Jeff Allen Jacob Biamonte

1. ECE 572/672 Project: Testing Jeff Allen Jacob Biamonte

2. Other important moments in the history of quantum test set generation • Original Idea from right here at PSU! • Markov/Hayes at the U.M.  work was done on reversible test set generation that at least made one think about quantum test set generation (We are the only group in the world to cite this paper thus far) This is one of the most fundamental papers published in recent times! • Many papers exist on regular testing, U.M. group and us are the only ones doing this research now (I think anyway), when reversible computers become commercial 1,000s of test set generation papers will be written, we just don’t know when this will happen but we hope it will be in our life times. • Ed Perkins wrote some reversible test generation software last year, he did a good job but I have a new method and new software

3. Goal Number One • Illustrate classical known method to detect and localize faults on a simple AND gate • Explain the classical fault model • Explain the use of this in large scale binary and analog circuit design

4. a a a a a a a y y y y y y y b b b b b b b Classical Fault Localization • Example 1: AND gate inserting stuck-at faults. Good Circuit • Seven possible situations: Stuck-at-1 at a Stuck-at-1 at b Stuck-at-1 at y Sa1 Sa1 Sa1 Stuck-at-0 at a Stuck-at-0 at b Stuck-at-0 at y Sa0 Sa0 Sa0

5. Classical Fault Localization Create truth table showing good circuit and all cases of faults  Indicates a faulty output (output different than that of good circuit) • Goal: minimize the number of test vectors needed to detect and localize all faults

6. T2 a b sa1 a sa1 y T1 0 0 X T3 1 0 X P F a b Good Circuit sa1 b sa0 a sa0 sa0 y T4 1 1 T1 0 0 T3 1 0 X T4 1 1 X X X Classical Fault Localization Input Vectors Chosen so as to detect and localize all faults with minimum number of test vectors a b Good Circuit sa1 a sa1 b sa1 y sa0 a sa0 sa0 y T1 0 0 X T2 0 1 X X T3 1 0 X X T4 1 1 X X X Fault Table of AND gate with stuck at faults Before Test: Possible: Good Circuit, sa1 a, sa1 b, sa1 y, sa0 a, sa0 b, sa0 y Applying Test Vector 01: After Test: Possible: sa1 a, sa1 y Possible: Good Circuit, sa1 b, sa0 a, sa0 b, sa0 y

7. P F T3 T2 T4 T3 F Sa1@a Sa1@b Sa1@y P F P a b sa1 a sa1 y T1 0 0 X T3 1 0 X a b Good Circuit sa1 b sa0 a sa0 sa0 y T4 1 1 a b Good Circuit sa0 a sa0 sa0 y T1 0 0 T1 0 0 T3 1 0 X T4 T4 1 1 P X X F X 1 1 X X X Good Circuit Sa0@a, Sa0@b, Sa0@y a b Good Circuit sa1 a sa1 b sa1 y sa0 a sa0 sa0 y Original Fault Table T1 0 0 X T2 0 1 X X T3 1 0 X X T4 Good Circuit, sa1 a, sa1 b, sa1 y, sa0 a, sa0 b, sa0 y 1 1 X X X Good Circuit, sa1 b, sa0 a, sa0 b, sa0 y sa1 a, sa1 y Good Circuit, sa0 a, sa0 b, sa0 y Because all of the stuck-at-0 faults have the same entries in the fault table, there is no way to localize them, unless we can measure all parts of the circuit. If we would have tested exhaustively it would have taken all 4 tests, we did it in 3, (and we know the type of error present!)

8. Goal Number Two • Show another example, but this time with a reversible circuit and Markov, Hayes stuck-at technique • Explain some of the differences between classical fault detection and reversible fault detection

9. Sa0 Sa1 Example 2, a reversible circuit  Stuck-at-0   Stuck-at-1  

10. Example 2 continued… Locations for faults 

11. Example 2 continued…

12. Example 2 B, localization

13. Example 2 B, localization

14. Compare Scaling Reversible v. Classical • Small reversible circuits have small gains compared to classical • Large Classical circuits often cannot be localized, where all reversible circuits can be localized

15. Reversible scaling • Each level of a reversible circuit can be partially tested with any test • A first test will test every C-NOT gates input, output and control half way

16. Example 2 revisited… Locations for faults 

17. Reversible scaling (cont) • The best each subsequent test can do is half of what is left. • Test overlap exists and can lower each successive test’s effectiveness

18. My Approach • Path propagation as opposed to fault tables • The Stuck-at model is incomplete • What about missing gate? • Bridging faults

19. What’s wrong with fault tables? • Memory requirements, a non linear increase exists when lines, stages, and or gates are added. • Good tables may require conditional branch solutions for localization.

20. Why Path Propagation? • Dynamic on the fly localization • Circuit can be loaded and testing can be started immediately. • Can be optimized to find known issues • Capable of providing OTF coverage specs

21. Path Propagation Example (Step One) 1 1 0 1 1 1 S0 S0 s0 s0 S1 s1 !s0 S0 S0 S0 S1

22. Path Propagation Example (Step Two) 1 1 0 0 1 1 0 S0 S1 S0 S1 s1 ! S0 s1 s0 S0 s0 s0 S0 S0 s0 s0 S1 S0 S1 S1 S0 1 0

23. Path Propagation Example (Step Three) 0 0 ** ** ** s1 S1 ** s1 ** ** s0 S0 ** s1 S0 S0 ** s0 ** S1 ** ** 0 0 1 1 Note Y-Stuck at 0 stage 1 never tested. And Stuck @ 0 can be missed in case of missing gates

24. Our Test Set • 110 • 011 • 001 • 10* to test Y-stuck@0 stage1

25. What does this mean? • The stuck at method can miss errors involving missing gates • 50% likely to miss, missing gate in stage 1 • Law of diminishing returns, how does it apply here?

26. Bridging Faults • In order to truly test bridging across lines one-hot and one-cool versions should be done • If single fault model used, implied bridging could be attempted

27. Implied Bridging Technique • At each connection and or xor two lists of all other lines for 1-0 and 0-1 oppositions • Percent bridge tested equals • ((tested 1-0) + (tested 0-1)) / (2 * totalnodes)

28. Thanks! • As can be seen by the example there are more calculations then what can be done by hand for larger circuits • Typically there is redundancy in the exhaustive method, this is far to complicated to be seen for people, but computers can remove it. The goal of this method is to remove the useless tests, and focus on the tests that give the most information about the circuit.