DIGITIMES: Physical AI reshapes automotive SoC design

The DIGITIMES report places the shift toward end-to-end autonomous driving alongside a broader change in how AI computing itself is organized. It describes global industry consensus forming around three stages of AI evolution, with hardware requirements differing at each stage. In the Agentic AI stage, systems handle not only inference but also auxiliary tasks like task scheduling and tool calling, leaning on the compute strengths of the CPU. As the industry looks ahead to Physical AI, systems must perform real-time, accurate inference in diverse and changing physical environments.

The report frames that transition as the reason automotive SoC design is changing. It lays out a 3x3 matrix of key indicators spanning underlying hardware specifications, mid-layer efficiency performance, and upper-layer solution compatibility, and builds a set of reference architectures and specifications around it. The discussion spans the end-to-end autonomous driving architecture and its technology evolution, FSD chip architecture and upgrade specifications, and design references for the vendors it names.

The stakes appear to sit with the chip designers and platform teams deciding what next-generation vehicles run on. Whether the report's framework becomes a practical guide hinges on how closely vendors and their customers align on the indicators it proposes, and on whether the AI accelerator indeed takes the central role DIGITIMES expects.