EE141
1
EE141 1
Design Metrics
EE141- Spring 2003Lecture 2
EE141 2
Last Lecture
� Moore’s law� Challenges in digital IC design in the
next decade.
Today� Design metrics
EE141
2
EE141 3
Administrivia
� Everyone should have a UNIX account on Cory! Thiswill allow you to run HSPICE!» If not, check
http://www-inst.eecs.berkeley.edu/usr/pub/jpg/How-to-Get-Named-Acct.jpg
� PC accounts for 353 Cory will be created early nextweek (when the classlist is completed)
� Discussions start in Week 2� Labs start in Week 3!� If you have not signed-in on the class roster, please
do so after the lecture.
EE141 4
The EE141 Week at a Glance
M
Tu
W
Th
F
8 9 10 11 12 1 2 3 4 5 6
Lab(TBD)353 Cory
Lab(TBD)353 Cory
Lab (TBD) 353 Cory
DISC*(Dejan)293 Cory
DISC*(Huifang)
203McLaughlin
Lec(Jan)
203 McLaughlin
<- ProblemSets Due
OH(Jan) 511Cory
Lec(Jan)
203 McLaughlin
* Discussion sections will cover identical material
EE141
3
EE141 5
Introduction (wrap-up)
� Why is designing digitalICs different today thanit was before?
� Will it change in future?
EE141 6
Frequency
P6Pentium ® proc
48638628680868085
8080800840040.1
1
10
100
1000
10000
1970 1980 1990 2000 2010Year
Fre
qu
ency
(Mh
z)
Lead Microprocessors frequency doubles every 2 years
Doubles every2 years
S. Borkar
EE141
4
EE141 7
Power
P6Pentium ® proc
486
3862868086
80858080
80084004
0.1
1
10
100
1971 1974 1978 1985 1992 2000Year
Po
wer
(Wat
ts)
Lead Microprocessors power continues to increase
S. Borkar
EE141 8
Power will be a problem
5KW18KW
1.5KW500W
40048008
80808085
8086286
386486
Pentium® proc
0.1
1
10
100
1000
10000
100000
1971 1974 1978 1985 1992 2000 2004 2008Year
Po
wer
(Wat
ts)
Power delivery and dissipation will be prohibitive
S. Borkar
EE141
5
EE141 9
Power density will increase
400480088080
8085
8086
286 386486
Pentium® procP6
1
10
100
1000
10000
1970 1980 1990 2000 2010Year
Po
wer
Den
sity
(W/c
m2)
Hot Plate
NuclearReactor
RocketNozzle
Power density too high to keep junctions at low temp
S. Borkar
EE141 10
Power delivery challenges
P6Pentium® proc
486386286
8086
80858080
800840040.01
0.10
1.00
10.00
100.00
1,000.00
1970 1980 1990 2000 2010Year
Icc
(am
p)
P6Pentium® proc
486386
286
8086
80858080
80084004
1.E-041.E-031.E-021.E-011.E+001.E+011.E+021.E+031.E+041.E+051.E+061.E+07
1970 1980 1990 2000 2010Year
L(d
i/dt)
/Vd
d
High supply currents at low voltage:Challenges: IR drop and L(di/dt) noise
S. Borkar
EE141
6
EE141 11
Productivity Trends
1
10
100
1,000
10,000
100,000
1,000,000
10,000,000
2003
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2005
2007
2009
10
100
1,000
10,000
100,000
1,000,000
10,000,000
100,000,000
Logic Tr./ChipTr./Staff Month.
xxx
xxx
x
21%/Yr. compoundProductivity growth rate
x
58%/Yr. compoundedComplexity growth rate
10,000
1,000
100
10
1
0.1
0.01
0.001
Lo
gic
Tra
nsi
sto
rp
erC
hip
(M)
0.01
0.1
1
10
100
1,000
10,000
100,000
Pro
du
ctiv
ity
(K)
Tra
ns.
/Sta
ff-M
o.
Source: Sematech
Complexity outpaces design productivity
Co
mp
lexi
ty
EE141 12
Design Abstraction Levels
n+n+S
GD
+
DEVICE
CIRCUIT
GATE
MODULE
SYSTEM
EE141
7
EE141 13
Challenges in Digital Design
“Microscopic Problems”• Ultra-high speed design• Interconnect• Noise, Crosstalk• Reliability, Manufacturability• Power Dissipation• Clock distribution.
Everything Looks a Little Different
“Macroscopic Issues”• Time-to-Market• Millions of Gates• High-Level Abstractions• Reuse & IP: Portability• Predictability• etc.
…and There’s a Lot of Them!
∝∝∝∝ DSM ∝∝∝∝ 1/DSM
?
EE141 14
Why Scaling?
