Tesla has been steadily ramping up the Full Self-Driving (FSD) capability of its bespoke Advanced Driver-Assistance System (ADAS), dubbed the Autopilot, and the FSD chips play a key role in this regard. Now, if the latest reporting on this matter pans out, it seems that TSMC is about to snatch away Samsung’s Tesla punch bowl. As a refresher, Samsung was responsible for manufacturing Tesla’s FSD 3.0 chips on its 14nm architecture. Moreover, the EV giant is expected to continue to rely on Samsung’s fabs for its FSD 4.0 chips that are to be built on 7nm nodes. However, for future chips that would rely on sub-7nm architecture, Tesla appears to be pivoting away from Samsung and toward TSMC, as per DigiTimes’ reporting. As evidence, DigiTimes cited the presence of Peter Bannon, Tesla’s VP of low voltage electronics, at TSMC’s recent Technology Symposium, which showcased the Taiwanese chip manufacturer’s prowess in industry-leading technologies, including its 3DFabric advanced system integration service. Bear in mind that TSMC is already cooperating with a number of automakers, including Volkswagen. Automotive chips are likely to constitute one of the fab giant’s strongest growth engines. For the uninitiated, Tesla introduced the vision-based iteration of its Autopilot with much fanfare a while back. The rationale here is that with eight high-resolution cameras and a high-tech neural network to interpret the visual cues, the Autopilot would mimic the way humans make decisions on the road. However, as the regulatory scrutiny around the Autopilot system seems to have grown lately, there are indications that Tesla might incorporate a radar into its vision-based ADAS as another layer of contingency in adverse weather conditions, where cameras often fail to function properly. During its AI Day 2022 event, Tesla announced that the Full Self-Driving (Beta) capability of its Autopilot system now has 160,000 customers as opposed to just 2,000 back in 2021. Last year, Tesla unveiled the 7nm-based D1 chip to power its in-house Dojo supercomputer, which the company uses to train its Autopilot neural network by feeding visual clips captured by millions of Tesla vehicles already on the road. The EV company revealed quite a lot of details about its Dojo supercomputer at the recent Chips 34 event. Tesla plans to build its first Dojo exapod in 2023. In total, the company plans to build 7 such exapods to speed up the training of its Autopilot neural network. Meanwhile, Tesla shares are under pressure today for missing analyst expectations regarding its Q3 deliveries. Tesla produced 365,923 vehicles and delivered 343,830 vehicles in Q3 2022. While the production number came in higher than Bloomberg’s estimate of 359,853, the deliveries missed the consensus number of 357,938. Tesla cited logistical constraints for falling short on its quarterly deliveries.
