Beyond the silicon horizon: breaking barriers in AI performance and efficiency

AI is the most important technology innovation of the last 50 years. Delivering future generations of highly-performant and energy-efficient systems will be critical to applying the power of AI to help solve the world’s most important challenges. The required technology will span silicon architecture, advanced packaging, memory design, networking and the fundamental coupling of hardware and software to power future inferencing and training applications. 

Silicon architecture has moved beyond monolithic chips and into 3.5D design built for petaflops of compute packed into dense modules with high bandwidth memory. The most advanced AI processors are pushing the boundaries of design tools, manufacturability, power delivery and cooling. At the same time, connecting thousands of AI processors at scale to enable maximum compute performance will require new innovations based on co-designing hardware and software as a solution. This presentation will explore developments and opportunities to chart a path for building the next generation of AI infrastructure.