Alphabet-owned autonomous driving company Waymo has revealed details of a custom 5-nanometre chip and computing architecture designed to process the large volumes of sensor data generated by its self-driving vehicles in real time.
The purpose-built application-specific integrated circuit, or ASIC, forms part of Waymo's sixth-generation autonomous driving system. The chip is designed to process and fuse raw information from cameras, lidar and radar before the data reaches the vehicle's core machine-learning system, helping the Waymo Driver make decisions within milliseconds.
Waymo said its custom accelerators can deliver more than 1,000 trillion operations per second, or over 1,000 TOPS, of machine-learning performance. The architecture is intended to handle the computational requirements of autonomous driving while maintaining low latency as vehicles continuously interpret their surroundings.
The custom chip does not mean Waymo is moving entirely away from external semiconductor suppliers. Its computing system combines internally designed technology with components from AMD, Micron, Nvidia, Samsung, SanDisk, Socionext and TSMC. The approach gives Waymo greater control over specialised sensor processing while retaining third-party CPUs, GPUs, memory, storage and other accelerators.
Waymo has increasingly moved processing complexity into custom silicon as it develops its autonomous driving hardware. Its sixth-generation Waymo Driver uses high-resolution cameras, imaging radar and lidar as a unified sensing system. The company said in February that the latest generation uses fewer than half the number of cameras found in its fifth-generation system while delivering improved capabilities and reducing costs.
The computing platform has also been designed around redundancy. Autonomous vehicles need to continue processing information even when individual components encounter problems, making backup systems an important part of the architecture. Waymo has focused the system around responsiveness, ruggedness and redundancy to support real-time driving operations.
The hardware disclosure comes as Waymo expands its commercial robotaxi operations. The company said earlier this year that the sixth-generation Waymo Driver was designed to support multiple vehicle platforms and larger-scale production. Its autonomous driving technology has accumulated nearly 200 million fully autonomous miles across more than 10 major cities.
Waymo is also adapting the sixth-generation system to vehicles including the Hyundai IONIQ 5. In July, the company said it had begun autonomously driving the model with an autonomous specialist present while preparing fully autonomous operations in Denver, Las Vegas, San Diego and Tampa.
The custom ASIC gives Waymo another internally developed component within its autonomous driving stack as it works to balance computing performance, power efficiency, hardware costs and the processing demands of AI-powered self-driving systems.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.