Resistive communications based on neuristors

David Alejandro Trejo Pizzo1

Abstract Memristors are passive elements that allow us to The original definition of the memristor is derived from
store information using a single element per bit. However, this circuit theory: besides resistor, capacitor and inductor, there
is not the only utility of the memristor. Considering the physical must exist a fourth basic two-terminal element that uniquely
chemical structure of the element used, the memristor can
function at the same time as memory and as a communication defines the relationship between the magnetic flux and the
unit. This paper presents a new approach to the use of the electric charge q passing through the device, or
memristor and develops the concept of resistive communication.
d = M dq
I. INTRODUCTION
A. ReRAM
The memristor is the fundamental non-linear circuit ele-
ment, with uses in computing and computer memory. Mem- ReRAM (Resistive Random Access Memory) is a resistive
ristors are basically a fourth class of electrical circuit, joining switching memory proposed as a nonvolatile memory. The
the resistor, the capacitor, and the inductor, that exhibit their phenomenon of resistive switching has been observed in a
unique properties primarily at the nanoscale. wide variety of materials, however, the mechanism responsi-
Theoretically, a memristor is a concatenation of a memory ble for the switching behavior seems to differ between them.
and a resistor that maintains a relationship between the time On the basis of I-V characteristic curves, switching behaviors
integrals of current and voltage across a two terminal ele- can be classified into two types, unipolar (non-polar) and
ment. Thus, its resistance varies according to a memristance bipolar.
function. In unipolar resistive switching, the switching process de-
pends on the amplitude of the applied voltage, but not on the
polarity. This type of switching behavior has been observed
in many binary metal oxides such as SiO, NiO and CuO.

Fig. 2. I-V curve of a CuO memristor

Fig. 1. Fundamental circuit elements
A memory cell initially in a virgin state, ie prior to the
The material implementation of memristive effects can application of any voltage pulse, is in a highly resistive state.
be determined in part by the presence of hysteresis (an When the application of a voltage pulse is high enough,
accelerating rate of change as an object moves from one it is put into the device in a low resistance state (LRS).
state to another) which, like many other non-linear anomalies This process is called forming. The rupture value of the
turns out to be less an anomaly than a fundamental property dielectric of the material tells us approximately at what range
of passive circuitry. of voltages we should find this threshold voltage for the
forming process. After the forming process, the cell from
*This work was not supported by any organization an LRS state is switched to a high resistance state (HRS)
1 David Alejandro Trejo Pizzo is with NyVind Research Labs and
School of Engineering, Universidad de Palermo, C1175ABT Buenos Aires, by applying a threshold voltage (reset process). Changing
Argentina dtrejopizzo@gmail.com from an HRS state to an LRS state is achieved by applying a
threshold voltage that is usually greater than the reset voltage, III. RESISTIVE COMMUNICATIONS
but less than the forming voltage. In the process of set and We know that neurons communicate with each other
forming, the current is limited by the compliant current (Icc) through the small spaces between them, in a process known
of the control system or, more practically, by adding a series as synaptic transmission (where synapses are the connections
resistor acting as a voltage divider. between neurons). Information goes from one cell to another
by neurotransmitters such as glutamate, dopamine or sero-
II. MEMRISTOR CIRCUITS tonin, which activate the receptors in the receiving neuron to
For almost fifty years, integrated electronic circuits built transmit excitatory or inhibitory messages.
with semiconductor devices have provided significant growth Three main types of synaptic transmission are distin-
in the number of processing elements and memory bits guished. The first two mechanisms constitute the main forces
available to system developers. governing the neural circuits and can be replicated using
memristor circuits, from now on called neuristors:
Exciting transmission: one that increases the possibility
of producing an action potential;
Inhibitory transmission: that which reduces the possi-
bility of producing an action potential;
Modulating transmission: that which changes the pattern
and / or frequency of the activity produced by the cells
involved.
As it is shown in A scalable neuristor built with Mott
memristors and in figure 5 a neuristor can be made with
Mott memristors. The channels consist of Motts M1 and
M2, each with a characteristic parallel capacitance (C1 and
C2, respectively) and are biased with opposite polarity d.c.
voltage sources.
Fig. 3. Designend array of neuristors contacts

