Digital VLSI Design

Full Automation Maximum benefit of scaling High speed , low power Robustness

Design metrics

INVERTER
STATIC CHARACTERISTICS

VTC DESIGN ISSUES

STATIC POWER CONSUMPTION FULL LOGIC LEVELS SHARP TRANSITION SWITCHING THRESHOLD NOISE MARGINS

PRACTICAL VTC

FIVE CRITICAL VOLTAGES

SWITCHING THRESHOLD
Vth Output changes its state

Noise Margins

Implementation Resistive load

Design for Vol

SAT. ENHANCEMENT LOAD INV.

LIN. ENHANCEMENT LOAD INV.

Static characteristics

Operating regions

VOH

VOL

VIL

VIH

V th -switching threshold

Critical voltage
Nothing We can design for wide noise margins Set Vth= Vdd

Why design for Vth Vdd?

Choose appropriate VM

Velocity saturated device

MOS

CONSTANT

Long Channel Vs. Short Channel

SAME

Long Channel Vs. Short Channel Id vs Vgs

Sub-threshold current

Sub-threshold operation
Required----

For velocity saturated device

Estimation of NM USING Piecewise lin. approx.

Determine g at Vin~Vm

Variation in VM by (w/L)

Impact Of Device Variations on Vm

Critical voltage
Nothing We can design for wide noise margins Set Vth= Vdd

Why design for Vth Vdd?

Choose appropriate VM

Effect on kR

Reducing supply voltage

Switching characteristics

Delay Definitions-with input slope

Capacitive load

Cgd

Cdb under transient conditions

Equivalence factor

m= for abrupt junction

Clock (Charge) feedthrough effect

Delay calculation method 1

CMOS Inverter Driving a Lumped Capacitance Load

CMOS Inverter can be viewed as a single transistor either charging the Cload or discharging the Cload Vin is assumed to switch abruptly If Vin switches high, the NMOS Tx discharges Cload while the PMOS Tx turns OFF If Vin switches low, the PMOS Tx charges Cload while the NMOS Tx turns OFF Cload is comprised of Cgate due to the gate capacitance of receiving circuits Cwire of the interconnect metal Cparasitics of the inverter output junctions

Switch Model of CMOS Transistor

MODEL-1
Approximate as a simple RC network where R is given as an equivalent resistance of the NMOS and PMOS devices and C is given as the total lumped Cload capacitance

|V GS|

Ron

|VGS| < |VT |

|VGS| > |VT |

CMOS Inverter: Transient Response

Switch model

Vout = VDD (1 e t / RONCL )

VDD

Vout = VDD (e t / RONCL )

tpHL = f(R [Link]) = 0.69 RonCL

Vout Vout CL Ron

1

ln(0.5)

VDD

0.5 0.36

Vin = V DD RonCL
t

Determination of Req

In velocity saturated device

Method 2

CMOS Inverter Propagation Delay

(AVERAGE CURRENT THROUGH LOAD)
V DD

t pHL = C

(V 50% -VDD) I av

V out

I av

CL

tpLH = C L (V50%-VOL)

Iav
V in = V
DD

WHERE

Iav, HL = [ic(VIN=VOH, VOUT= VOH)]+ ic(VIN=VOH, VOUT= V50%)]

Iav, LH = [ic(VIN=VOL, VOUT= V50%)]+ ic(VIN=VOL, VOUT= VOL)] SIMPLE Drawback-----neglects variation of cap. Load during the entire transition

Method-3
Differential equation approach
accurate

tpHL

tpLH

Impact of Rise Time on Delay

0.35

0.3

tpHL(nsec)

0.25

0.2

0.15

0.2

0.4 0.6 trise (nsec)

0.8

Input slope

or

0.25

Design for Performance-(speed)

Keep capacitances small Increase transistor sizes
watch out for self-loading!

