1. l i a b l eh kC o m p u t i n gL a b o r a t o r y On Effective and Efficient In-Field TSV Repair for Stacked 3D ICs Presenter: Li Jiang Li Jiang †, Fangming Ye *, Qiang Xu † Krishnendu Chakrabarty *, and Bill Eklow § † CUhk REliable Computing Laboratory The Chinese University of Hong Kong * Duke University § Cisco

2. Outline Introduction Related Works and Motivation In-field TSV Repair Framework Repair Algorithm Experimental Results Summary

3. TSV Latent Defects EM-induced Void Open Defect Signal Latency [Frank et al., IRPS’11] CTE-induced Crack Cooling Heating [Jung et al., ICCAD’11] crack

4. TSV Repair Schemes: Neighboring Repair To avoid Aging “hotspot”, we use signal-rerouting as our hardware infrastructure [Jiang et al., DATE’12] [Kang et al., JSSC’10]

5. Motivation Existing repair methods are deterministic Unaware of timing violating timing requirement after repair Hard to determine “faulty” TSV: A faulty TSV linking to a particular signal might be a good one if it links to another signal instead “Faulty” TSV propagation may render the entire TSV grid irreparable Repair solutions directly affect circuit lifetime reliability

6. Hardware Architecture Periodically On-line test In-field TSV Repair Repair Solution Validation Fail Success Circuit aging can also be detected

7. Repair Algorithm Signal-TSV pair graph: no confirmed timing violation Flow graph: routability checking Maximal Matching = #Signal ST-Graph Matched ST Pairs Potential ST Pairs Finding the maximal matching Repair Channels Residual Channels Test cost is too high in the runtime Flow-Graph

8. Next Matching Repair Algorithm To reduce test time Simultaneously Testing Previous Matching Current Matching Tested ST-Pairs Not Tested ST-Pairs Routable Avoid redundant test Test ST pairs from next matching in advance

9. Spare TSV Sharing How to solve the conflict of using shared spares? Merge STpair-graphs into connected STpair-graph

10. Experimental Setup Benchmark : IWLS 2005 OpenCore benchmarks data encryption standard (DES) circuit data encryption standard (DES) circuit fast-Fourier transform (FFT) circuit fast-Fourier transform (FFT) circuit Aging Effect: Characterized by additional latent delay in TSVs, reflected as resistance increase in terms of time t. Characterized by additional latent delay in TSVs, reflected as resistance increase in terms of time t. [Frank et.al, IRPS’11], [Ye et al. DAC’12] [Frank et.al, IRPS’11], [Ye et al. DAC’12] Parameters: TSV aging coefficient a, TSV initial resistance R Parameters: TSV aging coefficient a, TSV initial resistance R Following Normal Distribution Following Normal Distribution Modified Router based Repair Scheme MF: Continue repair if “new fault” occurs MF: Continue repair if “new fault” occurs MF’: Restore repair if “new fault” occurs MF’: Restore repair if “new fault” occurs Proposed Repair Algorithm MV: Match with Verified routability MV: Match with Verified routability MR: With test time reduction MR: With test time reduction MS: With spare TSV sharing MS: With spare TSV sharingComparison

11. Results Varied aging coefficients with fixed initial resistance

12. Summary First work targeting on in-field TSV repair First work targeting on in-field TSV repair An efficient TSV repair algorithm that is able to significantly improve MTTF of TSV through the judicious use of spares An efficient TSV repair algorithm that is able to significantly improve MTTF of TSV through the judicious use of spares Redundancy sharing technique can tolerate aging “hotspots” Redundancy sharing technique can tolerate aging “hotspots”

13. l i a b l eh kC o m p u t i n gL a b o r a t o r y Thank you for your attention !

14. Results 8x8 TSV grid size repair architecture with varied aging coefficients varied rerouting delay between two adjacent routers (ps)