Timing-Aware ATPG for High Quality At-speed Testing of Small Delay Defects

Pseudo-functional testing for small delay defects considering power supply noise effects

“…When comparing against patterns with maximum PSN effects without considering functional constraints, we can observe more than 14% less PSN effects on average for all benchmark circuits, and scan patterns for benchmark s15850 can result in up to 23% more power supply noises than our patterns that try to maximize PSN effects under functional constraints. This comparison indicates that it is crucial to take functional constraints into consideration when generating SDD test pat- Result in [6] Result of our method terns. Otherwise, circuits can be over-tested, leading to significant test yield loss.…”

“…In our method, however, we drop those undetected faults if and only if they are located on the path targeted in relevant transition justification, since we need to guarantee the dropped faults have been affected by sufficient PSN effects. Consequently, our solution generates more test patterns than [6]. We compare the pattern count between the two methods, as shown in the Fig.9.…”

“…In our second experiment, we randomly select six true critical paths sensitized by the SDD test patterns generated by [6] and then record the maximum TW and minimum TW when these paths are under test, as shown by "Orig. MIN" and "Orig.…”

“…In [6], once a SDD pattern is generated, fault simulation is conducted and all the faults that are propagated through long paths are dropped. In our method, however, we drop those undetected faults if and only if they are located on the path targeted in relevant transition justification, since we need to guarantee the dropped faults have been affected by sufficient PSN effects.…”

“…To generate pseudo-functional SDD test cube, we extend the conventional SDD test generation method presented in [6] by integrating the functional constraints checking and breaking mechanism, which is to sensitize SDD through critical paths whenever possible, meanwhile, it guarantees that no illegal states is included in the test cube by setting them as functional constraint as [10].…”