Post silicon analytics
#1
Hi
I'm a semiconductor specialist solutions architect, and I'm building an open-source post-silicon (yield) analytics solution. 
I'm hoping to partner with the PULP project, looking for parties that have taped-out an open-source chip, and have STDF files we can use to build this new tool...

Does the PULP project tape out chips? Is there a possibility to collaborate here?

Thanks,
Reply
#2
Hello,
I think I have also received from you the same request through an other channel, allow me to respond here.

Short version (see below for a longer discussion as to why): We could probably get access to STDF data for the runs we have participated, but we rarely do.

Even if we did have access to this data, I am pretty sure that we would not be able to share this with you (even if you were working for a non-profit organization) as this is as 'intimate' data as you can get from a manufacturer which traditionally are very conscious about the information they share. As more and more fabs are moving (or considering to move) to openPDKs, such data may also be more publicly available.

The best course of action would be to contact the manufacturers themselves to request this data. If a manufacturer needs our permission to release data to you, we would be more than happy to give our support.

Best Regards,
Frank

The longer version:

As the PULP team our IC manufacturing is almost exclusively through multi-project-wafer (MPW) runs for prototyping quantities. For most of the cases, this would result in us getting about 50 to 100 baredies, a fraction of which would be packaged. We have benefited tremendously from the Europractice IC service, (as well as Muse) as a customer. In these transactions we are a customer, and only in (rare) cases where we have issues with our results do we make requests to see STDF (https://en.wikipedia.org/wiki/Standard_Test_Data_Format) data from the runs we have participated. I believe this happened at least once, but not more than five times (out of the 50+ PULP chips, and 150 or so other chips that went from ETH Zurich) and in none of these cases did the information from STDF help us resolve our issues (other than showing that there were no unexpected variations among the dies we had access to). Part of the issue here is that although the fab has very good records for dies and their placement, for MPW services, once you dice, subdice and pick and place them in a gelpack or die carrier, the physical information which die ended up where in the gelpack goes away (as this is usually done by a subcontractor) and once that gelpack arrives at a packaging facility where they pick (say) 10 dies ouf the 50 and do the bonding, information is once again lost (maybe not completely, as the statistics are there, so we have a probability) . In any case, for one reason or another, we did not have any direct benefit from having access to STDF in our work (with the volume we have) and do not pursue to get this information for our runs.

The second complication is that we usually do not do full mask runs, meaning that the wafers where our chips are manufactured are shared with multiple tenants. I would think that the manufacturer would need separate releases from every tenant on an MPW (which is in the tens). Just the effort of obtaining these would be a significant burden. We would be happy to release data if a manufacturer would ask us permission, but I think the STDF data would have to come from the manufacturer.
Visit pulp-platform.org and follow us on twitter @pulp_platform
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)