1

Scaling of MOS Circuits CONTENTS 1. What is scaling? 2. Why scaling? 3. Figure(s) of Merit (FoM) for scaling 4. International Technology Roadmap for Semiconductors (ITRS) 5. Scaling models 6. Scaling factors for device parameters 7. Implications of scaling on design 8. Limitations of scaling 9. Observations 10. Summary

Scaling of MOS Circuits

1.What is Scaling? Proportional adjustment of the dimensions of an electronic device while maintaining the electrical properties of the device, results in a device either larger or smaller than the un-scaled device. Then Which way do we scale the devices for VLSI? BIG and SLOW or SMALL and FAST? What do we gain? 2.Why Scaling?... Scale the devices and wires down, Make the chips fatter functionality, intelligence, memory and faster, Make more chips per wafer increased yield, Make the end user Happy by giving more for less and therefore, make MORE MONEY!! 3.FoM for Scaling Impact of scaling is characterized in terms of several indicators: o Minimum feature size o Number of gates on one chip o Power dissipation o Maximum operational frequency o Die size o Production cost Many of the FoMs can be improved by shrinking the dimensions of transistors and interconnections. Shrinking the separation between features transistors and wires Adjusting doping levels and supply voltages. 3.1 Technology Scaling Goals of scaling the dimensions by 30%: Reduce gate delay by 30% (increase operating frequency by 43%) Double transistor density Reduce energy per transition by 65% (50% power savings @ 43% increase in frequency) Die size used to increase by 14% per generation Technology generation spans 2-3 years

Figure1 to Figure 5 illustrates the technology scaling in terms of minimum feature size, transistor count, prapogation delay, power dissipation and density and technology generations.

10

Minimum Feature Size (micron)

10

10

10

-1

10 1960

-2

1970

1980

1990

2000

2010

Year

Figure-1:Technology Scaling (1)

Figure-2:Technology Scaling (2)

Propagation Delay Figure-3:Technology Scaling (3)

100 Power Dissipation (W)

y /3 4 x rs ea

x1.4 / 3 y

ears

1000 Power Density (mW/mm2 )

3

0.7

10 1

100

10

0.1 0.01 80

MPU DSP 85 Year 90 95

(a) Power dissipation vs. year.

Scaling Factor normalized by 4 m design rule (b) Power density vs. scaling factor.

10

Figure-4:Technology Scaling (4)

Technology Generations

Figure-5:Technology generation

4.

International Technology Roadmap for Semiconductors (ITRS)

Table 1 lists the parameters for various technologies as per ITRS.

Table 1: ITRS

5.Scaling Models Full Scaling (Constant Electrical Field) Ideal model dimensions and voltage scale together by the same scale factor Fixed Voltage Scaling Most common model until recently only the dimensions scale, voltages remain constant General Scaling Most realistic for todays situation voltages and dimensions scale with different factors 6.Scaling Factors for Device Parameters Device scaling modeled in terms of generic scaling factors: 1/ and 1/ 1/: scaling factor for supply voltage VDD and gate oxide thickness D 1/: linear dimensions both horizontal and vertical dimensions

Why is the scaling factor for gate oxide thickness different from other linear horizontal and vertical dimensions? Consider the cross section of the device as in Figure 6,various parameters derived are as follows.

Figure-6:Technology generation

Gate area Ag
Ag = L *W

Where L: Channel length and W: Channel width and both are scaled by 1/ Thus Ag is scaled up by 1/2 Gate capacitance per unit area Co or Cox

Cox = ox/D Where ox is permittivity of gate oxide(thin-ox)= inso and D is the gate oxide thickness scaled by 1/ 1 = Thus Cox is scaled up by 1 Gate capacitance Cg C g = Co * L *W Thus Cg is scaled up by * 1/ 2 =/ 2 Parasitic capacitance Cx

Cx is proportional to Ax/d where d is the depletion width around source or drain and scaled by 1/ Ax is the area of the depletion region around source or drain, scaled by (1/ 2 ). Thus Cx is scaled up by {1/(1/)}* (1/ 2 ) =1/ Carrier density in channel Qon

Qon = Co * Vgs where Qon is the average charge per unit area in the on state. Co is scaled by and Vgs is scaled by 1/ Thus Qon is scaled by 1 Channel Resistance Ron
Ron = L 1 * W Qon *

Where = channel carrier mobility and assumed constant

Thus Ron is scaled by 1. Gate delay Td Td is proportional to Ron*Cg Td is scaled by

