Waymo details custom chip powering its autonomous taxis
Waymo revealed new details about its custom chip for robotaxis, a dual-ASIC system capable of 1,000 trillion calculations per second.
Waymo operates about 4,000 vehicles in 11 U.S. cities, most of them Jaguar I-Pace crossovers, with the rest being Zeeker minivans and Hyundai Ioniq 5 SUVs. Riders order trips through a standalone app and Uber.
In a blog post, Waymo executives Satish Jeyachandran and Daniel Rosenband said the stack is engineered for ultra-low latency, minimizing the delay from first pixel to action. The chips are manufactured by Taiwan Semiconductor Manufacturing Co. using its five-nanometer node, specifically the N5A automotive-optimized edition, which includes optimizations to reduce manufacturing fault risks and complies with three auto industry reliability standards. Waymo included two ASICs to mitigate hardware failures; if one goes offline, the other takes over.
The chips process data from a sensor suite of 13 cameras, four lidar devices and six radars. Cameras capture fine-grained details such as text on street signs, while lidar works more reliably in low-light conditions, and radars collect data such as nearby vehicles' speed. The ASICs are optimized for AI inference and run models for sensor fusion, combining readings from different sensor types into a single dataset while adjusting for different measurement angles. Sensor fusion also performs error correction using temporal denoising, which verifies whether a frame contains noise by checking adjacent frames.
In addition to the ASICs, the module includes central processing units and graphics processing units, housed in a ruggedized enclosure designed for challenging conditions. The system dissipates heat with the host vehicle's battery liquid cooling. The executives wrote that the hardware operates under constant vibration, shock and extreme temperatures, and is engineered for endurance from the component to system level.