Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

32
R. Dutton, B. Murmann 1 Lecture 29 Technology Scaling– Beyond EE114 MOSFETs R. Dutton, B. Murmann Stanford University EE114

Transcript of Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

Page 1: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 1

Lecture 29Technology Scaling–

Beyond EE114 MOSFETs

R. Dutton, B. MurmannStanford University

EE114

Page 2: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 2

MOS Level 1 Figures-of-Merit (FoM)

Long Channel Model

ds

m

g

g

Current Efficiency

Transit Frequency

Intrinsic Gain

D

m

I

g

OVV

2=

2

OV

L

V

2

3 µ=

gs

m

C

g

OVV

2

!"

•Simplicity of Level 1 has allowed near-perfect accuracyin having hand-calculations and SPICE-simulations agree

•What are the trends in state-of-the-art MOS technology?

•What modeling and methodology is needed foradvanced MOS devices?

!

IDS =KP

2"W

Leff" VGS #Vt( )

2

" 1+ $VDS( )

Vt =VTO + % 2& #VBS # 2&( )

EE114

Page 3: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 3

1.8V 3.3V

CMOS core

I/O for HV interface

DE core for current sources

Ids

??

STIDrain-Extended

gm/gds

Trends in Scaling

BSIM

Different “flavors” of Transistors:Core Digital, I/O, Special Devices

(we’ll comeback to special

devices)

Two Problems:

•Decrease in intrinsic gain

•Bias dependent

EE114

Page 4: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 4

Dependence on VDS

• The long channel model predicts that gds and gm/gds areindependent of VDS

– As long as device is biased in active region

• This is also no longer true in modern devices– gds (and therefore gm/gds) shows a significant

dependence on VDS

VDS

ID

OP1

Slope = gds1OP2

Slope = gds2

Physical effects beyond“channel length modulation”critical

EE114

Page 5: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 5

Observations – Intrinsic Gain

• gm/gds shows a strong dependence on VDS bias

– Mostly due to varying gds

• There is a gradual transition from triode to active

– Long channel model would have predicted an abrupt changeto large intrinsic gain at VDS = VOV

– Typically need VDS > VOV + 4kT/q to ensure at leastmoderate intrinsic gain

• At high VDS, gds increases due to SCBE (substrate currentinduced body-effect); this causes a decrease in gm/gds

– Highly technology dependent, and usually not present inPMOS devices

– If you are interested in more details, please refer to EE316or a similar course

EE114

Page 6: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 6

The Good News

•Large improvements in fT

•Product of gm/Id times fT isbecoming sharper; closer tothe sub-threshold region

EE114

Page 7: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 7

MOS R&D Prototypes vs. Production Versions

ProductionPrototypes:

R&D PrototypeTechnology:

Non-classical structures, physics limited, drasticvariations, and higher cost

Power- and robustness-constrained, adaptive,billion-scale integration, gigaHz operation

65nm(2005)

45nm(2007)

32nm(2009)

22nm(2011)

16nm(2013)

11nm(2015)

30nm(2000)

20nm(2001)

15nm(2001)

10nm(2003)

7nm(2005)

5nm(?)

Ground Plane

BOX (<20nm)Bulk wafer

FD Si film

Ground Plane

BOX (<20nm)Bulk wafer

FD Si film

Source: Intel, ITRS

EE114

Page 8: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 8

Example--45 nm Technology Node

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

1E-9

1E-8

1E-7

1E-6

1E-5

1E-4

1E-3

0.01 Fujitsu 45nm nodeL

eff=25nm (N)

35nm(P)1V

0.05V

Ids (

A/µm

)

Vgs

(V)

Data PTM

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

0

100

200

300

400

500

600

700

800

900

0.4V

0.55V

0.7V

0.85V

1VFujitsu 45nm nodeL

eff=25nm (N)

35nm (P)

Ids(µ

A/µ

m)

Vds

(V)

Data PTM

Published data;analysis K. Cao, ASU

•Vgs dependence more linear than quadratic

•Output conductance limits intrinsic gain

•Sub-threshold behavior limits gain and leakage

Sub-Vt

EE114

Page 9: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 9

Empirical Compact Models

2. Physical Parameters

Nch, Eta0, U0, Vsat, K1

estimated based on tech. specs.or extrapolated from data; thenrefined to fit typical IV curves

1. Tech. Specs.

Leff, Tox, Vth, VDD, Rds

(and Tsi for FinFET)

from literatureincluding ITRS

3. SecondaryParameters (~80)

either fitted orremained from

previous generations

Several Examples*:

•BSIM (UC Berkeley)--broadly used and supported (de-facto standard)

•PSP (Penn. State-Philips)--growing support (Europe)

•HISIM--strong support in Asia (Hiroshima U., Japan)

•PTM (Predictive Tech. Mod.)--evolution from BSIM (K. Cao, ASU)

*Compact Modeling Council (CMC) is an organization that benchmarks models

EE114

Page 10: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 10

Problems with Compact Models vs. Level 1

Circuits designed using hand calculations will typically show largediscrepancies between targeted specs and SPICE results

Specifications

Hand Calculations

Circuit

Spice

Results

Square Law

BSIMBSIM, PTM

EE114Model

EE214Model

EE114

Page 11: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 11

Alternative Approach (Solution…EE214)

Use pre-computed SPICE data in hand calculations

Specifications

Hand Calculations

Circuit

Spice

Results

Design Charts

BSIM

SpiceBSIM

CompactModel:

BSIM,PSP,

PTM…

BSIM, PTM…

Must be same

model!

