unintrusive aging analysis based on offline learning
play

Unintrusive Aging Analysis based on Offline Learning Frank Sill - PowerPoint PPT Presentation

Unintrusive Aging Analysis based on Offline Learning Frank Sill Torres* + , Pedro Fausto Rodrigues Leite Jr.*, Rolf Drechsler + *Universidade Federal de Minas Gerias, Belo Horizonte, Brazil + University of Bremen, Bremen, Germany Sill Torres -


  1. Unintrusive Aging Analysis based on Offline Learning Frank Sill Torres* + , Pedro Fausto Rodrigues Leite Jr.*, Rolf Drechsler + *Universidade Federal de Minas Gerias, Belo Horizonte, Brazil + University of Bremen, Bremen, Germany Sill Torres - Aging Analysis

  2. Motivation  Aging of integrated systems of rising importance  But: – (Still) less critical for customer applications V – Interest in low weight solutions (S.M.A.R.T. for HDDs, …)  This work: – Low-weight aging monitoring / remaining lifetime prediction – Based on (offline) learning Sill Torres - Aging Analysis 2

  3. Aging Monitoring  In-situ slack sensors – Detection / preview of failing timing C – Added invasively to (selected) critical paths  Online self-testing – Built-In Self-Test (BIST) during test mode – Additional circuitry (Scan chains, …)  Aging sensors – Report experienced aging – Ignores system’s activity Sill Torres - Aging Analysis 3

  4. Unintrusive Aging Analysis Architecture Software Layer Profiling VDD, Freq., Sleep Counter- Reporting Simulations measures Hardware Stress Test Prediction MDB APDB Stress sensors Field Data Compression Temp, V, Activity  APDB, MDB: Databases Sill Torres - Aging Analysis 4

  5. Unintrusive Aging Analysis Profiling  Sensors – Temperature, voltage, activity, … – Low area offset, unintrusive Profiling  – Simulations  Aging characterization at design time  Various scenarios (Temp, VDD, activity, …)  Parameter can vary – Also possible: Data from stress test / field Sill Torres - Aging Analysis 5

  6. Unintrusive Aging Analysis Compression and Profile Storage Set 4  Compression of Set 4 Sensor Value simulated / Set 3 Set 3 measured data Set 2 Set 2  Insertion in Set 1 Set 1 Databases Set 0 0 10 20 30 40 Time … Sensor S T,4 … MTTF  Data bases for in Set 0 … in Set 4 [%] [%] – Profile Data (APDB) – Measured Data (MDB) 20 % 32 % 2e2 h MTTF – Mean Time To Failure Sill Torres - Aging Analysis 6

  7. Unintrusive Aging Analysis Prediction Models Prediction MDB APDB  Prediction – Relate Measured data (MDB) to Profiling Data (APDB) for prediction of current Remaining Useful Lifetime (RUL) – Three Models (Linear, Euclidean Distance, Correlation) Sill Torres - Aging Analysis 7

  8. Results Accuracy of Prediction Best (Linear): 90.4% 100 % 80 % 60 % 40 % 20 % 0 % INV c499 c880 c1355 c5315 Linear Euclidian Correlation Static Sill Torres - Aging Analysis 8

  9. Conclusions  Methodology for low weight prediction of aging of integrated systems  Application of profiling data  Consideration of varying parameters  Simulation results: Prediction accuracy ca. 90 % → Not exact but – Enables proactive counter measurements – User can be warned Sill Torres - Aging Analysis 9

  10. Unintrusive Aging Analysis based on Offline Learning Thank you! www.asic-reliability.com franksill@ufmg.br Sill Torres - Aging Analysis 10

  11. Activity Sensor  [7] R. Baranowski, et al., "On-line prediction of NBTI-induced aging rates," in DATE 2015, pp. 589-592.  Monitoring of switching activity of the circuit’s primary inputs (PI) or pseudo-primary inputs (PPI) Sill Torres - Aging Analysis 11

  12. Aging 80 FIT (Failures in 10 9 h) 60 40 20 0 130 nm 90 nm 65 nm 40 nm 25 nm Stratix Stratix II Stratix III Stratix IV Stratix V Altera, RELIABILITY REPORT 56, 2013 Sill Torres - Aging Analysis 12

Recommend


More recommend