Increase VDD

Delay as a function of VDD

5.5 5 4.5 4 3.5 3 2.5 2 1.5 1 0.8

tp(normalized)

1.2

1.4

1.6

1.8

2.2

2.4

DD

(V)

Device Sizing
3.8 3.6 3.4 3.2
tp(sec)

x 10

-11

(for fixed load)

3 2.8 2.6 2.4 2.2 2 2 4 6 8 S 10 12 14

Self-loading effect: Intrinsic capacitances dominate

DELAY REDUCTION

Delay as a function of VDD()

28 24

Normalized Delay

20 16 12 8 4 0 1.00 2.00 3.00 4.00 5.00

VDD (V)

Delay as a function of CL()

DELAY CL Delay as a function of W/L() DELAY (W/L)-1

Need of simple delay model

Delay depends on many factorscharge, discharge, parasitic, w/L, fan in- fanout, topology Existing delay models do not give clear indication of contribution of each factor Circuit designers waste too much time simulating and tweaking circuits

85

Using LE in design of inverter chain

86

CKT DESIGN PROBLEMS

Chip designers face a bewildering array of choices. What is the best circuit topology for a function? How large should the transistors be? How many stages of logic give least delay?
87

Need of simple delay model

Circuit designers waste too much time simulating and tweaking circuits High speed logic designers need to know where time is going in their logic CAD engineers need to understand circuits to build better tools

88

Delay in a Logic Gate

89

Delay contributors
is speed of basic transistor p-intrinsic delay of the gate due to its own internal capacitances hcombines the effect of external load with sizes of transistors g effect of circuit topology

90

Observations
Logical effort describes relative ability of gate topology to deliver current [defined to be 1(best av. of charge and discharge both] for an inverter) Electrical effort is the ratio of output to input capacitance Delay increases with electrical effort Delay increases ---More complex gates have greater logical effort and parasitic delay
91

Estimation of

CMOS Ring Oscillator Circuit

An odd number of inverter circuits connected serially with output brought back to input will be astable and can be used an an oscillator (called a ring oscillator) Ring oscillators are typically used to characterize a new technology as to its intrinsic device performance Frequency and stage are related as follows: f = 1/T = 1/(2nP) where n is the number of stages and P is the stage delay

Ring OscillatorCOMPARING
DIFFERENT TECHNOLOGIES

T=2

tp N

2 N tp >> tf +tr

Computing Logical Effort

95

Different gates

96

Observations
More complex gates have larger logical efforts Logical efforts grow with increase in no. of inputs Complex gates exhibit high g, greater delay

97

Parasitic delay
It is fixed for a gate More complex gatehigher parasitic delay Ref. Pinv=1 (inverter parasitic delay ) For other gates , parasitic delay is written in terms of pinv

98

Parasitic delay

99

How to compute Pinv

For inv. g=1, dabs= (h+pinv)

In a given tech., plot d vs. h

Plot would be st. line with slope , & intercept(pinv )

Pinv can be estimated after obtaining

Draw similar plot for other gates Once is obtained , g and p of other gates can be found out.
100

Delay equation plot

101

Choice Of Standard Reference

Calculating delay of an inverter

103

Delay of 2 input Nand gate

Delay of 2 input NOR gate

Skewed Gates
Best WP/WN for min delay other then un/up

Using logical effort

Define three parameters for a gate r = wp/wn = p/n = pull up path/ pull down path = n / p

r=k

Case---==2 for inv.

