"Moore's Law Is Really Dead: What's Next?"
The 50-year reign of Moore’s Law, which delivered a billion-fold increase in transistors per chip, is finally over. Given that transistors are no longer getting much better, that the power budgets of microprocessors are not increasing, and that we’ve already replaced the single power-hungry processor with several energy-efficient ones, the only path to improve energy-performance-cost is specialized hardware. Microprocessors of the future will include special-purpose processors that do one class of computation much better than general-purpose processors. Accelerators for deep neural networks are but one of many potential targets. Panelists will discuss what old doors this seismic change will close and what new doors it will open.
John Hennessy, Stanford University (Moderator); Doug Burger, Microsoft Research; Norman P. Jouppi, Google; Margaret Martonosi, Princeton University; and Butler Lampson (1992 Turing Laureate), Microsoft