IBM Systems and Technology Group DATE 2006 © 2006 IBM Corporation VLSI Design for Yield on Chip Level Markus Bühler Jeanne Bickford Jason Hibbeler Jürgen Koehl
IBM Systems and Technology Group
DATE 2006 © 2006 IBM Corporation
VLSI Design for Yield on Chip Level
Markus BühlerJeanne Bickford Jason Hibbeler Jürgen Koehl
2
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Outline
Catastrophic Failures
Defect Mechanisms
State of the Art
Novel Techniques
Conclusion
3
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Catastrophic Failures
4
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Defect Mechanisms: Random Defects
caused by particles
quantitative prediction by Critical Area Analysis (CAA)given defect size x
calculate area where a defect of size x is catastrophic
sum CA over all shapes of design
weighted integrate over all defect sizes: p(x)
shorts opens
x
CA
5
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Defect Mechanisms: Systematic FailuresLithography issues
Things don't print as drawn
Design AnalysisLitho-SimulationExpert system approach
point to hot spots on designno quantitative yield prediction
Many other issues (e.g. with CMP)
6
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
State of the Art
Driven by short sensitivity on aluminumWire spreading increases net length
higher sensitivity to opens
Wire Spreading Redundant Vias
always a good thingslightly higher sensitivity to shorts
7
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
New Techniques for CopperChanging fail mechanisms
Aluminum: shorts/opens = 3:1 Copper: shorts/opens = 1:1Higher sensitivity to opens
Non-Tree Routing (Kang, 2004) Spreading + Wire Widening
Both approaches have significant timing impact
8
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Local Loops• Not every via can be made redundant
68%77.5%88.5%% of redundant vias
Design 390 nm
Design 2130 nm
Design 1130 nm
• More redundancy: local loops
9
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Advantages of Local LoopsSimple extension of redundant via approachCan be implemented w/o wrong way wiringLess fault correlation due to higher distance
Critical area reduction is higher than with redundant viasMinimal timing impact
0,0
0,2
0,4
0,6
0,8
1,0
1,2
1,4
1,6
Metal Open Via Open Metal Short
Rel
ativ
e C
ritic
al A
rea
Single viaRed viaRed via + LoopLoopLoop + red via
CAA: 5 ExperimentsSingle via onlyRedundant via onlyRedundant via + local loops Local loops onlyLocal loops + redundant vias
10
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Global LoopsKang 2004: augment existing wiring
100%15%100
100%27%20
100%38%10
protectedoverhead*Pins
Traveling Salesman approach*
36%4%50
47%17%10
54%33%5
protectedoverhead*Pins
*P. Panitz, M. Olbrich, IMS, U. Hannover
* compared to minimum Steiner tree * compared to minimum spanning tree
11
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Global Loops cont.EfficiencyAccumulated Netlength over Pincount
0%
20%
40%
60%
80%
100%
120%
1 2 3 4 5 6 7 8 9 10
Sinks
NetsNet Length
Still some research required!
6%11%overhead
16%17%red. wiring
105pin limit
TSMTree links
Timing issuesTSM in generalboth in case of fault
Comparison is not fair!
12
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
CAA on Library ElementsCAA has long runtimesCAA gives global yield number
no hints where problems arelate feedback
limited use for design tools
Critical area library characterizationperform CAA on each library elementcharacterize possible cell interactioncalculate fail probability for each cell
Possible short in adjacent cells
Perimeter failprobability
enables early yield estimationtools guidance
13
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
Feature VariationRelative feature variation increases with shrinking technology- dopant density- line width
Features vary independently- Example: wire width variation causes hardware failure
td=1.9nsTclock= 2ns
1.8
2.2
2.0
2.0
budget
-0.10.81.0fastslow
0.31.00.8slowfast
0.10.80.8fastfast
0.11.01.0slowslow
slackAT2AT1M5M4
Variation Aware TimingStatistical Timing
14
ASIC Design Center Böblingen
VLSI Design for Yield on Chip Level © 2006 IBM Corporation
ConclusionDesign for yield is design for low cost and quality
Traditional techniques are not sufficient
Multiple aspects- redundancy- defect robustness- variation robustness
DfY doesn't come for free- timing issues- wiring congestion increase- noise issues
There is a lot to be gained but also a lot to do!