108

Case---=2, =3

109

 Wp/Wn=P/N = r = k gives equal rise and fall delay =2; =2; k=1 for inv; k=1/2 for nand2, k=2 for Nor2
=2; =3;
k=2/3 for inv; k=1/3 for nand2, K=4/3 for Nor2

For inv.---Delay RC Delay n (1/Wn,p) (Wp+Wn) dd (1/n) (1/Wn) (Wp+Wn)

du (1/p) (1+r)

Condition for minimum average delay

Av =0 r For all gates

EE141

112

Optimum NMOS / PMOS ratio-rabaey

Smaller device size yields faster design

Cw> Cg chose w large Cw< Cg choose w small

Symmetrical transient response

MINIMUM POSSIBLE DELAY

Computing Intrinsic Transistor Capacitance

Intrinsic PN junction capacitance of the driving circuit must be added to the load capacitance Cload Consider the inverter example at left:
Area and perimeter of the PMOS and NMOS transistors are calculated from the layout and inserted into the circuit model
NMOS drain area = Wn x Ddrain PMOS drain area = Wp x Ddrain NMOS drain perimeter = 2 (Wn + Ddrain) PMOS drain perimeter = 2 (Wp + Ddrain)

SPICE simulations were done (bottom left) for a fixed extrinsic load of 100fF with increasing transistor width (Wp/Wn = 2.75)
Results show diminishing returns beyond a certain Wn (say about 6 um) due to effect of the increasing drain capacitance on the overall capacitive load

MINIMUM DELAY ~ ZERO DELAY

R= Wp/Wn

Non zero value

Area x Delay Figure of Merit

Increasing device width shows diminishing returns on propagation delay time

Define a figure of merit as area x delay for the inverter circuit

Increasing device width Wn shows a minimum in area x delay product

Unconstrained increase in transistor width in order to improve circuit delay is often a poor tradeoff due to the high cost of silicon real estate on the wafer!!

Design a chain of inv. for min delay

T-network Delay Model Of wire

Star-delta-transformation Vout=ZBC/(ZAB+ZBC) Vout=[(2/RC)/(S+2/RC)]*(1/S) =(1/s)-1/(s+2/RC) =U(t)[1-exp(-2/RC)t] FOR V50% delay tp=(RC ln2)/2=0.35 RC

Delay in the presence of long wires

Design Of Inverter Chain

For Min. Delay

Sizing a path for minimum delay

EE141

126

Branching effort along a path

Used for sizing for delay

Where BH is

EE141

127

Observations regarding F
F depends on only topology and loading F is Indep. of transistor sizes F is unchanged if inverters are added or removed

EE141

128

Path Delay D
Sum of delay of all stages

EE141

129

Condition for min. path delay

EE141

130

On Differentiation:

EE141

131

Thus, minimum stage effort of each stage reqd. for min. delay along a path is
We shd. choose transistor sizes such that stage effort is same for all blocks

Thus, minimum delay achievable along a path is

EE141

132

Example
Compute for each stage

Apply capacitance transformation backwards

EE141

133

Chain Of Inverters

In
1 Ci u u2 u N-1

Out

C1

C2 CL

uopt = e

Optimizing no of stages in a path for min. delay

135

To find optimum N

If pinv = 0,

EE141 136

For stages in chain with inverters Best delay per stage , d = gh + pinv d = + pinv

137

Graphical sol
As pinv grows, adding inverters become less advantageous

EE141 138

Chain Of Inverters

BEST NO OF STAGES

In
1 Ci u u2 u N-1

Out

C1

C2 CL

uopt = e

Chain Of Inverters
gu= gav x [2 / (+)] gd= gav x [2 / (+)]

BEST NO OF STAGES

In
1 Ci u u2 u N-1

Out

C1

C2 CL

uopt = e

For large N, delay expression-

For = 4 = log 4F X FO4

141

FO4 DELAY

Where FO4 = fanout of 4 inverter delay HERE = gh = 1 x 4 = 4; so d = 5

Thus for = 4 = log 4F X FO4 inverter delay

142

143

Wrong no of stages

144

EE141 145

Wrong size, L=1

W C=1

4sW C=16
=4/ s

16 W

C=4s = 4s

=4

Mis-sized

D = gh + pinv
= (4s + 4/s + 4 ) + 3 pinv = 15 units (s=1)
146

EE141 147

Power dissipation

Why worry about power? -- Heat Dissipation

microprocessor power dissipation

DEC 21164

Why worry about power Portability

Nominal Capacity (Watt-hours / lb)
50

Rechargable Lithium
40

Ni-Metal Hydride
30 20 10 0 65 70 75 80 85 90 95

Nickel-Cadium

BATTERY (40+ lbs)

Year Multimedia Terminals Laptop Computers Digital Cellular Telephony

Expected Battery Lifetime increase over next 5 years: 30-40%

Where Does Power Go in CMOS?