This growth has provided orders of magnitude improve-

ments in speed, power consumption, and reliability, together
with significant reductions in the cost per device but trends
like this are direct consequences of frequent miniaturiza-
tion in the semiconductor fabrication process. According to
Moore Law this will reach an end soon. When device sizes
are no longer scalable, microelectronic technology needs
innovations to support novel applications and here comes
the memristor.

Fig. 5. Neuristor made with Mott memristors

A. The conduction mechanism

To support the general function of the nervous system,
Fig. 4. Neuristor contact neurons have evolved unique capabilities for intracellular
signaling (communication within the cell) and intercellular
Memristors are used in hardware neural networks, both to signaling (communication between cells). To achieve long
implement different learning algorithms and back propaga- distance, rapid communication, neurons have evolved special
tion. It is also used to build neuromorphic systems (hardware abilities for sending electrical signals (action potentials)
that mimics the brain). Memristors can also be used in analog along axons. This mechanism, called conduction, is how the
circuits, for example as reconfigurable resistors to change the cell body of a neuron communicates with its own terminals
properties of the circuit. Another interesting application is the via the axon. Communication between neurons is achieved
use of memristors as part of logic circuits. at synapses by the process of neurotransmission.
 We can use the neuristor M to achieve this communication This signal is then propagated along the axon (and not,
as it is shown in figure 6. say, back to its dendrites) until it reaches its axon terminals.
An action potential travels along the axon quickly, moving
at rates up to 150 meters (or roughly 500 feet) per second.
Conduction ends at the axon terminals.
Axon terminals are where neurotransmission begins.
Hence, it is at axon terminals where the neuron sends its
Vout to other neurons. At electrical synapses, the Vout will
be the electrical signal itself. Neuristors communicate with
other neuristors with every set and reset action, changing
the resistance level and forming nano wires that can save
the memory of past memristance levels. This conducction
mechanism is what I would like to call resistive communi-
cation.
B. A nanoscale communication protocol
Fig. 6. Neuristor proposed circuit for synapses
Resistive communications is an approach to communicate
To begin conduction, a potential is generated near the cell between physical layers of neuristor with focus in creating
body portion of the axon, here the Vin of the Q transistor. neuromorphic circuits. To make this communication mech-
But whereas an electrical signal in an electronic device anism compatible with other nano devices it is essential
occurs because electrons move along a wire, an electrical to use a standard. A conceptual framework provides the
signal in a neuron occurs because ions move across the organization and structure required to develop conceptual
neuronal membrane. Ions are electrically charged particles. models of nanoscale communication. The IEEE Std 1906.1
The protein membrane of a neuron acts as a barrier to ions. Recommended Practice for Nanoscale and Molecular Com-
Ions move across the membrane through ion channels, here munication Framework provides this precise, common defi-
nanowires in a dielectric thin layer, that open and close due nition of nanoscale communication and a general framework
to the presence of neurotransmitters. that balances concepts with broad applicability. This includes
metrics, use-cases, and a reference model witch it is followed
here.
C. Reference model for molecular communications
