industry trends compare and contrast harry foster chief verification scientist
DESCRIPTION
Industry Trends Compare and Contrast Harry Foster Chief Verification Scientist. State of the Industry. Statistics. Mindless. There are three types of lies - lies, damn lies, and statistics. -Mark Twain. Myth vs. Reality?. Slowing Adoption of New Technology. 2005 1.72B - PowerPoint PPT PresentationTRANSCRIPT
Industry Trends
Compare and Contrast
Harry FosterChief Verification Scientist
HDF – HVC 2009
State of the Industry
MindlessMindless StatisticsStatistics
There are three types of lies - lies, damn lies, and statistics. -Mark Twain
HDF – HVC 2009
Myth vs. Reality?
HDF – HVC 2009
4
197929,000
Transistors8088
1982134,000
Transistors286
1985275,000
Transistors386
19891,290,000
Transistors486
19933.1M+
TransistorsPentium
19955.5M+
TransistorsPentium Pro
19977.5m+
TransistorsPentium II
19999.5M+
TransistorsPentium III
200042M
TransistorsPentium 4
2004592M
TransistorsItanium 2 (9MB cache)
20051.72B
TransistorsDual Core Itanium
2002220M
TransistorsItanium 2
Slowing Adoption of New Technology
2008 2Billion transistors
Tukwila Quad Core
2008 2Billion transistors
Tukwila Quad Core
HDF – HVC 2009
55
Frequent Statements About the Slowing of Technology Adoption
“The problem is that Moore's Law has collapsed," he says. Coburn asserts that there has been a slowdown in the previously steady move to smaller geometries and larger wafer sizes.
Pip Coburn, Coburn VenturesDecember 15th, 2008
Source: Barron’s “Why it’s going to get a lot worse,” Eric J. Savitz. Dec. 15, 2008http://online.barrons.com/article/SB122912495865802961.html?mod=gartner
“The slowdown in process technology transitions will mean that the semiconductor industry will be driven more by economics than technology …”
“You are not seeing these geometries rise and fall off the way they did before. Rather, they are living on.”
Len Jelinek, Director and chief analyst, Semiconductor Manufacturing, for iSuppli
Source: EE Times,“ISuppli: Gear costs to derail Moore's Law in 2014," Dylan McGrath, June 16, 2009 http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=217900102
“And, the customers have slowed down or delayed their technology transitions either by leveraging their existing installed base or just by delaying their new product introduction for later.”
Eric Meurice, Chairman,
President and CEO ASML Holding N.V.Source: Q4 2008 Earnings Call. January 15, 2009http://seekingalpha.com/article/115001-asml-holding-n-v-q4-2008-earnings-call-transcript?page=-1
HDF – HVC 2009
6
Design Completion TrendsDesign Completion Share by Linewidth
0%
5%
10%
15%
20%
25%
30%
35%
40%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Per
cen
t o
f T
ota
l D
esig
n C
om
ple
tio
ns
350nm 250nm 180nm 130nm 90nm 65nm 45nm
Source: VLSI Research, Design Completions, September 2008
6
HDF – HVC 2009
7
Volume of Wafer Starts by Linewidth (300mm wafer equivalents)
0%
5%
10%
15%
20%
25%
30%
35%
40%
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Per
cen
t o
f T
ota
l S
ilic
on
Dem
and
350nm 250nm 180nm 130nm 90nm 65nm 45nm
Source: VLSI Research, Silicon Demand, EDA Tech Forum June, 2009
Silicon Volume of Wafer Starts(in 300mm Wafer Equivalents)
7
HDF – HVC 2009
8
Reticle Sales TrendsReticle Revenue Share by Linewidth
0%
5%
10%
15%
20%
25%
30%
35%
40%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Per
cen
t o
f T
ota
l R
etic
le R
even
ue
Sh
