more out of alignment. Three of the 10 folding repetitions of the sample A and one of the ten

folding repetitions of the sample B while none of the 20 iterations with the samples C and D

showed alignment a period or more off. The second proposed benefit of the cascading was not

shown experimentally. Sample A, with only smaller nanomagnets, had worse average alignment

with a standard deviation of l5nm vs. 9-10nm for the other samples that had at least 25%

coverage of the longer magnets. There may be a number of reasons for this unexpected result.

The first is that the smaller nanomagnets were more prone to rounding in their lithography so

they were not the ideal rectangle shape. Secondly, the average alignment may have been lower

due to lithography errors in the magnets or the alignment of the optical features. These results

suggest that larger, rectangle nanomagnets are sufficient for accurate alignment without needing

the smaller magnets for error averaging. This is discussed further in Chapter 5 with additional

suggested improvements and experimentation

3.4. Membrane reconfiguration using nanomagnet arrays

The nanomagnet alignment approach showed that in some folding repetitions, the alignment

was off by one or more periods. This was initially considered to be a problem in the proof of

concept experiment. Figure 3-14(a-c) shows three configurations that were seen in the alignment

for the same sample. The folding in (a) is the proper position, while the second two (b-c) are one

period (10pm) off in each direction. This alignment occurred when the coarse alignment was

off by manipulating the folding axis to be at least 5 microns away from the designed position.

This approach shows promising results for reconfigurable systems where highly aligned final

116
positions are required. However, the requirement to fold and unfold the structure to achieve a

new position is prohibitive for most actuation applications. A more applicable approach is to re-

align the nanomagnet magnetizations such that the shift occurs without needing to unfold". This

was shown experimentally as in Figure 3-14(d). When the nanomagnets were patterned to be too

long for alignment, they aligned to a very unstable alignment position. By rotating the external

field, the internal magnetizations shifted further out of plane causing the magnet array to actuate

by roughly the length of the nanomagnets. This was viewed and recorded under an optical

microscope. Upon rotating the magnet back into plane the nanomagnets were re-magnetized in

their initial orientation. This caused the magnets to shift back to the other position. This could

be repeated over twenty times for the sample.

117
 C d Rotating
External field

Nanomagnets Optical features

Figure 3-14: A membrane folding into the correct position (a) and off by one period in both directions
(b-c). (d) Schematic and micrographs of switching between two stable positions with a rotating external
B-field

Although this experiment proved the concept of reconfiguration, further designs enable more

controlled switching. Figure 3-15 shows a configuration of diamond shaped nanomagnets.

Upon simply rotating the external field 90 degrees, the magnets shift orientation and become

attracted to a different nearby magnet. This can create a controllable switcher with switching

distance 1Os to 1000s of nanometers with alignment accuracy similar to that shown

experimentally in this research. Instead of a bistable switcher, the nanomagnets may be oriented

to allow for a step-wise actuation. Figure 3-15 shows two sets of nanomagnet circles in contact.

118
Using OOMMF simulation, I have shown that the internal magnetization of the magnets can be

rotated causing a translational force on the array. By rotating the external field either clockwise

or counter-clockwise, the stepper will actuate either direction. However, the geometry of the

nanomagnets causes an arc-shaped actuation motion for each period instead of linear actuation.

This is remedied by having an additional stepper underneath with actuation going in the opposite

direction. Further, the effective step size in the system is the difference in period between the

two layers of steppers. Therefore, the effective step size is not limited by the lithography (e.g.

size of the circles). For example, if the top stepper is 500nm circles and the bottom is 51 Onm

circles, the effective step size of the system is 1Onm. Other more complex arrays of

nanomagnets with different switching fields can also be used to realize similar benfits.

(a) (C)

Bext-0.1 T

1 nm
ext

(b) (d)Membrane 1.-- Bx

Membrane 2 *
Rotating external
Substrate I magnetic field

Bext P
Bext BeA B
Figure 3-15: (a-b) Method to make a simple switching actuator using a rotating external field. (c-d)
method to create a stepping actuator using a rotating external field and circular nanomagnets. (d)
Shows two overlayed steppers based on this strategy that actuate in opposite directions.

119
Chapter 4:

Novel membrane processing for 3D nanofabrication

In the course of the research a number of new or improved membrane processes were

developed. The previous section presented processes for dry release of nanopatterned silicon

nitride membranes without significant fracture or stress distortions. It became evident that the

ability to pattern nanomagnets onto both sides of a membrane would be advantageous for greater

alignment forces and balancing of the nanomagnet stresses. Section 4.1 shows the design and

results for a SEBL and PMMA-based dual-side patterning of silicon nitride hinges. Although all

of the previously presented research was done on rigid silicon nitride, there is a great motivation

for using highly compliant membranes as presented in Section 4.2. In particular, for very large

area alignment of nano-membranes (e.g. 1 cm), local alignment can be achieved using the

nanomagnet approach and a "stretchy" membrane. Section 4.2 shows the results of embedding

cobalt nanomagnets in compliant PMMA membranes.

4.1. Dual-side membrane patterning

One of the primary drawbacks to the nanomagnet alignment strategy is the nanomagnets

typically introduce a residual stress. This will distort the membrane as modeled in Section 2.2.1.

This limits the thickness of the magnets and therefore requires high compliance, such as that

provided by the hinges of Section 2.2.2. One improvement to this is to embed the nanomagnet

into the center of the membrane. Another solution is to pattern exactly matching nanomagnets

120
on the top and backside with the same shape, thickness and stresses. Since the processes used in

this research start with a free-standing membrane, as opposed to releasing the magnets as a final

step, there is access to the backside of the membrane for processing. However, repeating the

processes of the topside of the membrane onto the backside introduces alignment error between

the two and does not guarantee the matching of nanomagnets. Additionally, membranes hold the

advantage that electrons of high voltage for SEBL can transfer through the membrane with

relatively little deflection and energy reduction. This depends greatly on membrane thickness

and material, but I show in 4.1.2 that 30kV to 50kV electrons can be used to write both sides of

50-400nm membranes with relatively little error. In order to write on both sides, PMMA was

spun onto both sides of the membrane. Fortunately, the sloped sidewalls from the KOH backside

etch enabled relief for the backside PMMA spin. Thicknesses and uniformity were much

different than seen in typical wafer spinning. The quality of the spin results were highly

dependent on membrane size and spin speed. The results are shown in 4.1 .1.

The process developed for backside patterning is shown in Figure 4-1. Starting with a silicon

nitride membrane (i), PMMA is first spun onto the topside (ii). Next, the sample is flipped over

and held such that the membrane is not in contact with the spinning wafer. PMMA is spun onto

the backside (iii). This is followed by adding a conductive layer of evaporated gold or

Aquasave for the electron beam lithography step. It was later found that the high voltage

electrons resulted in negligible charging since they would pass completely through the

membrane and resist layers. Next, high voltage e-beam lithography is used to write onto both

sides using a single write (iv). The patterns in the respective resist layers are perfectly aligned,

although may have contrast or resolution differences if the process is not controlled as presented

in 4.1.2. The sample is developed (v) after the write. Finally, metal (a multi-layer cobalt stack

121
in this case) is evaporated onto both surfaces (vi). The final step is a lift-off of both sides. This

step was proven to be the most difficult because of membrane fracture. Processing tips to reduce

fracture at this step are given in 4.1.2.

e- e-
SiNX

0 Si®

PMMA

0D

O~PMM A 0CwOr
Figure 4-1: Experimental process used for double-sided patterning on SiNx membranes based on
spinning PMMA onto both sides and exposing using high voltage e-beam lithography.

4.1.1. Resist spinning

To the Author's knowledge, resist has never been spun onto the backside of <1Im

membranes for the purpose of lithography. After a few attempts on small membranes (<2 5 0mm),

it became clear that the process works much differently than typical wafer spinning. A modified

sample holder was created to allow holding of the sample without touching and damaging the

membrane. Two pieces of thick silicone PSA from a gel pack sample holder were placed onto a

dummy wafer with spacing between PSA pieces of roughly 5mm. The gap allowed for the

membrane to bridge across, while giving easy axis to the tweezers. Although most fabrication

122
for this research had previously been done on arrays of multiple membranes, it was very evident

that only one membrane on a sample can be evenly spun onto using backside spinning. Unlike

typical wafer spinning, placing a small amount of resist in the middle of the sample did not result

in even spreading over the multiple membranes. Additionally, completely coating the sample

with resist before spinning also did not result in even coverage. A mask was made for creating

single membranes in a lcm chip with the cleaving strategy of 3.1.1. Membranes ranging from

100 to 5000pum were fabricated in silicon nitride thin film of 35 to 1000nm thickness.

At first, the process of spinning seemed more like an art than a science. The location of the

membrane on the spinning wafer greatly affected the final uniformity. If the membrane was not

centered, there would be clear non-uniformity of thickness form the resist pulling outward. In

general, it was found that the membrane needed to be centered to better than 1/ 10 th its edge

length to see minimal thickness anisotropy. For samples in hundreds of microns size scale, that

meant the samples needed to be centered on the spin wafer to tens of microns. Fortunately, the

membrane is a clear difference in color compared to the surrounding substrate. Therefore, upon

starting to spin at roughly 500rpm a ring became clearly viewable. The center of the ring was

the goal location of the membrane. The radius of the ring was the initial alignment error, which

usually ranged in millimeters for the initial spin. After viewing the ring, it became very intuitive

to lightly adjust the sample, similar to how one would do for wafer spinning. However, in this

case the wafer would be highly out of centering, since it is attached to the membrane sample and

only the membrane is of concern. At the point of nearing perfect alignment, the spinning wafer

was tapped, spun and repeated roughly 10 to 25 times until the bolded color of the middle point

is seen. Once the membrane was adequately centered, the resist was deposited directly onto the

membrane and a 60 second spin was done. Since the membrane is so small, there is overfill of

123
the membrane trench, but this was not shown to be problematic. The overall process of

mounting, adjusting and resist deposition was roughly 10 minutes per sample, depending on size.