1

* =

Maximum operating frequency fo

fo =

W CoVDD * L Cg

fo is inversely proportional to delay Td and is scaled by

1 1 = 2

Saturation current Idss

I dss = Co W 2 * * (Vgs Vt ) 2 L 1

Both Vgs and Vt are scaled by (1/ ). Therefore, Idss is scaled by

Current density J I dss Current density, J = A where A is cross sectional area of the Channel in the on state which is scaled by (1/ 2). So, J is scaled by

1 1

Switching energy per gate Eg

E = 1 C V 2 g g DD 2 So Eg is scaled by

1 1 * 2 = 2 2

Power dissipation per gate Pg

Pg = Pgs + Pgd

Pg comprises of two components: static component Pgs and dynamic component Pgd: Where, the static power component is given by:
Pgd = E g f o
2 V DD Pgs = R on

And the dynamic component by:

Since VDD scales by (1/) and Ron scales by 1, Pgs scales by (1/2). Since Eg scales by (1/2 ) and fo by (2 /), Pgd also scales by (1/2). Therefore, Pg scales by (1/2).

Power dissipation per unit area Pa

1 2 2 Pg = Pa = = Ag 1 2 2

Power speed product PT

PT = Pg * Td =

1 1 = 2 2 2

6.1 Scaling Factors Summary Various device parameters for different scaling models are listed in Table 2 below. Table 2: Device parameters for scaling models NOTE: for Constant E: =; for Constant V: =1
Parameters VDD L W D Ag Co (or Cox) Cg Cx Qon Ron Idss Constant E General (Combined V Description and Dimension) 1/ 1/ Supply voltage Channel length 1/ 1/ Channel width 1/ 1/ Gate oxide thickness 1/ 1/ 2 2 Gate area 1/ 1/ Gate capacitance per unit area 1/ Gate capacitance /2 1/ Parsitic capacitance 1/ 1 1 Carrier density 1 1 Channel resistance 1/ 1/ Saturation current Constant V

1 1/ 1/ 1 2 1/ 1 1/ 1/ 1 1 1
2

10

Parameters

Description Conductor cross section area Current density Logic 1 level Switching energy Power dissipation per gate Gates per unit area Power dissipation per unit area Gate delay Max. operating frequency Power speed product

Ac J Vg Eg Pg N Pa Td fo PT

General Constant E (Combined V and Dimension) 2 2 1/ 1/ 2 / 1/ 1 / 2 1/

2

Constant V

1/ 2 1 1/ 1 2 2 1/ 2 1/

1/ 1 / 3 1/ 2 1 1/ 1 / 3
2

2 2 / 2 / 2 2 / 1 / 2

7.Implications of Scaling
Improved Performance Improved Cost Interconnect Woes Power Woes Productivity Challenges Physical Limits

11

7.1Cost Improvement Moores Law is still going strong as illustrated in Figure 7.

Figure-7:Technology generation

7.2:Interconnect Woes Scaled transistors are steadily improving in delay, but scaled wires are holding constant or getting worse. SIA made a gloomy forecast in 1997 Delay would reach minimum at 250 180 nm, then get worse because of wires But For short wires, such as those inside a logic gate, the wire RC delay is negligible. However, the long wires present a considerable challenge. Scaled transistors are steadily improving in delay, but scaled wires are holding constant or getting worse. SIA made a gloomy forecast in 1997 Delay would reach minimum at 250 180 nm, then get worse because of wires But For short wires, such as those inside a logic gate, the wire RC delay is negligible. However, the long wires present a considerable challenge. Figure 8 illustrates delay Vs. generation in nm for different materials.

Figure-8:Technology generation

12

7.3 Reachable Radius

We cant send a signal across a large fast chip in one cycle anymore But the microarchitect can plan around this as shown in Figure 9. Just as off-chip memory latencies were tolerated
Chip size Scaling of reachable radius

Figure-9:Technology generation

7.4 Dynamic Power Intel VP Patrick Gelsinger (ISSCC 2001)

If scaling continues at present pace, by 2005, high speed processors would have power density of nuclear reactor, by 2010, a rocket nozzle, and by 2015, surface of sun. Business as usual will not work in the future.

Attention to power is increasing(Figure 10)

13

Figure-10:Technology generation

7.5 Static Power VDD decreases

Save dynamic power Protect thin gate oxides and short channels No point in high value because of velocity saturation.