EE114

Page 12: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 12

Example--Intrinsic Gain Simulation (VDS)

* gm/gds versus vds

.param vt1=571.5m

mn1 d g 0 0 nch214 L=0.35um W=10um

vg g 0 dc 'vt1+0.2'vd d 0 dc 1.5

.op

.dc vd 0 3 10m

.probe gm1 = par('gmo(mn1)')

.probe gds1 = par('gdso(mn1)')

.options post brief

.lib './ee214_hspice.txt' nominal

.end

VGS=V

t+200mV

EE114

Page 13: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 13

Results

0 1 2 30

0.5

1

1.5x 10

-3

VDS

[V]

gm

[S]

0 1 2 30

2

4

6x 10

4

VDS

[V]

1/g

ds

[!]

0 0.5 1 1.5 2 2.5 30

10

20

30

40

50

60

70

80

NMOS, W/L=10/0.35, VOV

=200mV

VDS

[V]

gm

/g d

s

Intr

insi

c G

ain

Vds

Ids-Vds

(see commentabout VOV + 4kT/q)

EE114

Page 14: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 14

Example--fT Simulation

* ft versus gate overdrive

.param gs=1

vds d 0 dc 1.5Vvgs g 0 dc 'gs'mn1 d g 0 0 nch214 L=0.35um W=10um

.op

.dc gs 0.4V 1.2V 10mV

.probe ov = par('gs-vth(mn1)')

.probe ft = par('1/2/3.142*gmo(mn1)/(-cgsbo(mn1))')

.options post brief dccap* Note: "dccap" forces HSpice to recalculate caps in* each simulation step (instead of using constant .op* value). See HSpice manual for additional info.

.lib './ee214_hspice.txt' nominal

.end

10/0.35

VDD/2

EE114

Page 15: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 15

Results

22

3

2

1

L

Vf OVT

µ

!=Long channel model:

-0.2 -0.1 0 0.1 0.2 0.3 0.4 0.50

5

10

15

20

25

30NMOS W/L=10/0.35

VO V

[V]

fT [

GH

z]

EE214 technologyLong Channel FitEE114 technology

EE114

Page 16: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 16

More About Device Technology

-0.2 -0.1 0 0.1 0.2 0.3 0.4 0.50

5

10

15

20

25

30

35

40

VOV

[V]

gm

/I D [S/A]

EE214 technology

2/VOV

BJT (q/kT)Subthreshold

Operation

EE114 technology

Transition

Region

Sca

led

Tra

nsc

on

du

ctan

ce (

gm

/Id

s)

EE114

Page 17: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 17

Moderate Inversion

• In the transition region between subthreshold and stronginversion, we have two different current mechanisms

dx

dn

q

kT

dx

dnD - (BJT)Diffusion

E - (MOS)Drift

µ!

µ!

==

=

• Both current components are always present

– Neither one clearly dominates around Vt

• Can show that ratio of drift/diffusion current ~(VGS-Vt)/(kT/q)

– MOS equation becomes dominant at several kT/q

EE114

Page 18: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 18

Mobile Charge on a Log Scale

• On a log scale, we see that there are mobile charges before wereach the threshold voltage– Fundamental result of solid-state physics, not short channels

1.E-16

1.E-15

1.E-14

1.E-13

1.E-12

1.E-11

1.E-10

1.E-09

1.E-08

1.E-07

1.E-06

-1.00 -0.50 0.00 0.50 1.00

VOV [V]

Mo

bile C

harg

e [

C]

(where is Vt onthis plot?)

EE114

Page 19: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 19

BJT Similarity

• We have– An NPN sandwich, mobile minority carriers in the P region

• This is a BJT!– Except that the base potential is here controlled through a capacitive

divider, and not directly an electrode

EE114

Page 20: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 20

Subthreshold Current

• We know that for a BJT

)//( qkTVSC

BEeII !"