� Technology shrinks by 0.7/generation� With every generation can integrate 2x more
functions per chip; chip cost does not increasesignificantly
� Cost of a function decreases by 2x� How to design chips with more and more functions?� Design engineering population does not double every
two years…� Need to understand different levels of abstraction
EE141
8
EE141 15
Design Metrics
� How to evaluateperformance of a digitalcircuit (gate, block, …)?» Cost» Reliability» Scalability» Speed (delay, operating
frequency)» Power dissipation» Energy to perform a
function
EE141 16
Cost of Integrated Circuits
� NRE (non-recurrent engineering) costs» design time and effort, mask generation
» one-time cost factor
� Recurrent costs» silicon processing, packaging, test
» proportional to volume
» proportional to chip area
EE141
9
EE141 17
NRE Cost is Increasing
EE141 18
Die Cost
Single die
Wafer
From http://www.amd.com
Going up to 12” (30cm)
EE141
10
EE141 19
Cost per Transistor
0.00000010.0000001
0.0000010.000001
0.000010.00001
0.00010.0001
0.0010.001
0.010.01
0.10.111
19821982 19851985 19881988 19911991 19941994 19971997 20002000 20032003 20062006 20092009 20122012
cost:cost:¢¢--perper--transistortransistor
Fabrication capital cost per transistor (Moore’s law)
EE141 20
Yield
%100per waferchipsofnumberTotal
per waferchipsgoodofNo. ×=Y
yieldDieper waferDies
costWafercostDie
×=
( )areadie2
diameterwafer
areadie
diameter/2waferper waferDies
2
××π−×π=
EE141
11
EE141 21
Defects
α−
α×+= areadieareaunitperdefects
1yielddie
α is approximately 3
4area)(diecostdie f=
EE141 22
Some Examples (1994)
$4179%402961.5$15000.803Pentium
$27213%482561.6$17000.703Super Sparc
$14919%532341.2$15000.703DEC Alpha
$7327%661961.0$13000.803HP PA 7100
$5328%1151211.3$17000.804Power PC601
$1254%181811.0$12000.803486 DX2
$471%360431.0$9000.902386DX
Diecost
YieldDies/wafer
Areamm2
Def./cm2
Wafercost
Linewidth
Metallayers
Chip
EE141
12
EE141 23
Reliability―Noise in Digital Integrated Circuits
i(t)
Inductive coupling Capacitive coupling Power and groundnoise
v(t) VDD
EE141 24
DC OperationVoltage Transfer Characteristic
V(x)
V(y)
VOH
VOL
VM
VOHVOL
fV(y)=V(x)
Switching Threshold
Nominal Voltage Levels
VOH = f(VOL)VOL = f(VOH)VM = f(VM)
EE141
13
EE141 25
Mapping between analog and digitalsignals
V IL V IH Vin
Slope = -1
Slope = -1
V OL
V OH
Vout
“ 0” VOL
VIL
VIH
VOH
UndefinedRegion
“ 1”
EE141 26
Definition of Noise Margins
Noise margin high
Noise margin low
VIH
VIL
UndefinedRegion
"1"
"0"
VOH
VOL
NMH
NML
Gate Output Gate Input
EE141
14
EE141 27
Noise Budget
� Allocates gross noise margin toexpected sources of noise
� Sources: supply noise, cross talk,interference, offset
� Differentiate between fixed andproportional noise sources
EE141 28
Key Reliability Properties
� Absolute noise margin values are deceptive» a floating node is more easily disturbed than a
node driven by a low impedance (in terms ofvoltage)
� Noise immunity is the more important metric –the capability to suppress noise sources
� Key metrics: Noise transfer functions, Outputimpedance of the driver and input impedance of thereceiver;
EE141
15
EE141 29
Regenerative Property
v0
v1
v3
finv(v)
f(v)
v3
out
v2 inRegenerative Non-Regenerativev2
v1
f(v)
finv(v)
v3
out
v0 in
EE141 30
Regenerative Property
A chain of inverters
v0 v1 v2 v3 v4 v5 v6
2
V (
Vol
t)
4
v0
v1v2
t (nsec)0
�1
1
3
5
6 8 10Simulated response
EE141
16
EE141 31
Fan-in and Fan-out
N
Fan-out N Fan-in M
M
EE141 32
The Ideal Gate
Ri = ∞Ro = 0Fanout = ∞NMH = NML = VDD/2g = ∞
V in
V out
EE141
17
EE141 33
An Old-time Inverter
NMH
Vin (V)
V
o ut
(V)
NML
VM
0.0
1.0
2.0
3.0
4.0
5.0
1.0 2.0 3.0 4.0 5.0
EE141 34
Delay Definitions
Vout
tf
tpHL tpLH
tr
t
Vin
t
90%
10%
50%
50%
EE141
18
EE141 35
Ring Oscillator
v0 v1 v5
v1 v2v0 v3 v4 v5
T = 2 × tp × N
EE141 36
A First-Order RC Network
vout
vin C
R
tp = ln (2) τ = 0.69 RC
Important model – matches delay of inverter
EE141
19
EE141 37
Power Dissipation
Instantaneous power:p(t) = v(t)i(t) = Vsupplyi(t)
Peak power:Ppeak = Vsupplyipeak
Average power:
( )∫ ∫+ +
==Tt
t
Tt
t supplysupply
ave dttiT
Vdttp
TP )(
1
EE141 38
Energy and Energy-Delay
Power-Delay Product (PDP) =
E = Energy per operation = Pav × tp
Energy-Delay Product (EDP) =
quality metric of gate = E × tp
EE141
20
EE141 39
A First-Order RC Network
E0 1→ P t( )dt0
T
∫ Vdd isupply t( )dt0
T
∫ Vdd CLdVout0
Vdd
∫ CL Vdd• 2= = = =
Ecap Pcap t( )dt0
T
∫ Vouticap t( )dt0
T
∫ CLVoutdVout0
Vdd
∫12---C
LVdd• 2
= = = =
vout
vin CL
R
EE141 40
Next Lecture
� Manufacturing Technology� A First Glance at an Inverter