The core of the simulator has been also extended to model
molecular communications based on the pure diffusion pro-
cess. the UML diagram of classes modeling the Molecular
example. Also in this case, the diagram only reports the most
important data members and functions, whereas some details
about relationships among objects have been omitted.
It is assumed that molecules diffuse into the medium ac-
cording to Brownian motion. In that hypothesis propagation
of this pulse can be analytically modeled by Ficks laws of
diffusion, which expresses the concentration of molecules as
a function of distance and time. In particular, the molecular
concentration at any point in space is expressed in Equation
Q r 2
c(r, t) = 3/2
e 4Dt
(4 D t)
Fig. 7. Neuristor simplified grid
where
When the concentration of ions on the inside of the c is the molecular concentration
neuron changes, the electrical property of the membrane r is the distance between sender and receiver
itself changes. Normally, the membrane potential of a neuron Q is the number of molecules released by the sender
rests as -70 millivolts (and the membrane is said to be D is the diffusion coefficient
polarized). The influx and outflux of ions (through ion t is the time variable
channels during neurotransmission) will make the inside of
the target neuron more positive (hence, de-polarized). When IV. CONCLUSIONS
this depolarization reaches a point of no return called a The resistive communication presents a model on which
Vthreshold, a large electrical signal is generated. This is the to build neuromorphic networks that emulate the functioning
action potential. of the electrical synapse and the communication of data
 [8] S. N. Truong, K. Van Pham, W. Yang and K. S. Min, Memristor
circuits and systems for future computing and bio-inspired information
processing, 2016 IEEE Biomedical Circuits and Systems Conference
(BioCAS), Shanghai, 2016, pp. 456-459.
[9] M. Khalid and J. Singh, Memristor Crossbar-Based Pattern Recog-
nition Circuit Using Perceptron Learning Rule, 2016 IEEE Interna-
tional Symposium on Nanoelectronic and Information Systems (iNIS),
Gwalior, 2016, pp. 236-239.
[10] I. Gupta, A. Serb, A. Khiat and T. Prodromakis, Towards a memristor-
based spike-sorting platform, 2016 IEEE Biomedical Circuits and
Systems Conference (BioCAS), Shanghai, 2016, pp. 408-411.
[11] J. Secco and F. Corinto, Memristor-Based Binary Synapses for
Deep Neural Networks, CNNA 2016; 15th International Workshop
on Cellular Nanoscale Networks and their Applications, Dresden,
Germany, 2016, pp. 1-2.
[12] A. Ascoli, R. Tetzlaff, L. O. Chua, J. P. Strachan and R. S. Williams,
DC behaviour of a non-volatile memristor: part I, CNNA 2016;
15th International Workshop on Cellular Nanoscale Networks and their
Applications, Dresden, Germany, 2016, pp. 1-2.
[13] I. Vourkas, A. Abusleme, G. C. Sirakoulis and A. Rubio, 1-D
Memristor Networks as Ternary Storage Cells, CNNA 2016; 15th
International Workshop on Cellular Nanoscale Networks and their
Applications, Dresden, Germany, 2016, pp. 1-2.
[14] L. V. Gambuzza, M. Frasca, L. Fortuna, V. Ntinas, I. Vourkas and G.
C. Sirakoulis, A new approach based on memristor crossbar for syn-
chronization, CNNA 2016; 15th International Workshop on Cellular
Nanoscale Networks and their Applications, Dresden, Germany, 2016,
pp. 1-2.
Fig. 8. IEEE P1906 reference model [15] S. N. Danilin and S. A. Shchanikov, The research of operation
accuracy of a memristor-based artificial neural network with an input
signal containing noise and pulse interference, 2016 Dynamics of
Systems, Mechanisms and Machines (Dynamics), Omsk, 2016, pp.
between neurons. Much work remains to be done to build a 1-5.
[16] A. Siemon, S. Ferch, S. Menzel, R. Waser and E. Linn, Impact of