STATIC POWER---NIL
Dynamic Power Consumption
Charging and Discharging Capacitors

Short Circuit Currents

Short Circuit Path between Supply Rails during Switching

Leakage
Leaking diodes and transistors

Power consumption
4 components Static power consumption Short circuit power consumption Leakage power consumption Dynamic power consumption

 The total power in a CMOS circuit is given by Ptotal = Pd + Psc + Ps where

Pd is the dynamic average power (previous chart), Psc is the short circuit power, and Ps is the static power due to ratio circuit current, junction leakage, and sub-threshold Ioff leakage current

Short circuit current flows during the brief transient when the pull down and pull up devices both conduct at the same time where one (or both) of the devices are in saturation

Static power consumption

Short circuit power

CMOS Short-Circuit Power Dissipation Derivation

Short Circuit Path

Modelling

t1- t2, Mos operates in saturation At t2, current reaches its maximum value At this point vin=vdd/2, because inverter is symmetrical I mean= 2x [2/T] x Isat dt : Limits(t1, t2)

ConditionsVin(t)=(Vdd/) t;
--assume vin increases linearly with time

tr = tf = trf Psc = (/12) (Vdd 2Vt)3 (trf/tpin)

 For a balanced CMOS inverter with n=p= , and Vtn = |Vtp|, the short circuit power can be expressed by Psc = (/12)(Vdd 2Vt)3 (tr/f/tpin)

where tpin is the period of the input waveform and trf is the input rise time (or fall time) tr = tf = trf

Effect of load cap on short circuit power

P short circuit reduces Reason---- output start switching after input has completely stabilized

Effect of Cload

Dynamic energy consumption

Energy stored across capacitor

Dynamic power consumption-derivation

Average Dynamic Power in CMOS Inverter

Average dynamic power derivation:
On negative going input, pull-up device charges the load capacitance. On positive going input, pull-down device discharges the load into ground. Average power given by Pave = (1/T)CL (dvout/dt) (Vdd vout)dt + (1/T)(-1) CL (dvout/dt) vout dt where the first integral is taken from 0 to T/2 and the second integral is from T/2 to T completion of the integral yields

Pave = CL Vdd2 f

where f = 1/T

Note that the dynamic power is independent of the typical device parameters, but is simply a function of power supply, load capacitance and frequency of the switching!

Vdd

Vin CL

Vout

Energy/transition = CL * Vdd2 Power = Energy/transition * f = CL * Vdd 2 * f

Not a function of transistor sizes! Need to reduce C L , Vdd , and f to reduce power.

Reduce power consumption

Reduce Vdd Reduce swing at the output Reduce CL Reduce Switching activity

To keep same speed, can we reduce Vdd, increase (w/L)? No

Inc in W inc in CL

Dynamic Power Consumption - Revisited

Power = Energy/transition * transition rate = CL * Vdd2 * f01 = CL * Vdd2 * P01* f = CEFF * Vdd2 * f

Power Dissipation is Data Dependent Function of Switching Activity

CEFF = Effective Capacitance = CL * P0 1

Power Consumption is Data Dependent

uniform distribution of inputs
Example: Static 2 Input NOR Gate

Assume: P(A=1) = 1/2 P(B=1) = 1/2 Then: P(Out=1) = 1/4 P(0 1) = P(Out=0).P(Out=1) = 3/4 1/4 = 3/16 CEFF = 3/16 * CL

Transition Probabilities for Basic Gates

Non-uniform distribution of inputs

No feedback

Power consumptionCorrelated signals

Sizing for min. power consumption

For a given delay constraint

Sizing for power consumption

Vdd=Vddref

Vdd=Vddref

Vdd Vddref

Graphical solution

Why energy reduces for F increasing?

