Nvidia Presents New AI Research to Optimize Chip Design
On Monday, Nvidia Corporation unveiled its new research on how artificial intelligence (AI) can be used to enhance chip design. The chip design process involves placing tens of billions of tiny on-off switches called transistors on a silicon piece to make functional chips. The accurate positioning of transistors has a significant impact on the cost, speed, and power consumption of a chip. Complex design software, such as those provided by Synopsys Inc and Cadence Design Systems Inc, are used to optimize transistor placement.
Nvidia’s new paper revealed how AI techniques can be used to find better ways of placing big groups of transistors. The research aimed to improve on Google’s 2021 paper, which later became a subject of controversy. Nvidia added a second layer of AI to an existing effort developed by University of Texas researchers, who used reinforcement learning to achieve better results.
According to Nvidia’s Chief Scientist Bill Dally, the research is essential because per-transistor costs of new generations of chip manufacturing technology are now higher than in previous generations. This goes against Intel Corp co-founder Gordon Moore’s prediction that chips would always get cheaper and faster. Dally emphasized that to continue delivering more value to customers, cleverer chip design, instead of cheaper transistors, is necessary.