Similar to topside resist spinning, the resist post-baking was done for a higher amount of time

then a wafer version because of the reduced thermally conductive path to the membrane. The

resist was baked on a hot plate at 180C for 2 minutes for each sample, regardless of membrane

thickness or size. There was no visible difference in the time for the smaller vs. larger

membrane to show signs of solvent evaporation, so it is likely that the heating happens from the

air under the membrane and not conducted along the membrane length.

Two solutions of PMMA in anisole solvent were made with ~1% and ~3% PMMA by weight

to explore the effect of viscosity on spin uniformity. Membrane samples were made of 400, 800,

1200, 1700, 2200, and 4 2 0 0pm edge lengths. The 400pm membrane samples were able to be

spun onto, but the dependency on a centered sample during spinning, resulted in these being

discarded for the experiment. Samples were spun at 1k, 3k, 5k and 7k rpm for each membrane

width and PMMA solution. Figure 4-2 shows samples of the optical micrographs looking at the

resultant spins for the ~3% PMMA solution. Since the thicknesses ranged in hundreds of

microns, the newton rings from interference are easily visible showing non-uniformity. Sample

(a) was an example of a completely unusable spin. The combination of a small membrane

(8 0 0pm) and moderate spin speed (3k) resulted in the surface tension forces of the resist liquid

in the trench overcoming the spinning forces. The thickness is greater than multiple microns.

Sample (b) is improved, but only a small inner portion shows a relatively uniform coloring. The

spin appears to be well-centered in the membrane, but there are at least 6 Newton rings. Sample

(c) is a smaller membrane, but a higher spin. The centering is not as good as (b) but a slightly

larger area shows uniformity. Sample (d) is the same membrane as (b) but at a 5k spin speed.

124
 Sample (e) increases the spin speed to 7k with improved coverage even at the corners. Sample

(f) is a 4.2mm membrane showing very even coverage out to the edges at the 5k spin.

Figure 4-2: Examples of backside membrane spinning using 3%PMMA in anisole on the following
membrane sizes and spin speeds: (a) 800 lpm, 5k spin, (b) 1700 pm, 3k spin, (c) 1200 hlm, 5k spin, (d)
1700 pm, 5k spin, (e) 1700 pm, 7k spin and (f)4200 pm, 5k spin.

Figure 4-3(a) shows the approximate usable area depending on membrane size and spin

speed for the ~3% PMMA solution. For most critical SEBL lithography processes, this can be

considered the initial blue-green portion of the resist. Staying within that color region gives
125
uniformity of approximately +/- 20nm, which is adequate for many SEBL processes. Higher

uniformity is expected, but is not measureable from the Newton ring patterns. Referring to

Figure 4-3(a), the higher the spin speed and larger the membrane, the greater the amount of

usable space. The only samples that showed greater than 75% usable area were the

[2400pm/7k], the [2800pm/7k] and the [4200ptm/5k]. Since the thicknesses could not be

measured directly with the Dektak or ellipsometer, due to the small size, a secondary approach

was used. PMMA was spun at 0.1k RPM multiples onto the topside of multiple dummy spin

membranes. The thickness on the substrate portion was measured with the Dektak and an optical

micrograph was taken through the membrane to document its color. The color of the middle

point of the backside spin was then matched to a corresponding topside-spin micrograph to

estimate the thickness. Although not a perfect approach, this likely measured the thickness to +/-

15nm accuracy and was sufficient to move on to the e-beam writing. The results are shown in

Figure 4-3(b) for the 3k, 5k and 7k spins at 1.2, 1.7, 2.2 and 4.2mm membrane sizes. Using the

1%PMMA in anisole resulted in only a moderately higher amount of useable spin area. Figure

4-4(a) shows the same 17 0 0 pm membrane spun at 3k and 5k for the two solutions. The 1%is

approximately 50nm, so the Newton rings are not as evident until closer to the membrane edges.

Figure 4-4(b) shows a rough estimate of the combination of spin speed and membrane size that

result in 80% or higher uniformity for the two PMMA solutions. The plot provides a rule of

thumb, so a test sample should be done before committing to any membrane width, spin speed

and PMMA solution.

126
 a.)100%
b.) 240
1700pm
.a A 220
e75%
7 -
2200pmn
A - 200
- - 5k E 4200pmn
50%
o> L180 -No trench -
Ca 16 1200pm
w 25%
a.

120 .-

2500 5000
100
Membrane width (pm)
3 4 5
Spin speed ( x~jk6rpm)7 8

Figure 4-3: (a)The experimental results showing the percentage of usable area on a membrane
depending on the membrane width and spin speed for 3%PMMA in anisole. (b)The thicknesses at the
middle of the membrane for various membrane sizes and spin speed for the same resist.

(a) (b)
8000r

7000 -

6000

Lower viscosity
5000 -

3k spin 4000-
Okay
3000

2000- Not okay

1000
1%

0 1000 2000 3000 4000 5000 6000 7000 8000

5k spin
Spin speed (rpm)
Figure 4-4: (a)Comparison of spins on 1700pm membranes for the 1%and 3%PMMA in anisole
solutions for 3k and 5k rpm spins. (b)An approximation of spin speeds and membrane size that results
in a membrane with at least 80% useable area for processing.

127
4.1.2. Dual-side Electron-beam lithography

In order to understand and simulate the electron trajectory through the membrane stack, the

Casino Monte Carlo electron-beam simulation software 48 was used. This software is typically

used to look at backscattering of electrons through resist and thick substrate, but it also worked

well to show partial or full transmission through a PMMA/SiNx/PMMA membrane stack. The

full range of membrane (50-500nm) and PMMA thicknesses (50-200nm) were simulated at 10,

20, 30, 40 and 50kV electrons. Figure 4-5 shows simulation results for 70nm PMMA on both

sides of 50, 100, 200 and 300nm silicon nitride. A 50nm spot of electrons at 10, 30 and 50keV

are shown at the various membrane thicknesses. It is quite clear from the simulation that high

membrane thickness and low electron voltage results in substantial spreading of the beam and

backscattered electrons. The SEBL facility at MIT is equipped for 10 to 30 keV on the Raith

150 and at 50keV on the VS 26 from IBM. Figure 4-5 shows that 10keV is insufficient for even

50nm nitride membranes. On the other hand, even at 400nm membrane thickness, 50keV results

in most of the electrons deflecting minimally. This is shown in more quantitatively in Figure 4-

6. By looking at the two-dimensional spot size at exit (a), I measured the diameter at which 90%

of the energy passed. These are plotted on Figure 4-6(b). Not surprisingly, the beam did not

appreciably spread for the 50 and 100nm membranes at 3OkeV and 5OkeV. Further, none of

them showed significant spread of energy at 50keV, which is very promising for use in thicker

membranes or other material sets.

128
 50nm 100nm 200nm 400nm

4014n
- - --- 7.---.7

:I| o

1 1 -7

- I. .

--- - -- - -- -- -- -- -- --- - ..--

Figure 4-5: Results of the Monte Carlo simulation of electrons through a 70nm PMMA resist on the top
and bottom of SiNX at 50, 100, 200 and 400nm thick. The voltage of the electrons is at 10, 30 and 50keV
for the different rows. The widths of the x-axis are the following for the rows starting from the left:
250nm, 300nm 400nm, and 700nm.

129
 a.) b.)
Exit e-beam spot diameter (90% energy)

800
700 -
E
600
100 nm 55040u
500 400 nm

4) 400 200 n
300
200
W 100 1

I 200%j 250%0 0-
0 20 40 60
-250 -125 0 125 250 Electron energy (keV)
nm
Figure 4-6: (a) Example of the energy distribution at the output and (b) The exit spot diameter with 90%
of the electron beam energy contained for various thicknesses and electron energy.

Experimental samples were fabricated first starting from 50nm, 120nm and 400nm LPCVD

silicon nitride on silicon. Membranes were 1700pm width spun with the 1%PMMA in anisole

mixture at 5k rpm. PMMA was spun onto both surfaces. This was followed by spinning the

water-soluble conductor Aquasave. For each membrane, sets of five 5Onm wide spaced by 1Pm

were made at 10 different electron beam doses. This was repeated for 1OkeV and 30keV writing

in the Raith 150. After exposure, the Aquasave was rinsed in DI and the PMMA on both sides of

the membrane was developed in 3:1 MIBK:IPA for 60 seconds followed by 60 second IPA rinse.

The samples were viewed in the SEM. The proper dose was considered to be the set of five lines

that did not have any gaps along any of the lengths. The line widths were measured for each of

the proper dose sets and plotted in Figure 4-7 compared to the simulations. The data matches

well with the simulation at the 30keV write voltage and the thin membrane sizes. However, at

the 10 keV and 400nm thick membrane the simulation showed over 700nm compared to the

130
400nm measured experimentally. The reason for this is likely resist dose clipping and that

somewhere below the 90% energy range used in the simulation resulted in the most consistent

patterning experimentally.

a.)Minim 800
line size for 10 and 30 keV write
700 Simulation
-Experiment
--
S00-

80 nm Z

e00
10 keV
200 - 30 keV
100 -
0

0 100 200 300 400 500

Membrane thickness (nm)

Figure 4-7: (a) Example of the backside patterns written in PMMA and (b)The experimental results
compared to the Monte Carlo simulation for various membrane thicknesses at 10 and 30keV with
70nm of PMMA on both sides.