Vt must decrease to maintain device performance But this causes exponential increase in OFF leakage A Major future challenge(Figure 11)

Moore(03) Figure-11:Technology generation 14

7.6 Productivity Transistor count is increasing faster than designer productivity (gates / week)
Bigger design teams Up to 500 for a high-end microprocessor

More expensive design cost Pressure to raise productivity Rely on synthesis, IP blocks

Need for good engineering managers

7.7 Physical Limits o Will Moores Law run out of steam?

Cant build transistors smaller than an atom

o Many reasons have been predicted for end of scaling

Dynamic power Sub-threshold leakage, tunneling Short channel effects Fabrication costs Electro-migration Interconnect delay

o Rumors of demise have been exaggerated 8. Limitations of Scaling

Effects, as a result of scaling down- which eventually become severe enough to prevent further miniaturization. o Substrate doping

o Depletion width o Limits of miniaturization

15

o Limits of interconnect and contact resistance o Limits due to sub threshold currents o Limits on logic levels and supply voltage due to noise o Limits due to current density 8.1 Substrate doping o Substrate doping o Built-in(junction) potential VB depends on substrate doping level can be neglected as long as VB is small compared to VDD. o As length of a MOS transistor is reduced, the depletion region width scaled down to prevent source and drain depletion region from meeting. o the depletion region width d for the junctions is d = o si relative permittivity of silicon o 0 permittivity of free space(8.85*10-14 F/cm) o V effective voltage across the junction Va + Vb o q electron charge o NB doping level of substrate o Va maximum value Vdd-applied voltage
2 si 0V q NB 1

KT N B N D o Vb built in potential and VB = ln q ni ni

8.2 Depletion width N B is increased to reduce d , but this increases threshold voltage Vt trends for scaling down.

-against

Maximum value of N B (1.3*1019 cm-3 , at higher values, maximum electric field applied to gate is insufficient and no channel is formed. N B maintained at satisfactory level in the channel region to reduce the above problem. Emax maximum electric field induced in the junction.
Emax = 2V d

ln

16

If N B is inreased by Va =0 Vb increased by ln and d is decreased by

Electric field across the depletion region is increased by ln 1/

Where

Reach a critical level Ecrit with increasing N B d=

d=

si 0
q NB

2 si Ecrit .d q NB 2

( Ecrit )

Figure 12 , Figure 13 and Figure 14 shows the relation between substrate concentration Vs depletion width , Electric field and transit time. Figure 15 demonstrates the interconnect length Vs. propagation delay and Figure 16 oxide thickness Vs. thermal noise.

Figure-12:Technology generation

17

Figure-13:Technology generation

8.3 Limits of miniaturization minimum size of transistor; process tech and physics of the device
Reduction of geometry; alignment accuracy and resolution Size of transistor measured in terms of channel length L L=2d (to prevent push through) L determined by NB and Vdd Minimum transit time for an electron to travel from source to drain is

v drift = E
t= L Vdrift = 2d E

smaximum

carrier drift velocity is approx. Vsat,regardless of supply voltage

18

Figure-14:Technology generation

8.4 Limits of interconnect and contact resistance

Short distance interconnect- conductor length is scaled by 1/ and resistance is increased by For constant field scaling, I is scaled by 1/ so that IR drop remains constant as a result of scaling.-driving capability/noise margin.

19

Figure-15:Technology generation

8.5 Limits due to subthreshold currents

Major concern in scaling devices. I sub is directly praportinal exp (Vgs Vt ) q/KT As voltages are scaled down, ratio of Vgs-Vt to KT will reduce-so that threshold current increases. Therefore scaling Vgs and Vt together with Vdd . Maximum electric field across a depletion region is

Emax = 2{ Va + Vb }/ d
8.6 Limits on supply voltage due to noise
Decreased inter-feature spacing and greater switching speed result in noise problems

20

Figure-16:Technology generation

9. Observations Device scaling o Gate capacitance per micron is nearly independent of process o But ON resistance * micron improves with process o Gates get faster with scaling (good) o Dynamic power goes down with scaling (good) o Current density goes up with scaling (bad) o Velocity saturation makes lateral scaling unsustainable 9.1 Observations Interconnect scaling o Capacitance per micron is remaining constant o About 0.2 fF/mm o Roughly 1/10 of gate capacitance o Local wires are getting faster o Not quite tracking transistor improvement o But not a major problem o Global wires are getting slower o No longer possible to cross chip in one cycle 10. Summary
Scaling allows people to build more complex machines That run faster too It does not to first order change the difficulty of module design

21

 Module wires will get worse, but only slowly You dont think to rethink your wires in your adder, memory Or even your super-scalar processor core

It does let you design more modules Continued scaling of uniprocessor performance is getting hard -Machines using global resources run into wire limitations -Machines will have to become more explicitly parallel

22