Meta-Model based High-Dimensional Yield Analysis using...
Transcript of Meta-Model based High-Dimensional Yield Analysis using...
![Page 1: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/1.jpg)
Meta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation
Xiao Shi1, Hao Yan2, Qiancun Huang2, Jiajia Zhang2, Longxing Shi2, Lei He1
June 5, 2019
1 Electrical and Computer Engineering Department, UCLA2 Southeast University, China
![Page 2: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/2.jpg)
Outline
⚫ Preliminary of High Sigma Analysis and Existing Approaches
⚫ The Proposed Approach
⚫ Experiment Results
⚫ Summary
![Page 3: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/3.jpg)
Statistical Circuit Simulation
⚫ Process Variation First mentioned by William Shockley in his analysis of P-N junction breakdown[S61] in 1961
Revisited in 2000s for long channel devices[JSSC03, JSSC05]
Getting more attention at sub-100nm[IBM07, INTEL08]
⚫ Sources of Process Variation
⚫ Statistical Circuit Simulation helps to debug circuits in the pre-silicon phase to improve yield rate
[S61] Shockley, W., “Problems related to p-n junctions in silicon.” Solid-State Electronics, Volume 2, January 1961, pp. 35–67.
[JSSC03] Drennan, P. G., and C. C. McAndrew. “Understanding MOSFET Mismatch for Analog Design.” IEEE Journal of Solid-State Circuits 38, no. 3 (March 2003): 450–56.
[JSSC05] Kinget, P. R. “Device Mismatch and Tradeoffs in the Design of Analog Circuits.” IEEE Journal of Solid-State Circuits 40, no. 6 (June 2005): 1212–24.
[IBM07] Agarwal, Kanak, and Sani Nassif. "Characterizing process variation in nanometer CMOS." Proceedings of the 44th annual Design Automation Conference. ACM, 2007.
[Intel08] Kuhn, K., Kenyon, C., Kornfeld, A., Liu, M., Maheshwari, A., Shih, W. K., ... & Zawadzki, K. (2008). Managing Process Variation in Intel's 45nm CMOS Technology. Intel Technology Journal, 12(2).
![Page 4: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/4.jpg)
High Sigma Analysis
⚫ High sigma (rare event) tail is difficult to achieve with Monte Carlo method # of simulations required to capture 100 failure samples
⚫ High sigma analysis is critical for highly-duplicated circuits Memory cells (up to 4-6 sigma), IO and analog circuits (3-4 sigma) 1
⚫ How to efficiently and accurately estimate Pfail(yield rate) on high sigma tail?
1 Cite from SolidoDesign Automation whitepaper
![Page 5: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/5.jpg)
Existing Methods and Limitations
⚫ Draw more samples in the tail
⚫ Importance Sampling[DAC06]
Shift the sample distribution to more “important” region
⚫ Classification based methods[TCAD09]
Filter out unlikely-to-fail samples using classifier
⚫ Markov Chain Monte Carlo[ICCAD14]
Explore the failure region with a set of sample chains
[DAC06] R. Kanj, R. Joshi, and S. Nassif. “Mixture Importance Sampling and Its Application to the Analysis of SRAM Designs in the Presence of Rare Failure Events.” DAC, 2006
[TCAD09] Singhee, A., and R. Rutenbar. “Statistical Blockade: Very Fast Statistical Simulation and Modeling of Rare Circuit Events and Its Application to Memory Design.” TCAD, 2009
[ICCAD14] Sun, Shupeng, and Xin Li. “Fast Statistical Analysis of Rare Circuit Failure Events via Subset Simulation in High-Dimensional Variation Space.” ICCAD 2014
![Page 6: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/6.jpg)
Main Challenges – High Dimensionality
⚫ IS can be numerical instable at high dimension Curse of dimensionality[Berkeley08, Stanford09]
Computational complexity increases exponentially with circuit dimension
⚫ Classification based approaches Training classifiers is time-consuming and may perform
poorly at high dimension
⚫ Markov Chain Monte Carlo[ICCAD14]
It is difficult to cover multiple failure regions using a few chain of samples
[Berkeley08] Bengtsson, T., P. Bickel, and B. Li. “Curse-of-Dimensionality Revisited: Collapse of the Particle Filter in Very Large Scale Systems.” Probability and Statistics: Essays in Honor of David A. Freedman2 (2008)
[Stanford09] Rubinstein, R.Y., and P.W. Glynn. “How to Deal with the Curse of Dimensionality of Likelihood Ratios in Monte Carlo Simulation.” Stochastic Models25, no. 4 (2009): 547–68.
