1 / 55

Mahim Mishra and Seth C. Goldstein Carnegie Mellon University

International Test Conference Charlotte, NC, Sep 30-Oct 2, 2003 Defect Tolerance at the End of the Roadmap. Mahim Mishra and Seth C. Goldstein Carnegie Mellon University. Purpose. Future technologies: EUV Lithography and Chemically Assembled Electronic Nanotechnology

deiondre
Download Presentation

Mahim Mishra and Seth C. Goldstein Carnegie Mellon University

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. International Test ConferenceCharlotte, NC, Sep 30-Oct 2, 2003Defect Tolerance at the End of the Roadmap Mahim Mishra and Seth C. Goldstein Carnegie Mellon University

  2. Purpose • Future technologies: EUV Lithography and Chemically Assembled Electronic Nanotechnology • Single-digit nanometer feature sizes • Extreme device densities • Problem: much higher defect rates • Defect tolerance becomes key issue • We outline a defect tolerance strategy • novel testing method

  3. Talk Outline • Introduction – need for defect tolerance • Outline of defect tolerance strategy • Testing requirements • Description of proposed test strategy • Evaluation using simulations

  4. Red Brick Wall! Towards the end of the (ITRS) Roadmap • Feature sizes approach single-digit nanometers • Physical and economic limits to scaling • New Technologies • Chemically Assembled Electronic Nanotech. (CAEN) • Extreme Ultraviolet (EUV) Lithography

  5. New technologies: caveats • Extremely high defect densities • As high as 10% of fabric logic and routing resources • Cannot throw away defective fabrics • Defect-free yield: close to 0% • Must find a way to use defective fabrics

  6. Part of the solution: reconfigurability • Regular, periodic computing fabric • e.g., Field-Programmable Gate Arrays (FPGAs) • User programs ascircuits • Helps achieve defect tolerance

  7. New challenges for defect tolerance • New testing techniques to locate all the defects • generate a defect map • New, quick, place-and-route algorithms • utilize the defect map • Must scale with fabric size and number of defects

  8. Tester Fabric Defect Map Defect Aware Place-and- Route “Hard” Config. Defect Unaware Place-and- Route “Soft” Config. Proposed tool flow (Async.) Circuit Netlist

  9. Testing to locate defects • Required: scalable testing method to locate defects in large reconf fabrics • Capable of dealing with large defect densities, quick • Very different from previous FPGA testing methods • Similar approach: Teramaccustom computer at HP • Goal of this work • Show that such a testing method is possible • Develop some of the new, smart techniques required

  10. Testing: previous methods Defect-free, unknown Defective, unknown Defect-free, known Defective, known

  11. Testing: previous methods Defect-free, unknown Defective, unknown Defect-free, known Defective, known Returns success Returns failure

  12. Testing: previous methods Defect-free, unknown Defective, unknown Defect-free, known Defective, known Returns success Returns failure

  13. Testing: previous methods Defect-free, unknown Defective, unknown Defect-free, known Defective, known Returns success Returns failure

  14. Testing: previous methods Defect-free, unknown Defective, unknown Defect-free, known Defective, known Returns success Returns failure

  15. Testing with high defect rates • Previous method: works for low defect rates • Uses “binary” circuits • Requires significant number of defect-free test-circuits • Will not work for high defect rates • Each test circuit has multiple defects • Very, very few circuits with 0 defects

  16. Testing: high defect rates Defect-free, unknown Defective, unknown Defect-free, known Defective, known

  17. Testing: high defect rates Defect-free, unknown Defective, unknown Defect-free, known Defective, known Returns success Returns failure

  18. Testing: high defect rates Defect-free, unknown Defective, unknown Defect-free, known Defective, known

  19. Defect- Location Phase Probability- Assignment Phase 2 key ideas: • More powerful test-circuits • More than binary info; e.g. approximate defect counts • More powerful analysis techniques Dealing with high defect rates: our algorithm • Eliminates remaining defects • Deterministic; no mistakes • Finds probabilities of being defective • Eliminates components w/ high prob. “Probabilistic” Defect Map Defect Map Fabric

