Waymo Builds Custom ASIC Chip for Robotaxi Compute
Waymo is moving from off-the-shelf processors to custom silicon, betting that in-house design will let it optimize compute for its own operating conditions.
Reporting from 1 source: GIGAZINE.
Waymo has developed a custom ASIC chip for its robotaxis, built on TSMC's 5nm process. The machine learning focused chip processes raw sensor data in real time and runs advanced neural networks, delivering more than 1,000 TOPS of performance. Each robotaxi carries two chips for redundancy. Waymo will present the chip at the Hot Chips conference starting August 23.
Waymo has been using off-the-shelf processors in its robotaxis until now. The new chip, built on TSMC's 5nm process, is a machine learning processor designed to process raw sensor data in real time and run advanced neural networks. It delivers more than 1,000 TOPS of ML performance, and Waymo tuned it to maximize throughput in the low-batch environments where its operations run most often.
Waymo says the chip is one of several custom components it is developing, and it has reworked both the physical and architectural design to use it. Each robotaxi carries two of the chips for redundancy, so a hardware failure does not put passengers at risk. Waymo is to present the chip at the Hot Chips conference, which opens August 23.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.
Sources
- GIGAZINE Waymoがロボタクシーの性能向上のためカスタムASICチップを開発