RATS Reliability and Testability Seminar Effective TARO test
RATS (Reliability and Testability Seminar) Effective TARO test pattern generation By Intaik Park, RATS, Summer 2004
Outline • • Introduction Basic Concepts Algorithm Experimental Result Intaik Park, RATS, Summer 2004 2
Introduction a b 1 2 c 3 d • Normal transition fault patterns one path of each fault • TARO patterns all paths of each fault 4 5 e 6 f Fault site Output (PO+FF) Intaik Park, RATS, Summer 2004 3
TARO • TARO (Transition fault to All Reachable Outputs) test pattern + More thorough than a transition fault pattern + Can be generated without path analysis – Long test pattern generation time – Large test pattern size longer test time larger tester memory requirement Intaik Park, RATS, Summer 2004 4
Agenda • Need better way of generating TARO patterns • Need smaller pattern size Intaik Park, RATS, Summer 2004 5
Outline • • Introduction Basic Concepts Algorithm Experimental Result Intaik Park, RATS, Summer 2004 6
Required information • Need to propagate and detect faults through all reachable outputs of each fault requires information about reachable outputs of each fault Intaik Park, RATS, Summer 2004 7
Definition • Reachable outputs : outputs through which a fault can be propagated and be detected • Detectable faults : faults that an output can observe and detect Intaik Park, RATS, Summer 2004 8
Reachable outputs a b c 1 • Fault a output 1, 2 2 • Fault b output 2 3 d 4 5 e • Fault d output 4 • Fault e output 2, 3, 4, 5 … 6 f Fault site • Fault c output 2, 3 Output (PO+FF) Intaik Park, RATS, Summer 2004 9
Detectable faults a b c 1 • Output 1 Fault a 2 • Output 2 Fault a, b, c, e 3 d 4 • Output 4 Fault d, e … 5 e 6 f Fault site • Output 3 Fault c, e Output (PO+FF) Intaik Park, RATS, Summer 2004 10
Detectable faults (contd. ) • For TARO pattern generation, need reachable output list for each fault find detectable fault list for each output • Logical corn analysis on cut (hard, inaccurate) • ATPG with output masks (easy, accurate) Intaik Park, RATS, Summer 2004 11
ATPG with output masks a b c 1 • Output 2 unmasked 2 • Masks on all other 3 d 4 e f Fault site outputs • Pattern will detect all 5 faults in detectable fault 6 list of output 2 Output (PO+FF) Intaik Park, RATS, Summer 2004 12
Outline • • Introduction Basic Concepts Algorithm Experimental Result Intaik Park, RATS, Summer 2004 13
Simple algorithm • Run ‘ATPG with output masks’ on each output + Very fast ( no additional information necessary) – Large pattern size ( many don’t-care terms, unused outputs) need more efficient algorithm Intaik Park, RATS, Summer 2004 14
Definition • Fault-output pair : a pair of fault site and one of its reachable output (a transition path) Intaik Park, RATS, Summer 2004 15
Efficient algorithm • Use output masks to control to which output the fault will propagate • Assign fault-output pairs that have outputs not masked • Let ATPG try best on each round of pattern generation Intaik Park, RATS, Summer 2004 16
Efficient Algorithm (Contd. ) a b c • After a transition fault ATPG 1 • Not detected fault-output pair: 2 Fault a 1 output 3 c 2 4 e 2, 3, 4 5 e 4 6 f 5 d e f Fault site Output (PO+FF) Intaik Park, RATS, Summer 2004 Used path Unused path 17