[ICCAD14] Sun, Shupeng, and Xin Li. “Fast Statistical Analysis of Rare Circuit Failure Events via Subset Simulation in High-Dimensional Variation Space.” ICCAD 2014
![Page 7: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/7.jpg)
Outline
⚫ Preliminary of High Sigma Analysis and Existing Approaches
⚫ The Proposed Approach
⚫ Experiment Results
⚫ Summary
![Page 8: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/8.jpg)
Meta-Modeling for Yield Analysis
Input variables Output circuit spec
⚫ Construct a meta-model to substitute expensive SPICE simulation
![Page 9: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/9.jpg)
Meta-Modeling for Yield Analysis
Input variables Output circuit spec
⚫ Construct a meta-model to substitute expensive SPICE simulation
⚫ Polynomial Chaos Expansion (PCE)
Our solution:
Low Rank Tensor Approximation
![Page 10: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/10.jpg)
Low Rank Tensor Approximation Formulation
⚫ Rank-one tensor subset
⚫ LRTA meta-model in canonical form
truncate this canonical decomposition expression with small R and sufficient accuracy
⚫ Polynomial Representation
Linearly increase with dimension d!
![Page 11: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/11.jpg)
Solve for LRTA
⚫ Greedy Implementation
In each iteration, we perform alternating least-squares (ALS) to calculate the polynomial coefficients
Correction step
Sequentially minimize the approximation residual.
Normalization step
Update all previous normalization constant b for rank-one tensor component
![Page 12: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/12.jpg)
Solve for LRTA
⚫ Sparse Constraint The majority of circuit variables have only weak influence on performance. {zl} has many zero
elements or elements with extremely small magnitude
Generalized LASSO problem
⚫ Parameter Tuning Perform 3-folded cross validation to choose best tensor rank and maximum degree of polynomial
expansion.
![Page 13: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/13.jpg)
Adaptive Sampling Strategy
![Page 14: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/14.jpg)
Outline
⚫ Preliminary of High Sigma Analysis and Existing Approaches
⚫ The Proposed Approach
⚫ Experiment Results
⚫ Summary
![Page 15: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/15.jpg)
Experiments on 6T SRAM Bit Cell
About 6-9X speedup
over other methods
![Page 16: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/16.jpg)
Experiments on SRAM Column Circuit
About 2150X speedup
MC methods
![Page 17: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/17.jpg)
Computational Complexity vs. Dimensionality
⚫ Vary the number of bit cells in SRAM column
⚫ Simulation cost of LRTA grows only linearly with the dimension
![Page 18: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/18.jpg)
Summary
⚫ Develop a LRTA meta-model to substitute expensive SPICE simulation
Alternatively minimizes approximation residual and solve for polynomial coefficients
Adaptive sampling strategy to enrich sample set
⚫ Better accuracy and efficiency
6-9X faster than other existing methods
Complexity grows linearly with circuit dimension
![Page 19: Meta-Model based High-Dimensional Yield Analysis using ...eda.ee.ucla.edu/pub/C174slides.pdfMeta-Model based High-Dimensional Yield Analysis using Low-Rank Tensor Approximation Xiao](https://reader034.fdocuments.us/reader034/viewer/2022051603/5ff272643ead30598f57d3e7/html5/thumbnails/19.jpg)
Q&A
Thank you for your attention!