are
350nm 250nm 180nm 130nm 90nm 65nm 45nm
Source: VLSI Research, Reticles, September 2008
8
HDF – HVC 2009
9
Waffer in ProductionIC Fab Capacity by Node *
IC Fab Capacity by Node (MSI) January 2004 -to- June 2008
0
50
100
150
200
Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul-08
IC F
ab C
apac
ity
(MS
I)
45nm 65nm 90nm 130nm 180nmSource: VLSI Research, December 15, 2008
* Note: Realized
9
HDF – HVC 2009
10
Capacity Utilization Improvement Favors 65/45nm Technology
30%
40%
50%
60%
70%
80%
90%
100%
Jan-0
8
Feb-08
Mar
-08
Apr-08
May
-08
Jun-0
8
Jul-0
8
Aug-08
Sep-0
8
Oct
-08
Nov-08
Dec-0
8
Jan-0
9
Feb-09
Mar
-09
Apr-09
Uti
liza
tio
n %
180nm
130nm
90nm
65nm
45nm
Source: Selantek Capacity Analysis, May, 2009, and October 2009
90nm,19.0%130nm,
8.6%
65nm, 14.7%
150nm, 1.8%
40/45nm, 4.9%
180nm, 7.1%
>180nm, 27.1%
Capacity by Node - April
HDF – HVC 2009
11
197929,000
Transistors8088
1982134,000
Transistors286
1985275,000
Transistors386
19891,290,000
Transistors486
19933.1M+
TransistorsPentium
19955.5M+
TransistorsPentium Pro
19977.5m+
TransistorsPentium II
19999.5M+
TransistorsPentium III
200042M
TransistorsPentium 4
2004592M
TransistorsItanium 2 (9MB cache)
20051.72B
TransistorsDual Core Itanium
2002220M
TransistorsItanium 2
Adoption of Leading-Edge Semiconductor Technology Is at the Same Rate as in the Past
Slowing Adoption of New Technology
2008 2Billion transistors
Tukwila Quad Core
2008 2Billion transistors
Tukwila Quad Core
HDF – HVC 2009
Forecast
60% PCs & Cellphones
Huge growth in smartphones
2013 1/3 cells a smartphones
62.8 64.768.9
52.243.7
52.4
62.9 67.3 65.969.2 72.6 75.1
0
10
20
30
40
5060
70
80
$ B
illio
ns
1Q08 3Q08 1Q09 3Q09 1Q10 3Q10Quarter
World Semiconductor Market
Com
putin
g
Com
munication
Digital Video
ConsumerElectronics
HDF – HVC 2009
SoC Designs Dominate
HDF – HVC 2009
SOC Design & VerificationEmbedded Microprocessor Cores Trend
35%
14%
3%
11%
48%
7%
39%
6%
32%
7%
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
NONE 1 2 3 4 5 or MORE
2004 2007
HDF – HVC 2009
SOC Design & VerificationEmbedded DSP Cores
20%
9%
2% 2% 3%
65%
0%
10%
20%
30%
40%
50%
60%
70%
NONE 1 2 3 4 5 or MORE
HDF – HVC 2009
What Does 45-nm Mean To Us?Challenges and Opportunities
Active and standby leakage
accounts for 65% of overall
power consumption at 45-nm
Low-power techniques are
necessary at 45-nm
45-nm offers 2X reduction in
die size or 2X increase in gate
count over 65-nm300 mm2 45-nm wafer
45-nm Wolfdale65-nm Conroe
HDF – HVC 2009
17
Myth vs. Reality?Rising Design Costs Will Limit New Applications
HDF – HVC 2009
1818
Source: Technology Research Group – EDA Database, 1986, EDA TAM, 1989 & Gartner/Dataquest 2005 Seat Count Report, Gary Smith EDA, 2008 Seat Count Analysis VLSI Research, 2008 - Transistors Produced Analysis
Transistors Produced per Electrical Engineer Nearly 4-Orders of Magnitude since 1985
HDF – HVC 2009
19
EDA Revenue Is Flat 2% of IC Revenue
EDA License & Maintenance/IC Revenue (% Percent)
0.0%
0.5%
1.0%
1.5%
2.0%
2.5%
3.0%
3.5%
4.0%
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Source: Mentor Graphics, EDAC MSS & SIA WSTS
HDF – HVC 2009
2020
EDA Cost per Transistor vs Total IC Revenue per Transistor
Source: SIA, VLSI Research, Federal Reserve Note: EDA Cost Consists of EDA License and Maintenance revenue adjusted for Inflation… 1985 - 2007
1.00E-09
1.00E-08
1.00E-07
1.00E-06
1.00E-05
1.00E-04
