Moore’s law has helped us for 5+ decades through monolithic integration with doubling of transistors every two years. With the introduction of deep learning in 2012, compute demand has been out pacing Moore’s law. Progress in AI over the last decade has been hardware driven with large scale models exploiting the large transistor connectivity available. More recently there appears to be a shift towards developing software efficiencies, along with hardware advances. Nevertheless, the need for more and more transistors will never cease, and hence continuation of Moore’s law becomes a necessity. Heterogenous Integration (HI) is taking the front seat to continue Moore’s law.

But is HI ready to take on this challenge for the next decade? Are we geared towards saving our planet from an energy crisis? What are some of the key technologies that we need to focus on for HI? This presentation will address some of these questions by focusing on challenges as well as recent progress.