Efficient Algorithm (Contd. ) a b 1 2 c • Fault a output 1 • Fault c excluded 3 d 4 • Fault e 3 or 4 (or both) • Fault f 4 5 e 6 f Fault site • Mask on output 2, 5, 6 Used path Unused path Output (PO+FF) Intaik Park, RATS, Summer 2004 Assigned path 18
Efficient Algorithm (Contd. ) a b 1 2 c 3 d 4 a, e, f • Output masked 2, 5, 6 5 e 6 f Fault site • Fault assigned Used path Unused path Output (PO+FF) Intaik Park, RATS, Summer 2004 Assigned path 19
Efficient algorithm (contd. ) start reachable output info Verification by fault simulation Initial ATPG without masks All path detected? ATPG with masks No Assign faults / masks Yes end Intaik Park, RATS, Summer 2004 20
Assignment • Greedy algorithm 1. Pick a fault that has unused path 2. Add used outputs to mask list 3. Add unused outputs to unmask list • ‘Order of pick’ makes difference – which fault to pick first? Intaik Park, RATS, Summer 2004 21
Assignment heuristics • 1. 2. 3. 4. 5. Pick faults with … Most reachable outputs first Most unused path first Least unused path first Random Intaik Park, RATS, Summer 2004 22
Outline • • Introduction Basic Concepts Algorithm Experimental Result Intaik Park, RATS, Summer 2004 23
Experimental Result • ELF 35 cores – 4 combinational cores • 1 translator, 3 datapaths – 2 sequential cores with full scan • 2901’s Intaik Park, RATS, Summer 2004 24
Elf 35 Cores Core |Gates| |PO| |Scan FF| M 12 1309 12 NA SQR 538 6 NA MA 4499 33 NA PB 17468 12 NA LSI 2901 12338 64 544 TOPS 2901 18090 48 961 Intaik Park, RATS, Summer 2004 25
Core characteristics Core M 12 SQR PB MA lsi 2901 tops 2901 Transition faults 1, 629 625 57, 520 9, 973 11, 819 22, 131 TARO faults 16, 009 160, 176 2, 685, 293 93, 621 Gates 1, 309 538 17, 468 4, 499 12, 338 18, 090 Outputs 12 6 12 33 608 1, 009 2, 746 139, 420 * For seq. cores, output = PO + scan FF Intaik Park, RATS, Summer 2004 26
Number of TARO patterns (comb. ) Transition fault Simple Algorithm Heuristic 1 Heuristic 2 Heuristic 3 Heuristic 4 Heuristic 5 M 12 106 1, 057 586 451 480 886 629 SQR 50 252 189 158 179 218 163 PB 5, 633 26, 960 7, 357 7, 182 7, 185 7, 267 7, 260 MA 143 5, 564 NA NA NA * Number of patterns Intaik Park, RATS, Summer 2004 27
Number of TARO patterns (seq. ) Transition fault Simple Algorithm LSI 2901 624 561, 410 TOPS 2901 735 26, 630 * Number of patterns Intaik Park, RATS, Summer 2004 28
Issues • Long running time – Depends on number of outputs and fault observability • Large pattern size – Many don’t care terms Static compaction Intaik Park, RATS, Summer 2004 29
Future work • Finish TARO generation on sequential cores • Apply static compaction • Is TARO too thorough? Any way of excluding unnecessary thoroughness? Intaik Park, RATS, Summer 2004 30
Reference • Chao-Wen Tseng; Mc. Cluskey, E. J. ; Test Conference, 2001. Proceedings. International , 30 Oct. -1 Nov. 2001 Pages: 358 - 366 Intaik Park, RATS, Summer 2004 31
Previous work pattern length CUT TARO (Previous) (this method ) SQR Transition fault 157 222 158 M 12 213 246 451 MA 359 606 NA Intaik Park, RATS, Summer 2004 32
Core characteristics Core M 12 SQR PB MA lsi 2901 tops 2901 TARO / Transition Faults 9. 82 4. 39 2. 42 16. 06 227. 20 4. 23 Gates 1, 309 538 17, 468 4, 499 12, 338 18, 090 Outputs 12 6 12 33 608 1, 009 * For seq. cores, output = PO + scan FF Intaik Park, RATS, Summer 2004 33
- Slides: 33