1.00E+13 1.00E+14 1.00E+15 1.00E+16 1.00E+17 1.00E+18 1.00E+19 1.00E+20
ED
A C
os
t/T
ran
sis
tor
($)
IC R
ev
enu
e/T
ran
sis
tor
($)
1.00E-09
1.00E-08
1.00E-07
1.00E-06
1.00E-05
1.00E-04
1.00E+13 1.00E+14 1.00E+15 1.00E+16 1.00E+17 1.00E+18 1.00E+19 1.00E+20
AT
E C
ap
ita
l Co
st/
Tra
ns
isto
r ($
)
HDF – HVC 2009
SOC Design Costs Forecasted to Exceed $100 Million Within 3 Years
Impact of Design Technology on SOC Consumer Portable Implementation Cost
0
20
40
60
80
100
120
2007e 2008f 2009f 2010f 2011f 2012f
$ U
S M
illio
ns
Hardware Costs Software Costs1 2
1 Total Hardware Engineering Costs + EDA Tool Costs2 Total Software Engineering Costs + Electronic Software Design Tool CostsSource: 2007 ITRS Roadmap
Notes:
21
HDF – HVC 2009
Software Developers Outnumber Hardware Designers 2-to-1
Source: VDC - Embedded Systems Market Statistics 2007
(in 000’s)
22
HDF – HVC 2009
System Design Has Shifted to the Semiconductor Suppliers
Buses
CPU MEMHW
SWDriver
HW
SW Driver
HW
SWDriver
HW
SW Driver
OS
Middleware
Service Abstraction
Apps Apps Apps Apps AppsMuch of what was part of the end-system is now incorporated within a System-On-Chip
23
HDF – HVC 2009
24
Semiconductor Companies Are Assuming More of the System and Embedded Software
Engineering Design Responsibility
Rising Design Costs Limit New Applications
HDF – HVC 2009
Myth vs. Reality?Verification Is Keeping Up With Moore’s Law
HDF – HVC 2009
An Optimistic View of the Productivity Gap
Productivity Gap
Verify
Design
Manufacture
Time (years)
Siz
e (
# t
ran
sis
tors
)
Let’s assume…
Number of transistor doubles every 18 months (58% / yr)
Amount of logic we can design doubles every 2 years (41% / yr)
Amount of logic we can verify doubles every 2.5 years (25% per year)
HDF – HVC 2009
0
20
40
60
80
1988 1992 1996 2000 2005Ability to Verify
Ability to Design
Ability to Fabricate
DesignGap
VerificationGap
De
sig
n S
ize
in
Mil
lio
ns
of
Ga
tes
27
* Based on data from the 2003 ITRS, Collett International 2004 FV Survey, and customer surveys
The Verification Gap— Many companies still using 1990’s verification technologies
— Traditional verification techniques can’t keep up
Productivity Gap
HDF – HVC 2009
Imagine verifying a car using a directed-test approach— Requirement: Fuse will not blow under any normal operation
— Scenario 1: accelerate to 37 mph, pop in the new
Lady GaGa CD, and turn on the windshield wipers
The Verification Gap Directed Test
State-of-the-Art Verification Circa 1990
HDF – HVC 2009
Imagine verifying a car using a directed-test approach— Requirement: Fuse will not blow under any normal operation
— Scenario 1: accelerate to 37 mph, pop in the new
John Mayer CD, and turn on the windshield wipers
The Verification Gap Directed Test
State-of-the-Art Verification Circa 1990
A Few Weeks Later. . . . .
HDF – HVC 2009
Imagine verifying a car using a directed-test approach— Requirement: Fuse will not blow under any normal operation— Scenario 714: accelerate to 48 mph, roll down the window, and
turn on the left-turn signal
The Verification Gap Directed Test
State-of-the-Art Verification Circa 1990
HDF – HVC 2009
A purely directed-test methodology does not scale— Imagine writing a directed test for this scenario!
— Truly heroic effort—but not practical
Concurrency Challenge
HDF – HVC 2009
Today’s Concurrency Challenge
Packet-Based Design
TLP
DLLP
RetryMemory
Arbiter
Tx
Rx
FromFabric
ToPHY
FromRX
HDF – HVC 2009
What Are We Doing To Close The Gap?
