Li Auto Becomes First Chinese Automaker to Debut at ISCA, Unveils and Mass-Produces M100 Chip

Li Auto recently presented a paper titled "Mach M100: An Orchestrated Dataflow Architecture for General-Purpose AI Computing" and delivered a keynote speech at the International Symposium on Computer Architecture (ISCA) 2026. ISCA is recognized alongside MICRO, HPCA, and ASPLOS as one of the four premier global academic conferences in computer architecture. Since ISCA introduced its industry track in 2020, Li Auto has become the first Chinese automotive company to be selected for participation. The Mach M100 is a high-performance edge inference chip based on a dataflow architecture, fabricated using a 5nm automotive-grade process node. It delivers 1,280 TOPS of compute power with an impressive utilization rate of 82%. The chip features 8-channel LPDDR5X memory with a peak bandwidth of 273 GB/s and integrates a 24-core ARM Cortex-A78AE CPU cluster. Its architecture is built around five core principles: a dataflow execution model, a cache-less memory hierarchy, a tensor-granularity instruction set architecture (ISA), balanced hardware-software complexity, and a hierarchical tile-based interconnect structure. Additionally, the chip employs a fully redundant design with dual SoCs, dual MCUs, and dual power supplies, meeting the ASIL-D functional safety standard to ensure uninterrupted system operation even in the event of any single-point failure. Developed entirely in-house by Li Auto, the Mach M100 project began with feasibility studies in 2021 and was officially launched in 2022 with the goal of building a native AI chip architecture from first principles. Over the following four years, a dedicated team of 200 engineers completed end-to-end independent development—from architectural design and compiler development to thermal management systems and functional safety implementation. The Mach M100 is now in pre-production vehicle integration and has been deployed in the all-new Li L9 and L8 models.