Search

We have been following the dictum of Moore's Law for longer than most engineers have been alive. We're used to it, we know how to do it, and its focus on functionality, performance, and economics has yielded remarkable systems, from today's smartphones to supercomputers and internet servers, to engine controllers, anti-lock brakes, airbags, and the navigation systems in our vehicles. But software is notoriously buggy, and hardware isn't defect-free, either. Worse, we seem to only take security into account when some hacker perpetrates yet another outrage. The pending end of Moore's Law will greatly diminish the historical demand to replace systems every few years, which will put new pressure on system aging effects. Our military is using these same commercial systems, because they are the highest performance, most economical solutions, but this means that they, too, are now exposed to the same problems. By one estimate, more than half of automotive recalls are due to software defects. And here come self-driving cars, which could potentially suffer all of these threat axes simultaneously. Performance and efficiency enable new applications, and in the past, the computer design community could simply stop when those targets were within reach. In this talk, I will argue that we cannot get away with that attitude any longer, and need to turn our attention to areas of system design that are not good enough for how those systems will be used in the near future.