Li Auto CTO Details Mach M100 AI Chip at TSMC OIP Forum
Li Auto CTO Xie Yan delivered a keynote at the TSMC OIP Forum detailing the development and mass production of the Mach M100, a 5nm dynamic dataflow AI chip. This marks the first time an enterprise from mainland China has given a keynote speech at the event. The Mach M100 represents a major architectural shift away from traditional von Neumann computing toward data-driven edge inference for automotive systems. It enables automakers to handle both heavy autonomous driving algorithms and on-board large language models efficiently within strict power constraints. Built on a 5nm automotive-grade process, the Mach M100 delivers 1,280 TOPS of computing power per chip and features a dynamic dataflow architecture that prioritizes data-driven execution over control-flow overhead. It is already mass-produced and deployed across several new Li Auto vehicle models.
## BACKGROUND
The TSMC Open Innovation Platform (OIP) is a comprehensive design technology infrastructure that encompasses critical IC implementation areas to accelerate first-time silicon success. Dynamic dataflow architecture is a computer architecture where the execution of instructions is driven directly by the availability of input data, offering massive parallelism for AI workloads compared to traditional control-flow designs.