Increasing Efficiency through AI: TSMC Relies on Innovative Chip Designs

The continuously increasing demands for computing power and energy efficiency of artificial intelligence are driving developments in the chip industry in new directions. Taiwan Semiconductor Manufacturing Co (TSMC), the world's largest manufacturer of computer chips, presented an ambitious plan to improve the energy efficiency of these essential components at a conference in Silicon Valley. By using AI-driven software for chip manufacturing, TSMC hopes to increase the energy efficiency of its products tenfold. At Nvidia, a leading provider of AI servers, the current energy consumption during demanding tasks is up to 1,200 watts, equivalent to the continuous power supply of 1,000 US households. TSMC aims for significant energy savings through innovative chip designs, where multiple small "chiplets" of various technical concepts are combined into a single powerful package. This revolution in chip manufacturing is accompanied by AI trends, increasingly incorporating software tools from companies like Cadence Design Systems and Synopsys. Both companies introduced new products simultaneously with TSMC, developed in close dialogue with the chip manufacturer. The software allows complex design tasks to be completed in minutes, whereas human engineers previously required two days. "This significantly maximizes the potential of our technology," explained Jim Chang, Deputy Director of the 3DIC Methodology Group at TSMC, during his presentation. The conventional manufacturing methods so far are increasingly reaching physical limits, especially in the transmission of data over electrical connections in chips. Kaushik Veeraraghavan of Meta Platforms emphasized in his keynote that this is not just an engineering, but a fundamental physical problem. The challenge is to make new technologies like optical connections reliable in the practice of chips massively used in data centers.