DFT Digest

Test Compression makes test vectors for efficient - I wish there was such a thing for schedules… ITC Day 2 for me seemed like a lot of running around, trying to find people, trying to avoid people [joking! really...], answering e-mails, phone calls, finding cell phone chargers… I missed a lot, but I suppose I was productive otherwise.

The most fun I had all day Wednesday was witnessing Panel 3 - Will Test Compression Run Out of Gas? Moderated by Jeff Rearick of AMD, panelists Janusz Rajski of Mentor, Tom Williams of Synopsys, Somebody from Verigy standing in for Jochen Rivoir [update: Sandeep Goel, organizer of this panel, provided the name of the man standing in for Rivoir: Erik Volkerink - thank you Sandeep!], Sandeep Bhatia of Cadence, and Steve Pateras of LogicVision had a spirited discussion of whether test data compression technology has run its course or will continue to improve the efficiency of test patterns well into the future.

Mentor, as a company, is still improving on their technology, boasting orders of magnitude more compression than before, so understandably, Rajski’s postion is that compression has a ways to go, perhaps 10 more years. Both Bhatia and Williams took the opposite rail, and pronounced compression as good as it gets (unless the technology becomes something quite different).

Tom Williams argument, which I’ve heard him state several times in the last couple of years, it seems, is that once you have just 1% care bits in your test cube, you’ve got 100x, and that’s all you’re going to get. “The network just won’t support more than that”.

Steve Pateras used the panel to make his claim that everyone else was missing out, stating that he offered “infinite compression” with Logic BIST; with no vectors at all, he doesn’t see the point. Since LogicVision doesn’t have much of a stake in compression technology, Pateras just made sure that he promoted what LogicVision does bring to the table: BIST (although he also mentioned that they weren’t all out, since they do have a product called ScanBurst, which works together with Mentor’s TestKompress). He even got a plug in for LogicVision’s  mixed-signal BIST products when the subject of analog test came up.

The panel bantered back and forth on the information theory aspects, seemingly arguing somewhat orthogonal points - and the audience joined in with their take on the situation, for example, Grady Giles of AMD chimed in with the contention that we didn’t need a lot more compression, “we need more DFT”, and explained how with DFT work, the percentage of care bits can be reduced, and one can get great results out of even todays compression technology.

Well, that’s it for day 2 (it was now 2 days ago, I’m behind).  Next up,: day 3 (surprise!)