TSMC Unveils AI Strategy to Boost Chip Energy Efficiency

Taiwan Semiconductor Manufacturing Co unveiled a strategy at a Silicon Valley conference to increase the energy efficiency of AI computing chips by approximately 10 times. The approach utilizes AI-powered design software to create a new generation of chips composed of multiple chiplets, which are smaller components using different technologies packaged together.

To implement these designs, the company coordinated with software providers Cadence Design Systems and Synopsys, both of which released new products on Wednesday. TSMC reported that these AI tools solve complex design tasks faster and more effectively than human engineers, reducing some workloads from two days to five minutes.

The initiative addresses the high power demands of AI hardware, noting that flagship AI servers from Nvidia can consume up to 1,200 watts during demanding tasks. During the event, an engineer from Meta Platforms delivered a keynote address regarding the physical limits of chip manufacturing.