Professor Melvin Breuer presented an invited paper at the prestigious IEEE/ACM Design Automation Conference in June, 2010 held at the Anaheim Convention Center, Anaheim, Calif. Breuer discussed the controversial yet titillating concept of using defective VLSI chips in applications where they result in acceptable performance to the end user. The abstract can be found below.
"Hardware That Produces Bounded Rather Than Exact Results"
Technological achievements have made it possible to: fabricate CMOS circuits with over a billion transistors; implement Boolean operations using quantum devices and/or the spin of an electron; implement transformations using bio and molecular based cells. Problems with many of these technologies are due to such factors as process variations, defects and impurities in materials and solutions, and noise.
Consequently, many systems built from these technologies operate imperfectly. Luckily there are many complex and large-market systems (applications) that tolerate acceptable though not always correct results. In addition, there is emerging a body of mathematical analysis related to imperfect computation. In this paper we first introduce the concepts of acceptable error-tolerance and acceptable performance degradation, and demonstrate how important attributes of these concepts can be quantified. We interlace this discussion with several examples of systems that can effectively employ these two concepts. Next we mention several emerging technologies that motivate the need to study these concepts as well as related mathematical paradigms. Finally we will list a few CAD issues that are needed to support this new form of technological revolution.