ISCA used to publish retrospective evaluations of deployed hardware systems from industry that were influential and widely read. A recent blog post studied 2000+ architecture papers of the ISCAs from 1973-2018, and observed a decline of papers by first authors from industry. Moreover, while research groups in computer architecture (like at IBM, Intel, and NVIDIA) may have as much support for architectural exploration and publication as academic ones, product projects certainly don’t. Few industrial product architects have permission (let alone time or motivation) to write ISCA papers.
Many see great value in papers about working hardware that put novel ideas together that must work well together to help us understand the difficulty, cost, and performance of the ideas and the overall system. Such papers are unlikely to be accepted by the current ISCA reviewing ecosystem, since they are so different. Retrospective industrial product papers remain valuable complements to academic research papers, so we are experimenting with a new mechanism to encourage their return. While a worthy experiment, we should revisit the concept in 3 to 5 years to see if it works. If not, we would end it, and look at other ways to broaden participation.
Industry track papers will have a separate program committee to evaluate them, who understand the constraints of publishing papers on hardware from industry. The papers would still be held to the same high ISCA standards, but the program committee will recognize that company concerns about patent trolls or trade secrets may mean some details are not revealed.
David Patterson Google; UC Berkeley
|Sophia Shao||UC Berkeley|