After showing the dual-side spinning and patterning were possible, I proceeded to exploring

the lift-off process necessary to make nanomagnets on both side. The samples were prepared in

1200, 1700, and 2200 pm membranes with 400nm SiNx thickness. The 3% PMMA solution was

spun onto the membrane for 200 +/- 20nm thickness. Nanomagnets of various sizes were

patterned at 30keV. The PMMA was developed and nanomagnets of 3nm Cr/ 50 Co/ 5 Au were

evaporated onto both sides of the membrane. The samples were carefully placed in a hot NMP

bath at 90oC. After fifteen minutes of lift-off, the first sample was cleaned with the NMP

squeeze bottle while it was still in the NMP solution. This was a common practice I used for

other membrane lift-off to remove metal that is barely holding to the membrane. However, in

this case the membrane shattered before removal could happen. This was repeated for the other
131
membranes- even a small amount of agitation to the solution seemed to cause fracture. After

reviewing the simulation figures, it was evident that the sidewalls of the resist actually tilt

oppositely on the backside of the membrane as on the top. Instead of an undercut (which is not

always necessary for good lift-off) there is a opposite slant, which may result in a continuous

film evaporated onto the backside. In order to reduce this effect, a thinner 50nm membrane was

used. Also, the resist was spun to 150+!- 20nm and 30nm of Ni was evaporated on instead of the

58nm stack previously used. Lift-off was still problematic, now mostly because of the extremely

thin 50nm membrane. It was found that large metal films still attached to the membrane easily

caused fracture upon any agitation from the external environment (e.g. moving the dish around

or spraying NMP into the solution). The problem was solved, in part, by lowering the NMP

temp to 80CC and leaving the samples in solution for four hours. This resulted in 2 out of 4

samples successfully lifted off. The proposed reason for this working is the minor amounts of

currents are not sufficient enough to break the membrane when the metal flap is pulling on it, but

it is sufficient enough to slowly cause fatigue in the metal where it is attached. This results in the

metal being removed before any more drastic handling takes place. At this point the membrane

is significantly less susceptible to fracture. Overall, the dual-side process is promising for both

introducing nanomagnets but also other active features (e.g. nanophotonics or circuits) to each

side.

4.2. Nanomagnet embedding in ultra-compliant nanomembrane

The previous experimental results were all based on rigid silicon nitride membranes. When

compliance was needed, flexures arms and/or torsional hinges were patterned into the membrane.

Another method is to distort, stretch, and align ultra-compliant nanomembranes by forces

132
between nanomagnets. As shown in Figure 4-8(a), an array of magnets embedded within the

membrane is brought into proximity of a matching array of magnets patterned on a rigid

substrate. The attraction between substrate magnets and their corresponding membrane-

embedded magnets can be sufficiently strong to overcome the membranes' stiffness and result in

stretching and bending. Thus, pattern distortions on the membrane may be reduced and

repeatable, accurate overlay achieved. The technique is most promising for large area

lithography masks (Figure 4-8(b)) that would otherwise distort (e.g. x-ray lithography masks)

and for 3D nanomembrane stacking Figure 4-8(c) where layer-to-layer overlay alignment is

critical. Since compliance is preferred, this looks at the problem of membrane distortion from a

completely different angle than typically done. I worked with Martin Deterre to develop the

theoretical framework for predicting maximum membrane distortion from geometry and material

properties, and experimentally demonstrated the fabrication of cobalt nanomagnet disks

embedded in thin PMMA nanomembranes.

133
 a.) Ultra-compliant membrane
Radiation

Mag s Absorber
\Matching
- nanomagnet
arrays
?UI srate
Substrate

Nanopatterne c.)
Magnetic lignment Features

Aligned
layers

Figure 4-8: (a) Schematic of the stretch membrane alignment approach based on nanomagnets
embedded in a highly compliant polymer membrane. The use of this approach for (b) mask alignment
and (c) membrane stacking is shown.

The fabrication process for embedding nanomagnets in the polymer membranes is shown in

Figure 4-9(a). Nanomagnet pillars and disks 200nm thick were patterned on 200nm SiNx that are

1.6mm squares using e-beam lithography and lift-off. Next 200nm of PMMA was spun onto the

membrane and the underlying SiNx was removed with a backside reactive ion etch. Numerous

membrane and magnet sizes were explored; the stress of the deposited magnetic material and the

lift-off process were the critical processing points.

134
 PMMA

Backse

Figure 4-9: (a) Experimental process for creating a suspended PMMA membrane with nanomagnet
embedded within. (b) A fractured membrane (c)200nm nanomagnet pillars embedded into 100nm
PMMA suspended membrane before additional SEM exposure and (d) after additional SEM exposure
showing distortions from new stress within the PM MA.

4.2.1. Design

Numerical and analytical models show a near-linear response for both the membrane

distortion and inter-magnet forces at displacements smaller than the magnet diameters.

Therefore, the effective spring constants of an array of magnets and the membrane can be

compared to calculate the reduction in alignment error that the magnets provide. Figure 4-10

shows the force response for nanomagnet disks overlapping. The nearly linear portion with

small misalignment can be used to model them as a approximate linear spring. Figure 4-11(a)

shows the effective spring constant for various magnet diameters and a z-gap. This assumes

50% duty cycle for the magnets with 20% membrane coverage and that the magnet height equals
135
the diameter. This can be compared to the effective mechanical spring constants for the

membranes shown in Figure 4-11(b). This PMMA and PDMS lines assume a 200pm square

membrane with actuation at the center point. The flexure lines correspond to the four-bar

geometry shown with a 200 micron membrane. The reduction in alignment error is directly

scaled by the ratio of the nanomagnet to membrane spring constant. For example, if the

nanomagnets start 100nm out of alignment and the nanomagnet spring constant is 10x as high,

the expected final alignment would be 1Onm. Optimization is achieved by maximizing the

membrane compliance without making it prone to breakage, and then minimizing the amount

magnets needed for alignment.

Restoring force between magnetic pillars

6

0 Fr

c
4- I r
a)2 -'F
I_ r ccr
M F oc 2
r D -
0
0 200 400 600
r displacement (nm)
t=200nm thick magnets, D=250nm, Z-gap=25nm
Figure 4-10: Numerical simulation results of the alignment force between two
nanomagnets at the 200nm thickness, 250nm diameter and a gap of 25nm. This shows a
nearly linear portion followed by a maximum and a portion that scales with 1/r2, where r is
the misalignment.
136
 The nanomagnet approach to aligning the stretchy membrane is only as good as the

alignment of the magnets to the critical features. The magnets distort the membrane and align

locally, so the critical features only need to be aligned locally to the nanomagnets. In other

words, if 20nm alignment is desired everywhere, the magnets needs to be aligned to 20nm

accuracy locally and the critical features must be matched to them. This can be achieved by

setting a magnet grid that takes into account the patterning errors of the lithography. For

example, in the case of patterning the membrane and substrate with scanning electron beam

lithography, it would be most useful to have the magnets aligned to the critical features within

each SEBL write field. This gives flexibility for write-field stitching errors, which could be

reduced with the membrane stretching.

104 E104 PMMA

E 2Onm z_____
-; PDMS
3 1000 N/ ymer
75nm 102
Flexure 2tm arm
020n
Flexure 1Vm arm
1012 zoonm
z-gap z gp10 Q) Q Q) rm----
- N
50% Duty cycle
U 20% coverage .0
1010
200 E102U0
0 200 400 600 800 1000 100 200 300 400 500
Magnet diameter (nm) membrane thickness in nm

Figure 4-11: (a)Effective spring constant for 50% duty cycle and 20% coverage of magnets on a200pm
square membrane at the diameter and gap size shown. (b)The effective spring constants for 200pm
square membranes with aforce applied at the middle and (c)200pm square SiN. membranes at 1pm
thick and 1or 2pm wide flexure arms with substantially higher compliance.

4.2.2. Process and results

The new fabrication process for embedding nanomagnets in the polymer membranes is

shown in Figure 4-9(a). Nanomagnet pillars and disks 200nm thick were patterned on 200nm
137
SiNx that were 1.6mm squares using e-beam lithography and lift-off. Next 100nm to 200nm of

PMMA was spun onto the membrane and the underlying SiNx was removed with a backside

reactive ion etch. Quite amazingly, this resulted in a completely freestanding polymer

membrane with an aspect ratio of nearly 10,000:1. At the macros-scale this would be analogous

to a 1cm acrylic plate suspended over a football field. Numerous membrane and magnet sizes

were explored; the stress of the deposited magnetic material and the lift-off process were the

critical processing points. In order to better control the backside etch, thin 50nm membranes

were used for the sacrificial silicon nitride. However, this resulted in some of the same

problems as presented Section 4.1.2. In particular, the lift-off was done in 90*C NMP for 2 hours

versus the typical 10 to 15 minutes for regular lift-off. No agitation was used to remove

additional metal until the large visible flaps of metal were not attached. After lift-off, the

PMMA transfer was surprisingly robust. However, fracture did occur from over-etch or

membrane heating during the final silicon nitride release. The fractures appeared to instigate at

the magnet-PMMA interface. Figure 4-9(b) shows an example of membrane fracture that was

caused by poor release parameters. Figure 4-9(c) shows 200nm diameter nanomagnet pillars

embedded in the suspended 100nm thick PMMA membrane with 0.5mm dimension.

Figure 4-9(d) shows the same membrane as c after multiple views under the SEM49 . The

exposure by the SEM distorts the membrane by locally stressing the material. After significant

distortion within the SEM, there was no evidence of fracture near or away from the magnet

array. By comparing the membrane compliance to the effective spring compliance of the

10 x 10 array of 200nm diameter/height nanomagnets, it was shown that the expected reduction

in alignment error is 70%. However, the initial coarse alignment must be less than the diameter

of the nanomagnets. It should be noted that the strategy of feature transfer into a polymer

138
membranes is not limited to nanomagnets or even metal features. As long as the material is

compatible with silicon nitride and a PMMA spin, this process could be used with virtually any

common microtechnology material. Furthermore, voids can be patterned into the PMMA by

making sacrificial pillars on the silicon nitride that are removed before the backside etch.

