TSMC has reportedly started producing next-generation semiconductor chips to be installed in Tesla's AI supercomputer 'Dojo' for the American electric vehicle company.
On the 5th, Taiwanese media such as Liberty Times cited sources reporting that TSMC revealed this fact recently at a technology conference held in Santa Clara, California, USA.
Dojo is an AI supercomputer designed to process data and video materials collected by Tesla vehicles to train autonomous driving software.
Sources explained that TSMC has begun production of the next-generation training modules for Tesla's Dojo and will provide a complex system with more than 40 times the current computing performance by 2027 using an advanced packaging process called 'Chip on Wafer on Substrate' (CoWoS). They also mentioned that Tesla's new product differs from the system supplied to the American startup Cerebras.
Another source revealed that TSMC plans to create a wafer system integrating CoWoS advanced packaging technology and SoIC (System On Integrated Chips) advanced packaging by 2027.
The integrated wafer is said to offer 40 times the computing power, more than 40 photomasks, and space for over 60 high-bandwidth memory (HBM) units.
The next-generation Dojo is scheduled to be installed in New York, USA. Tesla plans to build a 100MW-scale data center at its Austin Gigafactory in Texas, where its headquarters are located, to train autonomous driving software, with hardware supplied by Nvidia.
Earlier, TSMC announced at a technology conference held recently in Santa Clara, California, that it plans to start semiconductor production using a 1.6-nanometer process from the second half of 2026.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