Assume delay reqd is tpref=5ns. As F inc CL inc. delay (tp) inc. and dyn. energy inc. linearly But as f inc delay reduces exponentially, energy inc. for F= 1 delay is already small and close to tpref). Inc in f does not cause much reduction rather energy increment is more For F large, delay and energy are large values Hence as f inc., delay reduces drastically (become less than tpref ). Hence to have given delay= tpref, energy is dec. which inc. delay to tpref. As f is increased further, delay reduction reduces, only energy increases

Design example0.25um technology, find f, Vdd for tpref=0.2ns. Cext=10Cg1, =1, Vref=2.5v

Design a chain of inv for min delay, min energy

Power delay product

Indicates that energy required 0 for Vdd 0 erroneous

Energy delay product

Shd. Be minimum

Energy delay product

optimum Vdd

Combinational logic
STATIC CMOS GATES

asynchronous design

Can Be Made Synchronous By Inserting Latches in between

Design Styles
Full Static CMOS or complementary logic

NAND

NOR

XOR/ XNOR

DRAWBACK complementary signals are required

F =

+ A. (B+C)

static CMOS gate VTC--Input data dependent

Tphl--Delay computation NAND state of intermediate nodes matter --worst case

Drawback of static cmos

2N devices required Prop delay inc with increase in fanin because of inc in Cint, large series chain

Uniform transistor sizing

For the gate, Find equivalent inverter model Find the required transistor w/L Hence estimate w/L of each transistor

Influence of fan-in / fanout on propagation delay

Other delay reduction techniques

Progressive transistor sizing Input reordering Logic restructuring

Reduce power consumption

Reduce switching activity

Power consumption due to glitches

Power reductionbalanced signal path for glitch reduction

Logic restructuring for lowering switching activity

Power reduction- Input reordering

affects

Power reduction- Time multiplexing of resourcesarea reduces, activity increases

Very low switching activity

Very high switching activity as bus toggles between 0 and 1

Other design styles--Pseudo NMOS

DCVSL

Xor/ Xnor
ADVANTAGE---TRANSISTOR SHARING

DCVSL is advantageous for full adder implementation Then static CMOS

NAND/ AND

Adder

USE OF LOGICAL EFFORT MODEL

GENERAL PATH

223

Transistor sizes

All stages shd have same sizes C = n W L Cox; n is a non zero no. Each stage load = 3 (w l) Cox L=min size
224

Transistor sizes
Inverter load at the input = (2pmos+1nmos) gate load [Wnmos + Wpmos ]Lmin Cox or [Wnmos + 2Wnmos ]Lmin Cox

Here Cz=C = [Wnmos + Wpmos ]Lmin Cox In a given tech., L is fixed, say 1um We take 1Cg= Wmin Lmin Cox If C= 100Cg; then Wpmos= Wnmos = 100 Wmin
225

Ca = Cb = [Wnmos + Wpmos ]Lmin Cox

EE141 226

Driving Large Capacitances -use LE

tpHL = CL Vswing/2
VDD

Iav

Vin

Vout CL

Transistor Sizing

Using Cascaded Buffers

In
1 Ci u u2 u N-1

Out

C1

C2 CL

uopt = e

design
Determine N, u(=) cL=un+1 cg (n+1)=ln (cL/cg) / ln(u) Delay=o (cd+u cg) / (cd+cg) Delay total = (n+1) o [(cd+u cg) / (cd+cg)] Delay total= [ln (cL/cg) / ln(u) ] * o [(cd+u cg) / (cd+cg)] u(ln u-1) = (cd/cg) ~0 U= e

Other logic design styles

Switch logic