Conversely, the PMMA can be patterned within e-beam while it is still attached to the silicon

nitride. Overall, this patterning and transfer approach appears to be a good platform for

membrane folding and stacking, particularly when large area alignment is required.

139
Chapter 5

Improvements to nanomagnet alignment

The nanomagnet alignment technique showed layer-to-layer alignment that met or exceeded

the requirements needed for many 3D nano-assemblies, including nanophotonic crystals.

However, there are straightforward improvements that will likely push the alignment to better

than 1Onm over multiple millimeter areas. The primary constraints to the alignment precision

were the mechanical interference of the physical magnets and the lithographic misalignment

between the nanomagnets and the features meant to be aligned. Solutions for both of these

issues are presented in section 5.1 with proposed fabrication processes based on concepts learned

through this research. Up until this point, the folding has been based on two segments folding

into face-to-face alignment. Section 5.2 presents methods to fabricate multi-layer structures

based on integrating the nanomagnet alignment approach with membrane folding methods

developed for the NanoStructured Origami 3D nanofabrication method. The other option for 3D

layering is shown in Section 5.3, which I have called the "Stack-N-Snap" approach, based on

aligning membrane segments that are suspended on flexures that fracture after stacking. The

final problem of the nanomagnet alignment method is that a metallic and magnetic material is

introduced into the 3D matrix. This can be particularly problematic for nanophotonic crystals,

where an absorbing material will reduce the benefits of the crystal. Solutions for this are

presented in Section 5.4 using dielectric magnets or "electrets". Overall, these improvements

140
combined with the previously presented models and alignment results make the nanomagnet

approach a serious improvement to many 3D membrane processes.

5.1. Improvements to overall alignment

The results shown in Section 3 of better than 30nm alignment required significant focus

on the lithography and stresses within the membrane and magnetic materials. For the process

and materials used, this was likely approaching the best-case scenario since the lithography and

stresses introduce moderate errors. Section 5.1.1. presents small improvements using the

previous approach of materials and processes, as well as a likely significant improvement based

on addition of a few processing steps. The previously mention mechanical contact between

magnets on matching arrays limited the amount of error averaging and processing tolerances. A

process for significantly improving the processing tolerances of the magnets and allowing for

elastic averaging is shown in Section 5.1.2. Overall, the author believes that alignment can be

pushed to sub-I Onm over millimeters-wide membranes with the introduction of these

improvements.

5.1.1. Process control

Substantial steps were taken to minimize the effect of stresses within the membrane and

magnetic material. However, there are still improvements to processes and materials that could

be made. First, any reduction in the residual stress of the membrane would improve the

alignment. This did not play a substantial roll in the smaller membrane sizes used in the

experiments, but over millimeter areas would likely show up in multiple tens of nanometers in

distortion. The first solution would be to further reduce the residual stresses within the

141
membrane. This comes down to process control22 and could be achieved post-production by

testing various lots of LPCVD silicon nitride for minimal distortion after release. Furthermore,

oppositely stressed material can be added to counter the membrane's intrinsic stress. Finally,

compliant membranes that allow the magnets to distort the membrane and locally align would

further reduce the effect of membrane stresses. The nanomagnets added to the membrane give

additional stresses that are dependent on the magnet material properties, pattern and thickness.

Although many metallic materials can be placed in tensile or compressive stresses depending on

deposition and annealing parameters, it is unlikely to fully eliminate the stresses. There are three

primary options available to reduce the effect of the magnet stress. The first is to balance the

stress by depositing a counter-stressed material onto the magnet. However, this adds to the

thickness of the magnet and is a difficult balance.2 3 The second solution is to pattern identical

magnets on both sides of the membrane as presented in Section 4.1. This balances the stress

moment across the membrane cross-section but still results in a net stress in-plane. The third

approach of embedding the magnets at the center of the membrane also eliminates the stress

moment across membrane but results in an in-plane stress.

If alignment is achieved by another method besides folding via torque on the nanomagnets,

the percent coverage and thickness of magnets can likely be significantly reduced. This is

because folding was the limiting factor in the process when comparing the mechanical and

magnetic forces. If the magnets arrays are brought into coarse alignment via the folding or

stacking approaches shown in Section 5.2 and 5.3, the magnet coverage can be cut to only a

fraction depending on the membrane compliance. This also is advantageous to increase the

usable area on the membrane for non-magnet features (e.g. nanophotonics). The other

disadvantage of using the nanomagnets as folding features is the external field needed to be at an

142
angle relative to the plane of the nanomagnets in order to provide a torque. This unfortunately

provides an added bias for the nanomagnets to partially magnetize out-of-plane and pull in the

arrays at a non-ideal position. This problem is eliminated with the approaches of 5.2 and 5.3 by

using a high strength >0.5T B-field that is magnetized strictly in-plane. One improvement that

would require significant experimentation would be to vibrate the specimen to assist in reaching

a local energy minimum. An additional coating might be necessary to reduce friction or van der

Waals attraction to allow for appreciable movement. Further, the vibration could easily cause

fracture in the fragile membranes, so the vibration approach is not promising.

The alignment of the nanomagnets to the nanopatterned features is a critical aspect of this

alignment approach. Writing the magnets and critical features in separate lithography steps

introduces alignment errors within respective membranes. This may be only a few nanometers

with introduction of features as used in SPLEBL, but typical mask alignment and e-beam

systems put this in tens of nanometers over larger areas. In order to eliminate this problem, the

processes shown in Figure 5-1 are suggested. The process shown in (a) results in nanopatterned

features being voids in the membrane, while (b) is additive features such as evaporated metal

circuits. Only the membrane is shown and not the connected wafer. Process (a) starts identically

to the experimental process used in Section 3.1. A thin Cr layer is coated onto the SiNx

membrane. PMMA is spun onto the membrane and patterned with both the magnets and

features(ii), followed by an etch of the underlying Cr (iii). The process diverges from the

previous work at step (iv), with the spinning and patterning of an additional resist. The resist

must have process and development selectivity with the PMMA so that the patterned PMMA

does not deteriorate with its addition. An optical lithography resist would be preferred but

PMGI has been shown to have good selectivity5 0 . The resist does not need to be accurately

143
patterned, but rather block voids in the PMMA for the critical features while leaving the magnet

patterned areas open. The magnet material, for example 3nm Cr/50nm Co, should be evaporated

onto the sample (v) with a Cr capping layer similar to the experiments of Section 3. This process

is finished exactly the same as the process used in the experiments. The resist is lifted off and

the sample is placed within a CF 4 reactive ion etch to etch through the membrane and eventually

the Cr layer. The process shown in (b) is based on the same strategy of writing the magnets and

critical features in the same step and then selectively protecting these areas while adding

materials via electron beam evaporation. These processes can be combined to make both

additive and subtractive patterns over the membrane aligned with the magnets. For a typical e-

beam lithography system, if the magnets and features are patterned within the same write-field

one would expect a placement error of less than 1Onm. Furthermore, the misalignment would

not be global (e.g. misaligned everywhere) but rather could benefit from elastic averaging.

144
 (a) (b)
Cr PMMA
11 ~~~SiN,
SiNX

PMMA -,r r - PR/PMGI

PMG
W
DDD LMG Metal

Magnetic
-] E Metal

r7_1 r-11i

DLI
Figure 5-1: Fabrication processes starting from suspended SiNx membranes for creating nanomagnets
and features in the same lithography step to reduce alignment error. The process in (a)results in
magnets and patterned voids in the membranes. The process in (b)results in nanomagnet and additive
features.

5.1.2. Magnet pattern

The results of the experiment indicated that the bar shaped magnet is favorable over the

rounded edges. The design options of the magnet shape and magnet array were presented in

Section 2.3. Angling the magnet faces at the poles allowed for magnets that were longer than

designed to still mechanically fit together at the cost of misalignment perpendicular to the length.

Further, the magnets have a slight amount of sidewall chamfer due to the e-beam evaporation of

the nanomagnet material. This is advantageous for aligning the magnets in-plane but dose errors

(e.g. too long of magnets) are translated into errors in the gap between membranes. This was
145
likely seen in the second experiment of Section 3 since the membranes could be brought in and

out of contact with an external torque. Overall, the location of the magnets on the surface is

problematic for multiple reasons: magnets must perfectly fit together mechanically, the magnets

create a gap between membranes, the magnets make distortion on the membrane, and the

magnets introduce friction for translations perpendicular to their length. For all of these reasons,

it is proposed that burying the nanomagnets within the membrane is the primary improvement

for future work. The greatest advantage is the dose of the magnets does not need to be perfectly

controlled because elastic averaging will be seen between the competing poles (Section 3.3).

The membranes are allowed to be brought into flush alignment and the magnets place minimal

stress on the membrane. A method to place the nanomagnet within compliant PMMA

nanomembranes was shown experimentally in Section 4.2. The process is more complicated for

silicon nitride membranes. Two approaches are shown in 5-2.

146
 ______________________SiNk ___________________ Cr

tw Si Q
SiN,
Cr
____ _______ ___
____ _______ __ ____ ___ ___PMMA

_________ED_________ PMMA
@owo
I @Mir0M

Electroplate Ni
Co

(a) 1W (b)
Figure 5-2: Processes for embedding nanomagnets into suspended SiN membrane. (a) PMMA is
patterned on SiNx membrane followed by an RIE etch and metal evaporation. After lift-off nanomagnet
are left embedded in the membrane. (b) Vias are etched through the membrane and after a series of
steps, Ni nanomagnets are electroplated into the membrane.