Productivity Gap
Verify
Design
Manufacture
Time (years)
Siz
e (
# t
ran
sis
tors
)
HDF – HVC 2009
34
197929,000
Transistors8088
1982134,000
Transistors286
1985275,000
Transistors386
19891,290,000
Transistors486
19933.1M+
TransistorsPentium
19955.5M+
TransistorsPentium Pro
19977.5m+
TransistorsPentium II
19999.5M+
TransistorsPentium III
200042M
TransistorsPentium 4
2004592M
TransistorsItanium 2 (9MB cache)
20051.72B
TransistorsDual Core Itanium
2002220M
TransistorsItanium 2
Moore with Less
Verification for Every Man, Woman, and Child in IndiaVerification for Every Man, Woman, and Child in India
HDF – HVC 2009
Myth vs. Reality?70% of the project effort is spent in verification….
HDF – HVC 2009
36
Design Engineers Are Becoming Verification Engineers
Source: 2007 Farwest Research IC/ASIC Functional Verification Study, Used with Permission
Verification 35%
Other14%
Design51%
Verification 46%
Design54%
HDF – HVC 2009
37
More and More Verification Cycles
HDF – HVC 2009
38
Faster Computers
0.01
0.10
1.00
10.00
100.00
1000.00
10000.00
1975 1980 1985 1990 1995 2000 2005
MIP
s
HDF – HVC 2009
39
Lots of Computers
Year # of Servers
1996 50
2006 5000+ (over 10,000 CPUs)
00
55
1010
1515
2020
2525
20012001 20022002 20032003 20042004 20052005 20062006
AMD GridAMD Grid Growth 2001-2006
(Relative to 2001 = 1.0)
Source: The AMD Grid: Enabling Grid Computing for the Corporation, August 2006
HDF – HVC 2009
So, with all this effort, what’s the results?
HDF – HVC 2009
Source: 2007 Far West Research and Mentor Graphics
41
Results2/3 Projects Miss Schedule
Designs completed on time according to project's original schedule
0.0
5.0
10.0
15.0
20.0
25.0
30.0
> +10% +10% 0 -10% -20% -30% -40% -50% > -50%
> +10%
+10%
0
-10%
-20%
-30%
-40%
-50%
> -50%
HDF – HVC 2009
6%8%
6%
1% 1% 2%
39%
17%
38%33%
20%
39%
28%
21%
42%
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
1 2 3 4 5 6 >= 7
2002 2004 2007
First Silicon Success Spins
Des
ign
s
Results77% Respins Due to Functional Bugs
Source: Collett International 2002, 2004,Farwest Research 2007 Functional Verification Study
HDF – HVC 2009
ResultsTypes of Flaws
22%27% 25%
13%
27%33%
26%
19%
5%
32%
14%
75%
11% 7%15%
11%7%
21%20%
19% 17%23%24%
77%
0%
20%
40%
60%
80%
100%
LOGIC ORFUNCTIONAL
CLOCKING TUNINGANALOGCIRCUIT
CROSSTALK-INDUCEDDELAYS,
GLITCHES
POWERCONSUMPTION
MIXED-SIGNALINTERFACE
YIELD ORRELIABILITY
TIMING – PATHTOO SLOW
FIRMWARE TIMING – PATHTOO FAST,
RACECONDITION
IR DROPS OTHER
2004 2007
Source: Collett International 2002, 2004,Farwest Research 2007 Functional Verification Study
HDF – HVC 2009
ResultsCauses of Functional Flaws
Source: 2007 Far West Research and Mentor Graphics
35%
41%
60%
18%15%
3%
0%
10%
20%
30%
40%
50%
60%
70%
INCORRECT orINCOMPLETE
SPECIFICATION
CHANGES INSPECIFICATION
DESIGN ERROR FLAW ININTERNALREUSED IP
FLAW INEXTERNAL IP
OTHER
HDF – HVC 2009
Myth vs. Reality?The industry is evolving its verification capabilities?
