News

[News] Taiwan’s Alchip Adjusts Chip Design for Chinese Automakers to Meet U.S. Standards


2025-01-21 Semiconductors editor

According to a report from Commercial Times, sources indicate that due to U.S. restrictions on AI computing power exports to China, the application for an ADAS (Advanced Driver Assistance System) chip by Taiwanese ASIC manufacturer Alchip was rejected.

The ADAS chip developed by Li Auto, named “Schumacher,” was co-developed with Alchip and originally planned for production using TSMC’s 5nm process. The chip was scheduled for tape-out at the end of last year, but the application was denied, and the project is now undergoing redesign, as stated in the report.

The report highlights that automotive chip manufacturers believe that, under the current trends for end-to-end large models, automotive intelligent driving chips with approximately 800 TOPs of computing power are considered industry-leading. It is estimated that excessively high computing power is unnecessary, with the focus instead placed on computing power requirements for the back-end training.

The report further points out that companies such as NVIDIA and Qualcomm offer autonomous driving solutions. However, to avoid potential supply chain issues, such as shortages or sanctions, major companies centered around intelligent driving, including NIO and XPeng Motors, are actively developing their own chips.

Alchip’s Focus on Co-Packaged Optics (CPO)

As noted by the report, supplier selection for AWS Trainium 4 has commenced. AWS is reportedly considering the use of SoIC (System on Integrated Chips) and CPO (Co-Packaged Optics) architectures for its Trainium 4 ASIC.

Notably, Alchip is developing related chip designs for AWS and has successfully completed the design finalization of a 2nm SoIC chip, which could position the company to secure next-generation chip orders from major clients, as the report indicates.

The report underscores that, from the perspective of major CSP (Cloud Service Provider) companies, CPO is expected to become a central focus for AI ASICs used in training and inference tasks. Similarly, general-purpose GPU manufacturers are increasingly focusing on silicon photonics, aligning with this emerging trend.

Read more

(Photo credit: Li Auto)

Please note that this article cites information from Commercial Times.

Get in touch with us