CAD Research Productivity and Impact: Gap Analyses and Illustrations

Abstract

Why does so much research activity (in CAD) seem oblivious or incomparable to previous work? Why didn't X cite/apply/compare against the work of Y? How much published research effort is essentially wasted or lost, and why?

CAD research is tied to practical applications. It is dominated by heuristic techniques and by engineering tradeoffs between quality of result and resource usage. Considering this context suggests some ways in which the CAD research community is less effective than it could be. I will discuss two kinds of "gap" -- (i) "barriers to entry" (scoping, testbed quality, etc.) and (ii) standards (for replicability, for experimental evaluation, for reporting, etc.) -- and why they are worsening fast.

I'll then illustrate how difficult it is to overcome these gaps (while engaged in academic research), using examples from VLSI physical design, and the implementation and evaluation of move-based hypergraph partitioning heuristics in particular. Issues that will be discussed (for partitioning) in some detail: obtaining an adequate software testbed; having a correct use model and maintaining relevance; understanding the body of previous work; and achieving meaningful experimental evaluation of new ideas. I will conclude with speculations regarding the types of culture change that will let us all increase our research productivity and impact.