1. On Diagnosis of Multiple Faults Using Compacted Responses Jing Ye 1,2, Yu Hu 1, and Xiaowei Li 1 1 Key Laboratory of Computer System and Architecture Institute of Computing Technology Chinese Academy of Sciences 2 Graduate University of Chinese Academy of Sciences

2. 2 Motivation With the exponential growth in the number of transistors: Multiple faults may exist in one circuit. Both of test data volume and test application time increase. SOURCE: L.-T. Wang, C.-W. Wu, and X. Wen, “VLSI Test Principles and Architectures: Design for Testability,” Morgan Kaufmann, San Fransico, 2006. It is necessary to diagnose multiple faults using compacted responses. ATE: Automatic Test Equipment CUT: Circuit Under Test

3. 3 Outline  Related Works  Terminology Explanation Capability Explanation Necessity  Issue of Compacted Responses  Diagnosis Method Suspect Fault Marking Suspect Fault Simulation Suspect Fault Evaluation Suspect Fault Ranking  Experimental Results

4. 4 Related Works Fault diagnosis approaches using compacted responses can be classified into three categories:  Bypassing diagnosis Bypass space compactor [ J. Ye, Y. Hu, et al., DATE10 ] [ F. Wang, Y. Hu, et al., ITC09 ]  Indirect diagnosis Mathematically de-compact the compacted responses [ H.P.E. Vranken, S.K. Goel, et al., DAC06 ] [ J. Rajski, J. Tyszer, et al., ITC03 ]  Direct diagnosis Use space compactor Compare actually observed responses with simulated compacted responses [ S. Holst, H.-J. Wunderlich, DATE09 ] [ W.-T. Cheng, K.-H. Tsai, et al., ATS04 ] Our method is a direct diagnosis method.

5. 5 Terminology Explanation Capability The explanation capability of a suspect fault reflects the match degree between its simulated responses and the actually observed responses. For one Observation Point (OP): Observed signal Simulated signal of a suspect fault Explanation Capability FailingPassing Failing The suspect fault explain the failing OP. Passing Failing The suspect fault contaminate the passing OP. ( Failing: A signal is different from the fault-free signal; Passing: A signal is the same as the fault-free signal. )

6. 6 Terminology Explanation Capability ( OP: Observation Point ) Following is an example of a CUT under one failing pattern. Observed signal Simulated signal of a suspect fault Explanation Capability Failing The suspect fault explain the failing OP. PassingFailing The suspect fault contaminate the passing OP.

7. 7 Terminology ( OP: Observation Point ) Explanation Capability Observed signal Simulated signal of a suspect fault Explanation Capability Failing The suspect fault explain the failing OP. PassingFailing The suspect fault contaminate the passing OP. Following is an example of a CUT under one failing pattern.

8. 8 Terminology ( OP: Observation Point ) Explanation Capability Two measures: σ: The number of explained failing OPs of one pattern. ι: The number of contaminated passing OPs of one pattern. A suspect fault with a high ( σ – ι ) reflects its high explanation capability.

9. 9 Terminology Explanation Necessity The explanation necessity of a suspect fault reflects its importance for matching the observed responses.

10. 10 Terminology N ce : The number of suspect faults which can contaminate/explain a passing/failing observation point. Each observation point is given a weight ω =. Explanation Necessity The explanation necessity of a suspect fault reflects its importance for matching the observed responses.

11. 11 Issue of Compacted Responses Using uncompacted responses, when multiple faults exist, the suspect faults with the highest explanation capability often hit the actual faults [9]. The situation, that the actual faults do not show the highest explanation capability, rarely occurs. This situation becomes serious when using compacted responses. [9] J. Ye, Y. Hu, and X. Li, “Diagnosis of Multiple Arbitrary Faults with Mask and Reinforcement Effect,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 885-890, 2010. Actual faults: a/0 (a-stuck-at-0), c/0 Failing observation points: Cell 11, q Suspect faults: a/0(d/1), c/0(q/0)

