Presentation on ML + IC Design
Date:
This talk tries to bridge the areas of Machine Learning and Chip Design.
I gave a presentation on the usage of Machine Learning in advancement of Integrated Circuit Design and how AI hardware can be significantly made better using specialized Circuit Design techniques.
Modern day AI systems are power hungry, this motivates the need for designing circuits that work highly efficiently. This has drawn engineers to take inspiration from the human brain, thus leading to the development of Neuromorphic computing. I collaborated with Prof. Anjan Chakravorty and Prof. Avhishek Chatterjee to estimate the energy consumed by data centers.
Circuit Design is extremely time intensive. Taping out a chip is often a year long endeavour. With the right use of Machine Learning, it is possible to significantly cut down on this design time. In my work with Prof. Sankaran Aniruddhan, I developed an algorithm to speed up the schematic level design of RF circuits.
This talk tries to answer the question of how chip design can make AI better and how AI can make chip design easier taking ideas from my works described above.