The first process (a) is straightforward and may replace the nanomagnet patterning steps used

in the experiment. Thick PMMA is patterned on to the silicon nitride membrane (i-iii). This is

followed by a CF 4 reactive ion etch using the PMMA as a mask. It is likely that only a thin

(<1 00nm) etch may be possible without significantly rounding the PMMA sidewalls making the

lift-off process problematic. The etch selectivity may be improved by selectively electroplating

the top PMMA layer with Ni after evaporating a thin 5nm Au layer. This is followed by plasma

147
etch to first etch through the Au and then a deeper SiNX etch. Either of these approaches would

then be followed by magnet evaporation and lift-off. The process shown in (b) is for thicker

magnets within the membrane. Chromium (1 Onm) is evaporated onto the top and backside of

the membrane. PMMA is spun onto the top and patterned followed by a backside spin. The top

PMMA pattern is transferred into the top Cr via wet etch. This is followed by a CF 4 reactive ion

etch through the membrane to the underlying Cr. This is followed by a selective electroplate

using the backside Cr as the seed layer. Magnetic Ni, for example, is electroplated to fill the

voids in the membrane. This is followed by removal of the PMMA and selective wet etching of

the thin Cr layers.

5.2. Integration with stressed folding

All of the design and experimental work on magnetic alignment up onto this point hav been

based on two segment origamis with matching arrays of nanomagnets. In order to reach the

primary goal of 3D nanofabrication via aligned membrane manipulation, additional steps must

be introduced. The first approach is to add additional membrane segments to the end of the

folding membrane segment. Theoretically, rotating the external field back and forth could result

in an "accordion" style fold. However, this requires that (1) the membranes segment stay

bonded after being brought into contact and (2) the magnets are embedded within the membranes

or on both sides. The first requirement is non-trivial and likely requires an additional bonding

material unless the van der waals forces can hold the faces together after contact without folding

apart upon subsequent field rotations. The second requirement was achieved experimentally in

the research of Section 4. This is a promising approach, but the moderately fragile hinges

required for nanomagnet folding would likely introduce yield problems when multiple are

148
connected in series. For these reasons this approach is not suggested for more than a few

origami segments.

A method to increase yield is to combine the nanomagnet alignment approach with other

more robust origami folding strategies. As discussed in Section 1.4.2, the primary stress-based

folding methods used in the NanoStructured Origami approach are using a stressed metal layer

on the membrane or to inject the membrane with ions using a focused ion beam. The stressed

bilayer approach is not a good match for the nanomagnet alignment strategy. There are currently

three problems with integrating these methods. First, the radius of curvature of the fold is in 1Os

of microns, which results in the membrane faces being out of the range of the nanomagnets.

Secondly, the folding only occurs in one direction, limiting its application for the "accordion"

style fold. Finally, the release steps for the membrane are based on a wet KOH etch. This is a

particularly violent etch since the underlying silicon etching creates gas bubbles and therefore

high surface stresses. Since flexures need to be built into the membrane for the nanomagnet

alignment, these would not hold up to the KOH etch. The ion-implantation strategy is a

significantly better approach to folding the membrane for nanomagnet alignment. First, the

radius curvature has been shown to be less than 2 microns 26 , which brings the membranes well

within the coarse alignment distance shown experimentally. Secondly, the membranes can be

folded both directions. This allows for the "accordion" shape fold to be made with the ion

implantation. Finally, the folding process is completely dry within the FIB. In fact, the exact

same process used in the experiment of Section 3 can be used to prepare the sample before ion-

beam folding. The main difference would be to replace the fragile ultra-compliant folding hinge

with a simple flexure to allow moderate deflection during alignment. This approach can be used

for origamis of multiple segments, particularly since supports can be added to hold long

149
sequences of membrane segments. The supports are cut with the FIB while in the chamber just

prior to folding. Although this approach is promising for prototyping, the requirement for a

serial-process of FIB, and the yield problems of multiple segments would likely limit this to

laboratory development. In order to utilize the nanomagnet alignment for volume production,

the stacking approach of the next section is proposed.

5.3. Stack-n-snap 3D nanofabrication approach

As discussed in Section 1.3.3, membrane stacking holds advantages over folding for multi-

layer structures with many layers. Each layer added via stacking can be inspected and unlike

folding, there is no requirement for identical materials for each stack layer. All of the theory,

models and results of this research readily apply to membrane stacking. Therefore, nanomagnet

alignment of membrane segments is very promising. The "Stack-N-Snap" approach to

membrane stacking utilizes the magnets' ability to pull-in, align and bond layers together. In the

simplest approach shown in Figure 5-3(a), nanomagnets and features are patterned onto

membrane segments that are held by some form of frame (e.g. membrane attached to a

substrate). The membranes are connected to the frame via one or more flexures that supply the

necessary compliance for nanomagnet alignment. Between the flexure and the membrane there

is a release mechanism, most likely a weakened fracture point, but could also be a more elaborate

sacrificial material or electrostatic holding. An external magnetic field is applied to keep the

nanomagnet in magnetic saturation during the entire multi-step assembly. The membrane

segments are brought into close proximity with a membrane stack with matching nanomagnet

arrays. The nanomagnets attract, align and bond the membrane to the stack. The stage and

frame are moved in such a manner to fracture the release point without affecting the alignment of

150
the membranes. This process is repeated for multiple segments and can include multiple

different sets of materials.

(a) Fractured Flexure

Flexure as
Membrane
Frame Segment Stack

(b) (c) (d)

ix
Lx
Fracture point

rFracture
dt 1 point------

Y Lz,max

F F,

Figure 5-3: (a)Schematic of the Stack-N-Snap process. Membrane segments with patterned
nanomagnets are held to a frame by compliant flexures. When brought into proximity with a
membrane stack, the nanomagnets pull-in, align and bond the membrane. Upon pulling the frame
away, the flexure fractures to leave the magnet. (b-d) The dimensions used for the simplified flexure
and fracture model.

There are clear advantages to the nanomagnet approach when compared to the state of the art

developed by Aoki et al. Their alignment takes place within a modified SEM chamber with a set

of electrostatic micro-manipulation probes for sample handling. Besides the problems of

fabricating accurate alignment pillars as discussed in Section 1.5.1, the micro-manipulator

approach requires slow sequential handling of each membrane segment. Instead of the magnets
151
providing pull-in and alignment, the probe must supply a force to push the membrane plate into

alignment. This does not result in fracture for the thicker plates that are used, but would be

problematic for fragile membranes. The coarse alignment required for the Aoki method is

considerably tighter since the nanomagnets provide a pull-in range. This results in higher

precision equipment needed and again slower sequential layering compared to the Stack-N-Snap

approach.

In order to better understand the fracture release mechanism in the Stack-N-Snap approach a

simplified mechanical design was modeled as in Figure 5-3(b-d). Assume that the membrane

segment is held by a flexure with arms of lengths L, and Ly connected by a radius of curvature

large enough so that fracture does not take place at the elbow. The membrane is brought into

contact with the stack and the frame is displaced upward in the z-direction. Now assume that the

flexure arms are flexible to the point that the flexure approximately stretches out to a straight

cord with length Lt ~ L, + LY . From geometry Lt 2 ~ Lz,max 2 + Lx 2 + Ly so Lz,max

2LXLx. Experimentally, we know there is a maximum hold down pressure that the

nanomagnets and/or van der Waals force provide. The vertical force required to peel two

membranes in contact is the maximum force that the flexure will encounter in the z-direction

during frame removal. In the folding and unfolding experiment of Section 3, the unfolding

torque necessary to pull apart the 1OOm membranes can be roughly translated into a 250+/-75

nN pealing force. Setting this to Fmax you find F = Fmax; F = xFma

2L~ maT and Fy = Fmax-
Lx m

Figure 5-3(d) shows a close-up of the connecting bar where the fracture takes place. A notch for

stress concentration creates a maximum stress in the location shown. The bending moments at

that point are given in Equation 5.1 with the maximum stress in Equation 5.2. This assumes a

152
stress concentration factor c=2, which may vary by the relative size of the notch compared to the

bar.

Mz = Fd ; M, = Fd ; My 0 A = wt (5.1)

Umax = C +Mz + Mx = 2 + 6F + 6 (5.2)

Using the same parameter used in the experimental process with SiNx of one membrane

thickness, the stress at the point of fracture is plotted in Figure 5-4(a) for various ratios of the

arm lengths. This also assumes a bar length of d=1Opm. The fracture stress of SiNx (af ~

400MPa) is designated on the plot. It can be seen in this simplified model that the magnets and

van der Waals stress provide the necessary bonding force to translate into a fracture at the point

of frame removal when the width of the fracture location is less than 0.2pm. In other words, the

membrane will not pull back up with the frame and stage when moving on to another membrane

segment. The fracture model was confirned via FEA in Cosmoworks 5 1 software. There is

substantial design freedom to further control the fracture, but the results from the model are

promising. Figure 5-4(b) shows an experimental example of 200nm tall nanomagnets patterned

on a 1 00pm membrane of 200nm thickness held by four flexure arms, that would be used in the

Stack-N-Snap assembly process.

153
 (a) Maximum stress dependency on LxILy
1.E+10 T

w=0.05
pm

0.1 PMrn n
1.E+09
E
E .2pm

1E+08 ----- r-- - -.

0 0.5 1 1.5 2 2.5 3 20p
Lx/Ly
d=10 pm, t=1 pm, w=0.05, 0.1, 0.2 .Lm
Figure 5-4: (a) Results of the model showing the stress at the fracture point depending on the ratio of
flexure arms and width of the fracture point. (b) 1pm thick SiN, membranes surrounded by four flexure
arms of approximately 1pm width with four sets of 200nm tall nanomagnet disks.