• In MOS case we have)//()(

0qnkTVV

DtGSeII

!"#

• n is given by the capacitive divider

ox

js

ox

oxjs

C

C

C

CCn +=

+= 1

where Cjs is the depletionlayer capacitance

• In EE214 technology n ≅ 1.5

EE114

Page 21: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 21

Subthreshold Transconductance

• Similar to BJT, but unfortunately n (≅1.5) times lower

kT

qI

ndV

dIg D

GS

Dm

!==1

kT

q

ndV

dI

I

g

GS

D

D

m 1==

-0.2 -0.1 0 0.1 0.2 0.3 0.4 0.50

5

10

15

20

25

30

35

40

VOV

[V]

gm

/I D [S/A]

EE214 technology

2/VOV

BJT (q/kT)~1.5x

(where did thisvalue come from?)

EE114

Page 22: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 22

Other Issues (and Research Perspective)

There are many, many other issues:

•Distortion

•Noise

•Emerging Technologies (“more than Moore”)

•Analog vs. Digital Scaling

THE Issues facinganalog scaling

EE114

Page 23: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 23

Noise in Electrical Elements

• Noise is unwanted signal fluctuations.– Equilibrium: Thermal noise (PSD=4kTg)– Non-Equilibrium: Who knows?

Equilibrium(thermal noise)

R=1/g 4kTg A famous Non-equilibrium case (shot noise)

ID 2qID

PSD

f

22,nnVI

Ec

N P

x

EE114

Page 24: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 24

Noise in MOSFETs

• MOSFET noise sources:Drain

Source

Gate inding

f

ind

!

2

f

1/f noise

White noisef

ing

!

2

f!

" Cgs#( )2

CarrierFluctuations

N+ N+

GateSource Drain

EE114

Page 25: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 25

MOSFET Noise Formulation

• Classical long-channel formulation:

( ) dond gkTf

i3/24

2

=!

( )( )dogs

nggCkT

f

i5/3/44

22

2

!="

j

ii

ii

ngnd

ngnd395.0

22

=

N+ N+

GateSource Drain

2

2

)(

4!"

#$%

&!"

#$%

&=

R

dR

dR

kTId eqn

dR

(Equilibrium Noise)

γ

δ

(a different“gamma”)

(correlation ofdrain and gate

noise)

EE114

Page 26: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 26

“What if” MOS Scaling* (45nm)

NML(increase L)

Asymmetry(omit HALO)

•Source/Drain dopinglevels and junctiondepths, based on ITRSprojections•Lateral dopingprofiles adjusted toachieve the necessarysub-threshold slope

*Dutton research group (TCAD) simulations

EE114

Page 27: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 27

TCAD Simulations: ML-, A-, NML-devices

•Minimum Length (ML)device follows traditionalscaling (Well-Tempered)•Non-ML (NML) devicebacks off on channellength (from minimum)but retains both HALOimplants•Asymmetric device (A)removes the drain-sideHALO and adjusts spacingto maintain sub-thresholdslope

(45nm)

NML

A Dev.

Intr

insi

c G

ain

(gmr o

)

EE114

Page 28: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 28

Looking at each part separately…

gm ro

EE114

Page 29: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 29

A Random PhD Quals Question

A device technologist has created aNEW DEVICE*-- part MOS, partBJT--and with “drive current”given by:IDrive=ID(MOS)+IC(BJT).The device cross-section is shownbelow

To achieve maximum small-signal (ac) gain, how shouldwe bias the NEW DEVICE?

Cgs C!

VinVout

RS

NEW

DEVICE

*Comment: you might ask THEcritical question, “is this device reallyuseful?” (I.e. what is the down-side ofdoing this, aside from technology)

EE114

Page 30: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 30

Hypothetical MOS-BJT Device

LL

Gate 1

Gate 2(=Base)

Common

(Ground)Output

Node

W

N+

P-typeN+

COX (F/cm2)=

4x10-6 F/cm2

EE114

Page 31: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 31

(dc) Log I vs V--MOS vs BJT

Log 1

0 I (I

DS o

r IC)

Drive Voltage(VGS or VBE)

slope =q/kT

“sub

-Thr

esho

ld”

“power-law”

~(VGS-VT)m

VT

MOS

BJT

!

gm "#Ioutput

#Vinput$drive

(Larger slopeimplies larger gm)

VBE(on)

!

"eVBE

Vthermal#

$ %

&

' (

!

"eVGS

mVthermal#

$ %

&

' (

EE114

Page 32: Lecture on Technology Scaling (R. Dutton. B. Murmann, Stanford)

R. Dutton, B. Murmann 32

Summary

•It’s not likely that this lecture covered all the slidesshown above but…what’s important?

•EE114 technology has made it (relatively) easy to get amatch between hand-calculations and SPICE but…MOSscaling is creating much more complex devices (practicalmodels are now primarily empirical)

•EE214 technology and modeling will deal with thesecomplexities using a methodology that creates designcurves (I.e. gm/Id etc.) that can match SPICE and supportpractical circuit design

•Other issues--distortion, noise, etc.--are critical and willbe addressed as well (ongoing challenges for research)

EE114