12. 12 [9] J. Ye, Y. Hu, and X. Li, “Diagnosis of Multiple Arbitrary Faults with Mask and Reinforcement Effect,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 885-890, 2010. Actual faults: a/0 (a-stuck-at-0), c/0 Failing observation points: OP 1, q Suspect faults: b/0 Failing observation points: Cell 11, q Suspect faults: a/0(d/1), c/0(q/0) Explanation capability alone is not enough for selecting right suspect faults. Using uncompacted responses, when multiple faults exist, the suspect faults with the highest explanation capability often hit the actual faults [9]. The situation, that the actual faults do not show the highest explanation capability, rarely occurs. This situation becomes serious when using compacted responses. Issue of Compacted Responses

13. 13 Each suspect fault is given a score ε under one given pattern: Combine explanation capability and explanation necessity Issue of Compacted Responses

14. 14 Diagnosis Method Marking Simulation Evaluation Ranking

15. 15 Diagnosis Method Marking Simulation Evaluation Ranking Suspect Fault Marking Search failing observation points’ related logic cones.

16. 16 Diagnosis Method Marking Simulation Evaluation Ranking Suspect Fault Simulation Each single suspect fault is simulated under all the given patterns to record their compacted responses. The structure of space compactor is known.

17. 17 Diagnosis Method Marking Simulation Evaluation Ranking Suspect Fault Evaluation Each suspect fault is given a score ε under each given pattern: The Final Score (FS) of a suspect fault is the sum of its scores under all given patterns.

18. 18 Diagnosis Method Marking Simulation Evaluation Ranking Suspect Fault Ranking ( OP: Observation Point )

19. 19 Experimental Results Experimental Setup Benchmark circuits: ISCAS’89, ITC’99 Test patterns: Generated by a commercial ATPG tool Injected fault types: Stuck-at faults, Transition faults Space compactor: For each circuit, each number of multiple faults, each type of faults, and each compaction ratio, 100 diagnosis cases are conducted. [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009. Circuits9234s13207s15850s35932s38417 Number of test patterns14227513224104

20. 20 Experimental Results Evaluation Metrics Success rate In this work, only the accuracy of the most likely suspect faults is concerned. A diagnosis case is considered successful if the top-ranked suspect faults hit at least one actual fault. Experimental focus Success rate vs Compaction ratio Success rate vs The number of injected faults Comparison with [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009.

21. 21 Experimental Results Success rate vs Compaction ratio S (%): Success rate of our method ΔS (%): Success rate of our method - Success rate of the method proposed in [20] T (seconds): Average run time of diagnosis in our experiments [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009.

22. 22 Experimental Results [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009. Success rate vs Compaction ratio S (%): Success rate of our method ΔS (%): Success rate of our method - Success rate of the method proposed in [20] T (seconds): Average run time of diagnosis in our experiments

23. 23 Success rate vs Compaction ratio Compaction ratio = 2 Compaction ratio = 32 Experimental Results [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009. 94.98%99.94% 86.47%98.64% s9234, s13207, s15850, s35932, s38417, s38584, b17, b20, b22

24. 24 Experimental Results [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009. Success rate vs The number of injected faults Compaction ratio = 32 S (%): Success rate of our method ΔS (%): Success rate of our method - Success rate of the method proposed in [20] T (seconds): Average run time of diagnosis in our experiments

25. 25 Success rate vs The number of injected faults Number of injected faults = 2 Number of injected faults = 32 Experimental Results [20] S. Holst, H.-J. Wunderlich, “A Diagnosis Algorithm for Extreme Space Compaction,” Proc. of Design, Automation, and Test in Europe (DATE), pp. 1355-1360, 2009. ( Compaction ratio = 32 ) 91.4%99.2% 49.8%93.3% s9234, s13207, s15850, s35932, s38417, s38584, b17, b20, b22

26. 26 Propose the explanation necessity. Combine the explanation capability and the explanation necessity: 98.8% of top-ranked suspect faults hit actual faults. Conclusion

27. 27 Thank You For Your Attention! Questions?