5.4. Dielectric magnets or electrets

The final problem that remains is the metal magnets act as an absorber or a conductor in

assemblies where this is not desired. Further, the leaking magnetic field can affect electrical,

optical or quantum 3D nanomanufactured devices. The effect of this can be reduced by locating

the magnets away from the critical regions, but this reduces the amount of alignment force and

places the magnets away from the locations where alignment is the most important. What would

be most desirable is that the magnets provide alignment at the point of assembly and then are

removed. Unfortunately, cleanly etching out the magnets from the 3D matrix after assembly

would likely be prohibitively difficult for most multi-layered structures. The closest thing to

removing the magnet is to demagnetize the nanomagnets by bringing the structure above the

curie temperature. However, the Curie temperature is typically very high (TcFe = 7704C and

Tc,co = 1130*C) and would likely result in material damage or warping in a 3D matrix. Further,

154
the magnets would return to some magnetized state upon cooling. In the last 15 years there has

been development on polymer magnets. Note: The term "Polymer magnet" is also used to refer

to magnetic metallic powder dispersed into a polymer matrix. In our case, we are interested in

pure polymer materials. Rajca52 developed a polymer that acts magnetic under 10*K at which

point it saturates to roughly 1/ 2 0 h of iron. Above 10*K, the material is non-magnetic.

Unfortunately, magnets of this material would provide 1/400* the force of equivalent iron

nanomagnets and the requirement from such cold processing introduces more material problems

and distortions than it benefits. No significant breakthrough in polymer magnets has been seen

that overcomes both of these problems.

Another option is to use an array of "electrets" in place of the magnets. Electrets are quasi-

permanent groupings of electrostratic charge. Electrets can be dipoles similarly to magnets, but

they also can be positive or negative monopoles. Fortunately, electrets take place in dielectric

materials and are more transient in nature than magnets, since the charge typically migrates over

time to reduce the electret's strength. In other words, they can be temporary during the

alignment steps and removed after. Figure 5-5(a) shows the two formats in which electrets could

provide alignment for stacked membranes. In the first case, an array of positive and negative

electrets is patterned onto the membranes in an alternating fashion. In this case, the electrets

theoretically provide a pull-in force similarly to the nanomagnet approach. If the electrets

patterning is limited to only positive or negative electrets the other approach would be used. In

this case, the electrets only provide a repulsive force so an additional bonding force would be

necessary. The important improvement over the nanomagnet approach would be that the charges

could feasibly be removed after alignment to leave strictly a dielectric material.

155
 340, rv-m
a I I I-iI N I

I I- I I- I I- I I

I I-I I-L I -i i

-/

b NEG (-) POS (+)

1 E-06 - - - - - -

1 E-07
150 nm SiN
z 1 E-08

©IthDeff
0
LL 1 E-09 iL
5
1 E-10 - ----
-
100nm PMMA
1.E-11 -

1.E-12 Dlitho
0 1 2 3 4
Gapdinstance(pim)
Figure 5-5: (a)Schematics of the two versions of electrets alignment based on alternating positive or
negative charge or otherwise all positive or negative. (b) Theoretical force between lpm diameter
electrets based on literature results of surface charges for 150nm SiNX and 80nm PMMA. (c)Example of
lpm period electret array in PMMA by Whitesides. (d)(i) Schematic for negative and positive electret
charging in a suspended membrane using e-beam exposure at low and high voltage.(ii) Example of
mechanical islands that can be used to ensure electrets are constrained to prescribed sizes and locations
in the suspended membrane.

In order to understand the potential alignment forces that the electrets could provide, a

number of electret materials in the literature were explored. The most common materials for

bulk electrets are fluoropolymers, which have high resistivity and retain charge. The charging of

thick films, for example by corona discharge, is well understood and used in products ranging

from microphones to air filters. However, limited work has been done at the nano-scale for the

156
properties of the electrets. Some work has been done on charging submicron thin films for use in

data storage5 3 Electrets have been patterned for MEMS power generators at thicknesses in

multiple microns5 4 . This work gave experimental insight that the properties of bulk electrets

cannot be simply scaled to the micro or nano-electrets. Charging of the films is encountered by

applying a current through the resistive material. Charge builds up in the volume or at the

surface resulting in an effective surface charge density d, and surface voltage V. Assuming

patterned electrets of a certain A, the electrostatic pole is estimated as Q, = 6-,A. Using the

electrostatic dipole approximation the force between two electrets separated by a distance d and

diameter D such that d>>D is given by:

2
QaQ2_ ~ 2

Fe= 2
(5.3)
47AE~d2 47rE0d2

Comparing Equation 5.3 with Equation 2.5, it can be shown that the equivalent

magnetization of a surface charge density to provide the same force from a pole is given in

Equation 5.4 where Meq is in tesla and &,-is in C/m 2 . Unlike the magnetization of the material,

the surface charge density is dependent on the sample thickness. A thicker sample typically

provides more charge than a thinner sample. This does not appear to be a linear effect when

comparing bulk vs. thin film because of various charge migration mechanisms.

Meq = =376 - [m] (5.4)

Two examples of thin film dielectrics were shown in 150nm Si 3N4 5 and 80nm PMMA".

Most interestingly, Whitesides et. al showed that submicron electrets could be patterned into the

thin PMMA using a flexible PDMS electrode with multiple points of contact. Figure5-5(c)

157
shows the resultant pattern of 350nm wide electrets measured with a conductive AFM. The

electrets show a 2V surface voltage. The authors use this surface voltage to estimate a surface

charge density of 0.0016 C/m 2 . Using equation 5.4, this corresponds to a magnetization of 0.6T.

Although this is less than the 2.2T and 1.8T for Fe and Co, this is nearly equivalent to that of Ni.

This is extremely promising that the electrets can provide a pull-in force for alignment. Zhang et

a155 showed that thin film silicon nitride can act as an electret. The entire thin film was charged

showing an initial 50 C/m 2 surface charge density, which quickly fell to nearly zero within ten

minutes. With increased boron ion within the Si 3N4 , part of the charge could be retained after 15

minutes. These results are for the entire film and nanopatterning as shown by Lo54 would result

in significantly lower values. Figure 5-5(b) gives the force calculation for the 80nm PMMA

electrets and 150nm fully charged Si 3N4 electret assuming a 200nm diameter electret. The

nitride forces give a high end value, much greater than seen in the nanomagnets. On the other

hand the PMMA electrets, which have been realized experimentally in submicron form, show

performance similar to nanomagnet (see Figures 2-4 and 2-5).

158
 100nm 200nm

2keV

5keV

Figure 5-6: Monte Carlo electron beam simulation results showing the mechanisms
for positive and negative charging 100nm and 200nm suspended PMMA membranes.
In the 2keV simulations, the electrons give a net negative charge since they do not
pass through the membrane. At the 5keV and 100nm thick membrane, one would
expect positive charging since the magnets pass through and kick off secondary
electrons. The 5keV and 200nm isunclear and would result in a wider electrets
because of the beam spreading.
The use of electrets may hold some obvious benefits but there are still a number of questions

regarding what happens at the point of fine alignment. Using nanomagnets and an external

magnetic field, the internal magnetization of the nanomagnets is well controlled and magnets

within close proximity do not affect their neighbor's internal state. On the other hand, bringing

electrets into close proximity could easily result in charge migration, dielectric breakdown, and

non-ideal alignment. It is unclear if arcing may occur at the point of contact, which, if

controlled, could be beneficial to create a bond, but could easily cause heating problems. It is

also unlikely that an external field could be used to fold the electrets since they are not

constrained to being dipoles as magnets are. One benefit of the electrets approach is the

membrane process is well suited for charging since there is not underlying silicon which can

remove charge. In fact, membrane charging has been shown in the lab on free-standing PMMA

membranes by charging with the electron dose from an SEM. I propose that arrays of negative
159
electrets can be patterned with low voltage bias voltage. Figure 5-6 shows examples of 100nm

and 200nm PMMA membranes patterned with 2keV and 5keV electrons. Positive electrets may

be possible by using high voltage electrons. High voltage electrons completely pass through the

membrane and cause secondary electrons to leave the membrane. This results in a net positive

charge, although it may take a high dose to achieve the necessary electret performance. One

method to reduce the effect of charge migration, give higher alignment tolerance and better

control the shape of the electrets is shown in Figure 5-5(d.ii). In this case, discs held by thin bars

are patterned to be suspended in the membrane. The discs (and possibly the thin arms) are

charged within the SEM. This creates controlled mechanical dimensions of the electrets and the

charge will have more difficulty migrating in the thin bars. The bars can also flex to provide

elastic averaging if necessary. The electret alignment approach is very promising for use in

multi-layer dielectric stacks such as needed for 3D photonic crystals. Although most of the same

principles of the nanomagnet alignment apply, a greater understanding of the charge migration

and process flexibility will need to be obtained through future research.

160
Chapter 6

Conclusion

The use of nanomagnets as alignment features for membrane stacking and folding has

been presented. There is clear evidence both theoretically and experimentally that the approach

is useful for self-aligning membrane segments to high accuracy for use in 3D nanofabrication. In

the pursuit of proving the utility of this approach, a number of additional processes and concepts

were developed ranging from new membrane handling to a new style of MEMS hinge. A greater

understanding of the forces and interactions at this scale was found through modeling the various

forces. The nanomagnet interactions with an external field and at coarse and fine proximity to

another magnet were modeled to optimize array design. This was also used to assist in

mechanical designs of novel hinges and flexures that provide the required system compliance.

New membrane processes included a dry release process, a double-sided patterning approach and

creation of a thin polymer membrane with embedded magnetic features. The research provides

strong groundwork, which with the additional improvements suggested in Section 5, will provide

a breakthrough in 3D nanofabrication. Future areas of application include 3D nanophotonics, 3D

biotech devices and other 3D hybrid systems.