neuromorphic circuit that emulates brain functions, but we Quantized Conductance Effects of ReRAM Devices on Neuromorphic
are closer to achieving small neural networks to physically Networks, CNNA 2016; 15th International Workshop on Cellular
build pattern recognition algorithms. The memristor and the Nanoscale Networks and their Applications, Dresden, Germany, 2016,
pp. 1-2.
neuristor construction are the ideal elements for its physical [17] M. Prezioso et al., Spiking neuromorphic networks with metal-oxide
characteristics to build neural networks and emulate the memristors, 2016 IEEE International Symposium on Circuits and
electrical synapse using the resistive switching phenomenon. Systems (ISCAS), Montreal, QC, 2016, pp. 177-180.
[18] A. Ascoli, R. Tetzlaff, V. Lanza, F. Corinto and M. Gilli, Memristor
plasticity enables emergence of synchronization in neuromorphic net-
works, 2014 IEEE International Symposium on Circuits and Systems
R EFERENCES (ISCAS), Melbourne VIC, 2014, pp. 2261-2264.
[19] V. Erokhin, Organic memristive devices: Architecture, properties and
[1] Zheng Fan, Xudong Fan, A. Li and Lixin Dong, Resistive switching applications in neuromorphic networks, 2013 IEEE 20th International
in copper oxide nanowire-based memristor, 2012 12th IEEE Interna- Conference on Electronics, Circuits, and Systems (ICECS), Abu
tional Conference on Nanotechnology (IEEE-NANO), Birmingham, Dhabi, 2013, pp. 305-308.
2012, pp. 1-4. [20] O. Turel, I. Muckra and K. Likharev, Possible nanoelectronic imple-
[2] Y. Zhang, Y. Li, X. Wang and E. G. Friedman, Synaptic Charac- mentation of neuromorphic networks, Proceedings of the International
teristics of Ag/AgInSbTe/Ta-Based Memristor for Pattern Recognition Joint Conference on Neural Networks, 2003., 2003, pp. 365-370 vol.1.
Applications, in IEEE Transactions on Electron Devices, vol. 64, no. [21] C. Liu, Q. Yang, C. Zhang, H. Jiang, Q. Wu and H. H. Li, A
4, pp. 1806-1811, April 2017. memristor-based neuromorphic engine with a current sensing scheme
[3] M. R. Azghadi, B. Linares-Barranco, D. Abbott and P. H. W. Leong, for artificial neural network applications, 2017 22nd Asia and South
A Hybrid CMOS-Memristor Neuromorphic Synapse, in IEEE Trans- Pacific Design Automation Conference (ASP-DAC), Chiba, 2017, pp.
actions on Biomedical Circuits and Systems, vol. 11, no. 2, pp. 434- 647-652.
445, April 2017. [22] A. Vandesompele, F. Walter and F. Rhrbein, Neuro-evolution of
[4] H. Abunahla, D. Shehada, C. Y. Yeun, C. J. OKelly, M. A. Jaoude spiking neural networks on SpiNNaker neuromorphic hardware,
and B. Mohammad, Novel microscale memristor with uniqueness 2016 IEEE Symposium Series on Computational Intelligence (SSCI),
property for securing communications, 2016 IEEE 59th International Athens, 2016, pp. 1-6.
Midwest Symposium on Circuits and Systems (MWSCAS), Abu [23] T. E. Potok et al., A Study of Complex Deep Learning Networks on
Dhabi, United Arab Emirates, 2016, pp. 1-4. High Performance, Neuromorphic, and Quantum Computers, 2016
2nd Workshop on Machine Learning in HPC Environments (MLHPC),
[5] C. Liu, Q. Yang, C. Zhang, H. Jiang, Q. Wu and H. H. Li, A
Salt Lake City, UT, 2016, pp. 47-55.
memristor-based neuromorphic engine with a current sensing scheme
for artificial neural network applications, 2017 22nd Asia and South
Pacific Design Automation Conference (ASP-DAC), Chiba, 2017, pp.
647-652.
[6] S. Wang, W. Wang, C. Yakopcic, E. Shin, T. M. Taha and G. Subra-
manyam, Memristor devices for use in neuromorphic systems, 2016
IEEE National Aerospace and Electronics Conference (NAECON) and
Ohio Innovation Summit (OIS), Dayton, OH, 2016, pp. 253-257.
[7] M. M. Dakheel, A. M. Hassanein, R. A. Fouad and A. G. Radwan,
Memristor-based data converter circuits, 2016 28th International
Conference on Microelectronics (ICM), Giza, 2016, pp. 357-360.