HDF – HVC 2009
0
20
40
60
80
1988 1992 1996 2000 2005Ability to Verify
Ability to Design
Ability to Fabricate
DesignGap
VerificationGap
De
sig
n S
ize
in
Mil
lio
ns
of
Ga
tes
46
* Based on data from the 2003 ITRS, Collett International 2004 FV Survey, and customer surveys
The Verification Gap— Many companies still using 1990’s verification technologies
— Traditional verification techniques can’t keep up
Productivity Gap
Ability to Adopt?
HDF – HVC 2009
Dynamic Techniques Usage
Source: 2007 Far West Research and Mentor Graphics
7%
9%
10%
10%
10%
21%
23%
24%
25%
26%
27%
33%
36%
37%
40%
41%
48%
51%
67%
0% 20% 40% 60% 80% 100%
PROTOTYPING WITH SPECIAL TEST CHIPS
RF SIMULATION
EMULATION (CUSTOM BUILT SYSTEMS)
ACCELERATED SIMULATION
EMULATION (COMMERCIAL SYSTEMS)
EMBEDDED CHECKERS TO TRAP ILLEGAL CONDITIONS IN THE DESIGN
TRANSACTION-LEVEL SIMULATION
TRANSISTOR-LEVEL SIMULATION
SYSTEM C SIMULATION
HARDWARE/SOFTWARE CO-VERIFICATION
ANALOG/MIXED-SIGNAL SIMULATION
C/C++ SIMULATION
FUNCTIONAL SIMULATION ABOVE RTL LEVEL
ASSERTIONS
FUNCTIONAL COVERAGE
FPGA PROTOTYPING
FUNCTIONAL SIMULATION AT GATE LEVEL
TIMING SIMULATION AT GATE LEVEL
FUNCTIONAL SIMULATION AT RTL LEVEL
HDF – HVC 2009
Static Verification Techniques Usage
Source: 2007 Far West Research and Mentor Graphics
1%
19%
35%
40%
43%
44%
48%
57%
62%
83%
0% 20% 40% 60% 80% 100%
Other
FORMAL PROPERTY CHECKING
DESIGN FOR VERIFICATION TECHNIQUES
EQUIVALENCE CHECKING ABOVE GATELEVEL
LINT CHECK
CODING GUIDELINES THAT AREENFORCED
CODE COVERAGE ANALYSIS
CODE REVIEWS
EQUIVALENCE CHECKING AT GATELEVEL
STATIC TIMING ANALYSIS
HDF – HVC 2009
49
Stop, time to recap. . . .
HDF – HVC 2009
Recap
1. Industry fails to mature its processes
2. Concurrency is difficult to verify
3. Throw lots of bodies at the problem
4. Throw lots of computers at it too
5. All this . . . and poor results
HDF – HVC 2009 WCR February, 2008
51
Verification Challenges Keep Coming
Low power
Clock domain crossing
Hardware/software
Network-on-chip
Multi-level verification
Multi-core verification
System verification
IP reuse (black box)
Mixed-signal (RF/analog/digital)
HDF – HVC 2009
Myth vs. Reality?The biggest bottleneck in the flow is simulation performance?
HDF – HVC 2009
Debugging is the Bottleneck
Effort Allocation of Dedicated Verification Engineers by Type of Activity
52%
34%
14%
Verif ication Debug Testbench Development Other
HDF – HVC 2009
Hey. . . how about some possibilities and prescription!
HDF – HVC 2009
SoC Designs DominateCoverage Confusion
Coverage — How big of a deal is it?
HDF – HVC 2009
Coverage ConfusionHow big of deal is it?
Not A Problem!
Not A Problem!
Coverage Adoption
48%40%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
CODE COVERAGE FUNCTIONALCOVERAGE
ConfusedConfused Major Deal!
Major Deal!
HDF – HVC 2009
Challenges Users Want Solved3 of the top 6 changes are related to coverage!
38%
18%
15%
10%
9%
9%
2%
0% 5% 10% 15% 20% 25% 30% 35% 40%
CLOSING COVERAGE
MANAGING COVERAGEDATA
MANAGING THEVERIFICATION PROCESS
ISOLATING BUGS
TIME TO DISCOVER THENEXT BUG
CREATING FUNCTIONALCOVERAGE MODEL
Other
HDF – HVC 2009
Summary
Today I shared some common themes and insight of
what we are seeing and hearing in the industry
HDF – HVC 2009 WCR February, 2008
59