For decades, Moore's Law has delivered the ability to integrate an exponentially increasing number of devices in the same silicon area at a roughly constant cost. This has enabled tremendous levels of integration, where the capabilities of computer systems that previously occupied entire rooms can now fit on a single integrated circuit. It's well documented that Moore's Law is slowing on all its dimensions and corollaries. While improvements in density scaling continue at a reduced pace, the industry is simultaneously observing increases in manufacturing costs that are threatening to eliminate the cost per transistor generational improvement that has been fundamental to the value proposition of semiconductor compute products. In response to this existential threat to the industry, the solution is, as it always has been, to innovate.

This talk will focus on several compelling vectors of innovation that will enable our industry to thrive within the limits of physics and silicon scaling. The first is around extraction of more value from each manufactured transistor through modular chiplet architectures that enable application-specific choice of process technology and market-specific personalization of products. The second opportunity extracts more performance from each yielded transistor through targeted use of special purpose accelerators that provide dramatic efficiency gains over general purpose compute. Finally, there is the growth in applications of AI to all aspects of computing, making use of reduced precision accelerators that add non-linear value to compute capabilities and user experience in all key market segments. Putting all these together with the right software solutions will provide synergistic improvements in compute capability that can enable a thriving industry for the foreseeable future.