Tesla reportedly taps TSMC to produce new CPU for autonomous driving — 3nm next-gen CPU may feature N3P process node

(Image credit: Tesla)

Tesla is set to use TSMC's N3P (performance-enhanced 3nm-class) fabrication process to build one of its next-generation Full Self-Driving (FSD) hardware, according to a rumor published by China Times and noticed by analyst Dan Nystedt. If the information is accurate, Tesla plans to use a non-automotive grade production node for its FSD system-on-chip (SoC) or system-in-package (SiP).

TSMC's N3P process is a cutting-edge manufacturing technology that offers high performance, high transistor density, and relatively lower power. For demanding processors, such as Tesla's FSD hardware, N3P could be just what the doctor ordered. Several companies are reportedly committed to using production nodes, so Tesla will not be alone. But there is a catch about N3P and FSD chips: when exactly does Tesla plan to use this technology?

A previous report indicated that Tesla planned to use TSMC's N4 technology to build its Full Self-Driving Hardware 4 system-on-chip in Arizona sometime in 2024. Meanwhile, since TSMC had to delay production at its Fab 21 to 2025, Tesla either decided to port its FSD 4 design to N3P and make it in Taiwan or slow production of FSD 4 hardware in the U.S. by six or more months. Perhaps the company is also prepping an FSD 4.5 or even an FSD 5.0 platform that will use N3P, but this is speculation.

Since the information about Tesla's usage of TSMC's N4 or N3P technology comes from an unofficial source, it should be taken with a grain of salt. Meanwhile, it should be noted that neither N4 nor N3P are aimed at automotive applications. Instead, TSMC is set to offer N5A and N3A technologies designed for automobiles.

TSMC's N5A process technology has been qualified for the AEC-Q100 Grade 1 standards and is known for its rigorous chip quality and reliability criteria. SoCs compliant with AEC-Q100 Grade 1 must function in extreme temperatures ranging from -40 to +150 degrees Celsius and undergo numerous tests for reliability and durability to ensure long-term performance under harsh conditions. Additionally, N5A adheres to the requirements of ISO 26262 for functional safety and IATF16949 for quality management, defect prevention, and variation reduction. TSMC's N3A is set to follow in N5A's footsteps regarding quality, reliability, and longevity.

To enhance its N3 and N5 production nodes for automotive-grade applications, TSMC had to modify both technologies, so it is impossible to use N3 or N5 IP for N3A and N5A technologies, which is why it would be strange for Tesla to design FSD SoCs or SiPs on non-automotive-grade nodes if it plans to adopt proper nodes later on.

Anton Shilov
Freelance News Writer

Anton Shilov is a Freelance News Writer at Tom’s Hardware US. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

  • kjfatl
    Using a non automotive node is a viable option. For cold environments, the obvious solution is to provide a heater that keeps the computer at the minimum safe operating temperature. If it is -40C outside, a small amount of power can be used to keep the computer warm. Worst case, if the system was in cold shutdown at -40C, would be a warmup time in the 5 minute range before FSD was available. It probably takes longer for the heaters on the cameras to defrost the lens covers.
    On the high end, various cooling approaches can be use to keep the chip cooled.
    Automotive grade parts are needed in an open engine compartment or tail light assembly, but not for the FSD computer.