The modeling of the nanomagnet interactions with an external field and nearby magnets

was modeled at various levels of detail. The magnetization states were modeled using Object

Orientated MicroMagnetic Framework to ensure single domain nanomagnets with magnetization

in a controlled direction. The magnet shape, material and proximity to nearby magnets all
161
played important rolls. By placing the nanomagnets within a strong external magnetic field, the

magnetization could be further constrained along their long axis, even with coupling effects of

nearby nanomagnets. Using this approach to create magnetic dipoles, the force interactions

between nanomagnets could be easily modeled to understand the pull-in and alignment forces. It

was shown that the pull-in force scales favorably and can overcome mechanical forces. Further,

various arrangements to assure that the inter-magnet alignment force has a favorable dependency

on misalignment distance were shown and modeled. The net interaction between arrays at

coarse alignment for these designs showed favorable elastic averaging effects. It was discovered

that the same external field that magnetizes the nanomagnets could be used to provide a torque

for folding actuation, which greatly simplified some of the later experimental processes. After

fully understanding the nanomagnet interactions, the mechanical forces involved in membrane

distortion, folding and alignment were modeled. This led to the design and modeling of a novel

NEMS-style torsional hinge that provides sufficient folding and translational compliance, while

allowing face-to-face contact of folded segments.

After understanding the magnetic interactions and mechanical forces, a series of

experiments to prove the nanomagnet alignment approach was performed. Actuation of simple

cantilevers with patterned nanomagnets was used to certify that the elongated magnet geometry

provided sufficient anisotropy to apply a torque with an external field. Previous nanopatterned

silicon nitride fabrication processes were modified to have a dry final release step. This enabled

the fabrication of the sensitive torsional hinge structures with incredible structural compliance.

Next, the nanomagnet alignment approach was proven, showing clear evidence of pull-in when

the arrays are brought to close proximity. This alignment was further improved by taking

particular care in the lithography and controlling the nanomagnets' internal state. This resulted

162
in alignment between patterned optical components on the membranes to better than 30nm over

100 micron segments for a set of samples. The residual alignment error is well understood, and

steps for further improving the alignment were presented.

In order to achieve 3D stacking of membranes aligned with nanomagnets, two new

approaches for membrane patterning were developed and shown experimentally. The first was

for patterning nanomagnets on both the top and backside of the suspended membrane. The

approach was based on spinning resist on the top and backside of the membrane, followed by

high voltage e-beam writing, material deposition and lift-off. The angled sidewalls of the

anisotropic KOH etch of the silicon wafer enabled relief for the spinning resist. The top and

backside of the membranes were exposed at the same time with good fidelity between the layers.

This is a promising approach for not only introducing nanomagnets but also other optical or

functional features to the membranes. Instead of patterning onto both sides of the membranes, a

second approach was developed based on implanting the nanomagnets within compliant polymer

membranes. It was shown experimentally that cobalt nanomagnet disks could be transferred into

suspended PMMA films less than 150nm thick. This enables an approach described as

"stretchy" membrane alignment, where the nanomagnets align on both the global and local level,

giving greater fabrication tolerances.

A number of further improvements to the nanomagnet alignment approach have been

proposed to enable membrane stacking over large areas with alignment potentially better than a

few nanometers. The first critical improvement is to ensure that the alignment force scales

similarly to a spring. The easiest way to achieve this is implanting elongated nanomagnets

within the membranes. This allows greater fabrication tolerances by reducing the mechanical

interference of the magnets while also giving the benefit of elastic averaging. The second

163
important improvement is to eliminate the alignment error between the nanomagnets and critical

features. Multiple processes for defining these in the same lithographic step have been proposed.

An approach called "Stack-n-snap" has been proposed based on the nanomagnets pulling-in and

aligning membrane segments which are held to a frame by flexures. After alignment, the frame

is pulled away, resulting in controlled fracture as the nanomagnets hold the segments together.

Finally, the use of electrets has been proposed using many of the same advantages of the

nanomagnet approach with potentially greater material choices. In particular, temporary

alignment features based on implanted charges enables alignment without the use of absorbing

magnetic metals that can be problematic for some optical devices.

The ability to quickly and accurately align and stack nanopatterned membranes is highly

impactful to 3D nanofabrication. Today, most 3D nanofabrication is limited to simple material

sets and slow prototyping. Contrary to other approaches, membrane stacking does not sacrifice

throughput or feature resolution for material choices or pattern complexity. However, membrane

stacking is difficult and has been limited to small overall sizes or large layer-to-layer

misalignments. The addition of self-alignment features to membrane stacking will improve

throughput, alignment and/or overall device size. This will likely first be seen in prototyping

novel 3D photonic or electronic systems and potentially later as a tool for industrial scale

manufacturing. The ability for complex material sets opens the door for various hybrid systems

combining one or more photonic, electronic, quantum, biological, magnetic, mechanical, or

chemical components. Since self-alignment between membranes has been proven, the biggest

hurdles look to be processing, handling and cleaning the fragile membranes. Steps were taking

in this research, but dedicated equipment must be developed based on the unique properties of

164
the membranes. An industry push, perhaps driven by an application such as 3D photonic

crystals, would likely overcome these limitations.

If the membrane layering tools are developed to their full extent, nanotechnology will

certainly grow into the 3 rd dimension with faster systems, new functionality, greater information

densities and hybrid systems. One path of development would be from perhaps 3D photonic

crystals, to electro-photonic hybrid systems into optical or quantum computing decades down the

road. A second path of development may be simple cell matrix scaffolding into full-scale tissue

engineering using nanopatterned structural precursors. The ultimate goal is that membranes can

be treated simply as building blocks at the lab, prototype, or production scale for building the

next generation of nano-devices. This present thesis helped the goal along by providing a

reliable mechanism for membrane alignment and bonding.

165
Bibliography

[1] Website: http://www2.warwick.ac.uk/fac/sci/eng/research/microtec/3dmicrostructures.

[2] M. Deubel, M. Wegener, A. Kaso, J. Sajeev, "Direct laser writing and characterization of
'Slanted Pore' Photonic Crystals," Applied Physics Letters, vol. 85, pp.1895-1897, SEP
2004.

[3] Y.-T. Chen and M.T. Lee, "OPC for edge post structure using chrome-less phase shifting
mask in 3-D memory," ProceedingsofSPIE, vol. 5992, pp. 1-11, 2005.

[4] Website: http://naturalnano.com.

[5] H.J. In, S. Kumar, Y. Shao-Horn, G. Barbastathis, "Nanostructured OrigamiTM 3D

fabrication and assembly of electrochemical energy storage devices," Nanotechnology,
2005. 5th IEEE Conference on Nanotechnology, vol.1, pp. 374- 377, JULY 2005.

[6] S. Cao, K. L. Yeung, J. K.C. Kwan, P. M.T. To, S. C.T. Yu, "An investigation of the
performance of catalytic aerogel filters," Applied Catalysis B: Environmental,vol. 86, pp.
127-136, FEB 23, 2006.

[7] A. Mata, E. J. Kim, C. A. Boehm, A. J. Fleischman, G. F. Muschler, S. Roy, "A three-

dimensional scaffold with precise micro-architecture and surface micro-textures,"
Biomaterials,vol. 30, pp. 4610-4617, SEPT 2009.

[8] S. Shoji and S. Kawata, "Photo-fabrication of 3D photonic crystal by muli-beam laser

interference," Micro- and Nano-photonicMaterialsandDevices, vol. 3937, pp. 12-17,
2000.

[9] K. Khuen Seet, V. Mizeikis, S. Matsuo, S. Juodkazis, and H. Misawa, "Three-dimensional

spiral-architecture photonic crystals obtained by direct laser writing," Advanced Materials,
vol. 17, pp. 541-545, MAR 8, 2005.

[10] Website: http://materials.usask.ca/photos.

[11] K. Aoki, H. T. Miyazaki, H. Hirayama, K. Inoshita, T. Baba, K. Sakoda, N. Shinya, and

Y. Aoyagi, "Microassembly of semiconductor three-dimensional photonic crystals,"
Nature Materials,vol. 2, pp. 117-121, FEB 2003.

[12] G. Zhou and F. Siong Chau, "Three-dimensional photonic crystal by holographic contact
lithography using a single diffraction mask," Applied Physics Letter, vol. 90, pp. 1-3,
2007.
166
[13] Y. Fu, Z. Jin, G. Liu, Y. Yin, "Self-assembly of polystyrene sphere colloidal crystals by in
situ solvent evaporation method," Synthetic Metals, vol. 159, pp. 1744-1750, 2009.

[14] C.J. Summers, E. Graugnard, and J.S. King, "3D luminescent photonic crystal structures,"
ProceedingsofSPIE, vol. 5801, pp. 142-149.

[15] A.D. Apostol, V. Damian, F. Garoi, T. Tordache, P.C. Logofatu, V. Nascov, A. Sima, B.
Dana Cristea, and R. Muller, "2D multiple beam interference lithography," International
Semiconductor Conference ,pp. 151-154, 2006.

[16] S. Shoji, H.-B. Sun, S. Kawata, "Photo-fabrication of wood-pile three-dimensional

photonic crystal by using laser interference," ProceedingsofSPIE, vol. 5000, pp. 35-42,
2003.

[17] S. Ahn, S. Kim, H. Jeon, "Single-defect photonic crystal cavity laser fabricated by a
combined lithography of laser holography and focused ion beam," OSA/CLEO/QELS,
2010.

[18] M. Deubel, M. Wegener, S. Linden, G. von Freymann, and S. John, "3D-2D-3D photonic
crystal heterostructures fabricated by direct laser writing," Optics Letter, vol. 31, pp. 805-
807, MAR 15, 2006.

[19] A. Patel and H. Smith, "Membrane stacking: A new approach for three-dimensional
nanostructure fabrication," Journal of Vacuum Science & Technology B: Microelectronics
and Nanometer Structures, vol. 25, pp. 2662-2664, NOV 2007.