  20. Probability assignment: example Defect-free, unknown Defective, unknown Defect-free, known Defective, known

  21. Probability assignment: example Defect-free, unknown Defective, unknown Defect-free, known Defective, known Test circuits: • Counter • None-some-many More than binary information! 1 2 1 2 1 2 1 3

  22. Probability assignment: example 1 2 1 3 2 1 1 2 Defect-free, unknown Defective, unknown Defect-free, known Defective, known 1 2 1 2 1 2 1 3

  23. Probability assignment: example 1 2 1 3 2 1 1 2 Component Defect Probabilities Analysis Method Analysis methods: • Sorting • Bayesian 1 2 1 2 1 2 1 3

  24. Probability assignment: example Defect Probabilities higher lower

  25. Probability assignment: example Removed: & Assumed Defective

  26. Probability assignment: example

  27. Defect location: example Returns success Returns failure Assumed Defective Binary Information

  28. Defect location: example Returns success Returns failure Assumed Defective

  29. Defect location: example Returns success Returns failure Assumed Defective

  30. Defect location: example Returns success Returns failure Assumed Defective

  31. Final defect map

  32. Test circuits for prob. assignment • Idealized countercircuits • Conceptual circuits • Return defect counts, upto threshold t • Higher threshold  more powerful circuit • None-some-many circuits • Tell if none, some or many defects • Less powerful than counters, easier to realize • e.g., our LFSR-based design

  33. (a) (b) (c) no no at most 1 of (b), (c) wrong? (a) right? many yes yes some none None-some-many circuits

  34. Analysis methods • Sorting analysis • Example • Bayesian analysis • See paper • Comparison

  35. Analysis methods: Sorting analysis 1 2 1 3 2 1 1 2 1 2 1 2 1 2 1 3

  36. Analysis methods: Sorting analysis 1 2 1 3 2 1 1 2 5 1 2 1 2 1 2 1 3

  37. Analysis methods: Sorting analysis 1 2 1 3 2 1 1 2 5 2 1 2 1 2 1 2 1 3

  38. Analysis methods: Sorting analysis 1 2 1 3 2 1 1 2 5 Higher probability of being defective 2 1 2 1 2 1 2 1 3

  39. Analysis methods: comparison • Ease of implementation • Bayesian: harder to implement (restricted circuits) • Sorting: no restrictions • Complexity • Bayesian: O(n2) best case • Sorting: O(n logn ) • Quality of results: ~10% better for Bayesian

  40. Algorithm: discussion • Minimally-adaptive algorithm • Minimal rerouting required at test time • No false negatives • After defect-location phase, all defects identified • Algorithm complexity: • circuit size k • defect rate p • k× k fabric • requires O(kp) test-circuit orientations

  41. Evaluation • Quality metric: recovery • percentage of defect-free components markednot defective • Each simulated test circuit: 100 components • Simulated defect rates: 1 to 13% • 1 to 13 defects on average per test circuit • Results valid for circuits with this many defects

  42. Evaluation results: comparison sorting Bayesian Ctr Nsm

  43. Eval.: counter circuits, Bayesian anal.

  44. Evaluation results: comparison sorting Bayesian Ctr Nsm

  45. Each line: represents a different number of pieces into which the large LFSR is broken Eval.: n-s-m circuits, Bayesian anal.

  46. Evaluation results: comparison sorting Bayesian Ctr Nsm

  47. Evaluation results: clustered defects • So far: uniformly distributed defects • In VLSI: defects often clustered • Clustered defects higher recovery

  48. Conclusions • New manufacturing paradigm • Reduced manufacturing complexity and cost • Increased post-fabrication testing and defect-tolerant place-and-route effort • Defect tolerance is a major challenge • Locate defects and configure around them • Scalable testing with high recovery is possible

  49. Backup slides • Fabric architecture • Algorithm • Probability calculation • Wave testing • Individual results graphs

  50. Control, configuration &defect mapping seed cluster long-lines nanoBlock switch-block nanoFabric architecture (ISCA’01)

More Related