1. Reduced Complexity Test Generation Algorithms for Transition Fault Diagnosis Yu Zhang Vishwani D. Agrawal Auburn University, Auburn, Alabama 36849 USA Oct 10th1

2. Outline Purpose (motivation) Problem statement and contribution Introduction and background Representing a transition fault as a single stuck-at fault Exclusive test patterns for transition faults –One and two time frame models Experimental Results Conclusion Oct 10th2

3. Purpose Oct 10th3 Many modern design failures behave as non-classical faults. Most failures are timing related. Transition fault model is widely used due to its simplicity. There exist a need for diagnosis using the transition fault model.

4. Problem Statement and Contribution Oct 10th4 Modeling and test generation for transition faults: –Detection of single transition faults –Exclusive tests for fault-pairs Present contribution: –A diagnostic ATPG system for transition faults using conventional fault-detection tools.

5. Introduction Oct 10th5 Previous work*: –A diagnostic coverage metric –Diagnostic fault simulation –Exclusive test generation for stuck-at fault –Diagnostic ATPG system * Yu Zhang, V. D. Agrawal, “A Diagnostic Test Generation System,” in Proc. Int. Test Conf., 2010. Paper 12.3.

6. Introduction Oct 10th6 Given a set of vectors, we define: Diagnostic Coverage: Where g 0 is the group of undetected faults. Fault group: Set of faults detected by same vectors at same outputs (hence indistinguishable). Fault coverage (conventional)

7. DC vs. Fault-Pair Coverage – s27 Oct 10th7

8. Introduction Oct 10th8 line x 2 PI PO Single circuit copy ATPG: find a test vector to distinguish fault f1 (line x 1 s-a-a) from fault f2 (line x 2 s-a-b) s-a-b line x 1 s-a-a x1’x1’ x2’x2’

9. Introduction Oct 10th9 PI PO Single copy exclusive test generation: y x1x1 CUT C x1’x1’ a x2x2 b x2’x2’

10. Introduction Oct 10th10 Suppose a is 0 and b is 1, the model can be simplified: PI PO y x1x1 x1’x1’ x2x2 x2’x2’ CUT C

11. Representation of a Transition Fault Oct 10th11 x’ Slow to rise x x’ x MFF init. 1 01 MFF Model:

12. Detection Test Generation Oct 10th PI x x’ MFF init. 1 Using MFF Model: y PO 12 s-a-1 Test for y sa1 is also a test for xx’ slow to rise

13. Two-time-frame Model (Simplified): Oct 10th13 xx’ Detection test for xx’ slow-to-rise Useful for equivalence identification x x’ PIPO s-a-1 Detection Test Generation y

14. Single Copy Exclusive Test Generation Oct 10th14 x1x1 x1’x1’ x2’x2’ x2x2 MFF init. 1 s-a-0/1 0 1 MFF init. 0 0 1 PO PI Exclusive test for x 1 x 1 ’ slow-to-fall and x 2 x 2 ’ slow-to-rise:

15. Single Copy Exclusive Test Generation Oct 10th15 x1x1 x1’x1’ x2’x2’ x2x2 s-a-0/1 MFF init. 0 PO PI MFF init. 1 Simplified version:

16. Advantages of Exclusive Test Algorithm Reduced complexity: Single-copy ATPG model is no more complex than a single fault ATPG. No need for especially designed diagnostic ATPG tools. Can take advantage of various existing fault detection ATPG algorithms. Oct 10th16

17. Experimental Results Oct 10th17

18. Need for Equivalence Identification Oct 10th18 Some fault-pairs are functionally equivalent. Exclusive test ATPG may leave many undiagnosed fault pairs as aborted faults causing low DC. Many techniques have been proposed for fault equivalence identification: –Structural analysis –Exhaustive enumeration –Learning & implication –Branch & bound –Circuit transformation & symmetry identification

19. Conclusion A new diagnostic ATPG system for transition fault is constructed. Only conventional tools are used: –Exclusive test generation for transition fault requires only single stuck-at fault detection. –Fault equivalence checking is important for DC; requires effective algorithm. Oct 10th19

20. Oct 10th20