[20] K. Aoki, D. Guimard, M. Nishioka, M. Nomura, S. Iwamoto, and Y. Arakawa, "Coupling

of quantum-dot light emission with a three-dimensional photonic-crystal nanocavity,"
Nature Photonics,vol. 2, pp. 688-692, NOV 2008.

[21] H. Jin In, W. Arora, P. Stellman, S. Kumar, Y. Shao-Horn, H. Smith, and G. Barbastathis,
"The nanostructured origami 3D fabrication and assembly process for nanopatterned 3D
structures," ProceedingsofSPIE, vol. 5763, pp. 84-91, 2005.

[22] P. Temple-Boyer, C. Rossi, E. Saint-Etienne, and E. Scheid, "Residual stress in low

pressure chemical vapor deposition SiNx films deposited from silane and ammonia,"
Journalof Vacuum Science & Technology A, vol. 16, pp. 2003-2007, JUL/AUG 1998.

[23] W. Seok Yang, S. M. Cho, H. Ryu, S. Hoon Cheon, B. Gon Yu, C. Auck Choi,
"Deformation Reduction of A MEMS sensor by stress balancing of multilayer," The
Second InternationalConference on Sensor Technologies andApplications, 2008.

167
[24] W. Arora, A. Nichol, H. Smith, and G. Barbastathis, "Membrane folding to achieve three-
dimensional nanostructures: Nanopattemed silicon nitride folded with stressed chromium
hinges," Applied Physics Letter, vol. 88, pp. 1-3, 2006.

[25] K. Kubota, T. Fleischmann, S. Saravanam, P. Vaccaro, and T. Aida, "Self-assembly of

micro-stage using micro-origami technique on GaAs," Jpn. J. Appl. Phys. 2003

[26] W. Arora, S. Sijbrandi, L. Stem, and J. Notte, "Membrane folding by helium ion
implantation for three-dimensional device fabrication," Journal of Vacuum Science &
Technology B, vol. 25, pp. 2184-2187, NOV/DEC 2007.

[27] E. Hui, R. Howe, and M. Steven Rodgers, "Single-step assembly of complex 3-D
microstructures," Micro Electro MechanicalSystems, 2000. MEMS 2000. The Thirteenth
Annual InternationalConference on, pp. 602-607, 2000.

[28] A. Slocum and A. Weber, "Precision passive mechanical alignment of wafers," Journalof
MicroelectromechanicalSystems, vol. 12, pp. 826-834, DEC 2003.

[29] S.Y.Y. Leung and D.C.C. Lam, "Parametric analysis and design of ultra fine assembly of
micro dies," ElectronicMaterials andPackaging,2007. EMAP 2007. International
Conference on , vol., no., pp.1-6, 19-22 Nov. 2007

[30] 0. Cugat, J. Delamare, and G. Reyne, "Magnetic micro-actuators and systems," IEEE
Transactionson Magnetics, vol. 39, pp. 3607-3612, NOV 2003.

[31] D. Meeker. Finite Element Magnetic Model. www.FEMM.info. 1998.

[32] G. Kartopu, 0. Yalcin, M. Es-Souni, A.C. Basaran. "Magnetization behavior of ordered

and high density Co nanowire arrays with varying aspect ratio," JournalofApplied
Physics , vol.103, no.9, pp.093915-093915-6, May 2008

[33] Porter, D.G.; Donahue, M.J.; McMichael, R.D.; Blue, J.L.;, "OOMMF: Public Domain
Micromagnetic Software," MMM-Intermag Conference, 1998. Abstracts., The 7th Joint,
vol., no., pp. 113, 6-9 Jan 1998

[34] R. Riesenberg, A. Unkroth, and D. Dietrich, "Stress controlled adhesion and magnetic
properties of ion implanted Co films," physica status solidi (a). Volume 112, Issue 2,
pages K85-K89, 16 April 1989

[35] M. Jergel, I. Cheshko, Y. Halahovets, P. Siffalovic, I. Mat'ko, R. Senderak, S. Protsenko,

E. Majkova, and S. Luby, "Annealing behavior of structural and magnetic properties of
evaporated Co thin films," JournalofPhysics D: Applied Physics, vol. 42, pp. 1-8, 2009.

168
[36] N.S. Shaar, G. Barbastathis, C. Livermore. "Cascaded mechanical alignment for
assembling 3D MEMS," Micro Electro MechanicalSystems, 2008. MEMS 2008. IEEE
21st InternationalConference on , vol., no., pp.1064-1068, 13-17 Jan. 2008

[37] Y. Bai, J. Yeow, and B. Wilson, "Design, fabrication, and characteristics of a MEMS
Micromirror with sidewall electrodes," JournalofMicroelectromechanicalSystems, vol.
19, pp. 619-631, JUNE 2010.

[38] E. Kolesar, M. Ruff, W. Odom, S. Ko, J. Howard, P. Allen, J. Wilken, R. Wilks, N.

Boydston, "Polysilicon surface micromachined structural entities with continuous hinges
and microrivets for assembling three-dimensional MEMS devices," Broadband
Communicationsfor the InternetEra Symposium digest, 2001 IEEE Emerging
Technologies Symposium on , vol., no., pp.74-78, 2001

[39] L.J. Hornbeck. "Digital Light Processing and MEMS: an overview," Advanced
Applications of Lasers in MaterialsProcessing/BroadbandOptical Networks/Smart
Pixels/OpticalMEMs and Their Applications. IEEE/LEOS 1996 Summer Topical
Meetings: , vol., no., pp.7-8, 5-9 Aug 1996

[40] M. Culpepper, S. Kim. "A framework and design synthesis tool used to generate evaluate
and optimize compliant mechanism concepts for research and education activities,"
ProceedingsofDETC '04. Salt Lake City Sept 28, 2004.

[41] F.J. Castano, C.A. Ross, C. Frandsen, D. Gil, H.I. Smith. "Magnetization reversal in sub-
micron nanoring arrays," Magnetics Conference, 2003. INTERMAG 2003. IEEE
International, vol., no., pp. GC- 06, 28 March-3 April 2003

[42] S. Hashi, S. Yanase, Y. Okazaki, M. Inoue. "Magnetic properties of FeRh alloy films
prepared by ion-beam sputtering," Magnetics Conference, 2002. INTERMA G Europe
2002. Digest of Technical Papers.2002 IEEE International, vol., no., pp. CC5, 2002

[43] M. Farhoud, J. Ferrera, A. J. Lochtefeld, T. E. Murphy, M. Schattenburg, J. Carter, C. A.

Ross. H. I. Smith, "Fabrication of 200 nm period nanomagnet arrays using interference
lithography and a negative resist," Journalof Vacuum Science & Technology B:
Microelectronicsand Nanometer Structures , vol.17, no.6, pp. 3 182-3185, Nov 1999

[44] A. J. Nichol, W.J. Arora, G. Barbastathis. "Thin membrane self-alignment using

nanomagnets for three-dimensional nanomanufacturing," Journal of Vacuum Science &
Technology B: Microelectronicsand Nanometer Structures , vol.24, no.6, pp.3 128-3132,
Nov 2006

169
[45] A. J. Nichol, P. S. Stellman, W. J. Arora, G. Barbastathis, Two-step magnetic self-
alignment of folded membranes for 3D nanomanufacturing, Microelectronic Engineering,
Volume 84, Issues 5-8, Proceedings of the 32nd International Conference on Micro- and
Nano-Engineering, May-August 2007

[46] A. J. Nichol, G. Barbastathis. "Sub-30nm alignment accuracy between layered photonic

nanostructures using optimized nanomagnet arrays," Optical MEMs and Nanophotonics,
2008 IEEE/LEOS InternationallConference on , vol., no., pp.9-10, 11-14 Aug. 2008

[47] A. J. Nichol, W.J. Arora, G. Barbastathis. "Reconfigurable Nanophotonic Systems By

Tunable Alignment Between Nanomagnet Arrays," Optical MEMS and Nanophotonics,
2007 IEEE/LEOS InternationalConference on , vol., no., pp. 5 1-52, Aug. 12 2007-July 16
2007

[48] Casino Software. http://www.gel.usherbrooke.ca/casino/What.html

[49] A. J. Nichol, M. Deterre, and G. Barbastathis, "Alignment and stretching of compliant

nanomembranes by embedded nanomagnets" EIPBN 2009, Marco Island, Florida, paper
7A.3, June 2009

[50] B. Cord, C. Dames, K.K. Berggren, and J. Aumentado, "Robust shadow-mask evaporation
via lithographically controlled undercut," Journalof Vacuum Science & Technology B,
vol. 24, pp. 1-5, 2006.

[51] Cosmosworks Finite Element Modeling Program. Dessault Systems. 2008

[52] A. Rajca, S. Rajca, J. Wongsriratanakul, C. R. Ross, II, "4,6-Bis-(trifluoromethyl)-N, N':-

di-tert-butyl-1,3-phenylenebis(aminoxyl) and its
Bis(trifluoroacetylacetonato)manganese(II) Complex: Synthesis, X-ray Crystallography,
and Magnetism", Polyhedron, 2001

[53] H. 0. Jacobs and G. M. Whitesides, "Submicrometer patterning of charge in thin-film

electrets," Science, vol. 291, pp. 1763-1766, MAR 2, 2001.

[54] H. Lo and Y Tai, "Parylene-HT-Based electrets rotor generator," MEMS 2008. IEEE 21st
InternationalConference on , vol., no., pp.984-987, 13-17 Jan. 2008

[55] X. Zhang, Y. Zhang, Y. Pan, Z. Xia, X. Liu, B. Li, and Z. Lin, "Mechanistic property and
charge storage in amporphous Si 3N4 electrets film based on silicon by boron ion
implantation," Proceedingsof the 6th InternationalConference on Propertiesand
Applications ofDielectricMaterials,pp. 525-528, June 2000.

170