HC2023-K1: Exciting Directions for ML Models and the Implications for Computing Hardware

"During the session, the following key points were discussed:1. Optical computing can significantly improve computation speed, reducing the number of chips required for the same task, leading to better performance and energy efficiency.2. There may be a trend towards using single models to handle different workloads, and interconnects will become more important.3. New developments in Network on Chip (NoC) technology are gaining significance, with advancements in optical on-chip technology potentially leading to breakthroughs in the future.4. Machine Learning (ML) for Electronic Design Automation (EDA) has the potential to automate different phases of the design process, but there is still work to be done in integrating all these pieces into a more automated workflow with humans in the loop.5. Liquid cooling can improve performance and reliability, but it also adds complexity and deployment time from design to production, which can be challenging.6. The physical footprint of liquid cooling systems needs to be transformed across the world, which has been a challenge for many years.7. The speaker believes that there is potential in designing specialized chips for accelerating neural network inference, which is a much narrower problem than designing a single chip for the entire world."