Transfer Tesla HW4.0 autopilot chip to hitchhiker to accumulate 5nm

According to a number of Taiwan media reports, Tesla's HW4.0 self-driving chip will be built using TSMC's 4nm or 5nm process. This means that compared to the 14nm HW3.0, the performance of the new generation of chips will achieve a significant leap. At present, TSMC is also responsible for the production of some Tesla chips, and if it takes over Samsung's OEM orders in the future, its relationship with Tesla is naturally not the same.

First of all, let's briefly review Tesla's partner change history:

In 2019, Tesla officially broke up with NVIDIA, released self-developed Hardware 3.0 hardware and said that "this is the best chip in the world";

Ø In 2020, it was reported that Tesla is cooperating with Broadcom to develop Hardware 4.0 chips, which are expected to use TSMC's 7nm process and be fully mass-produced in the fourth quarter of 2021;

Ø In 2021, Tesla confirmed that the self-driving chip will continue to be handed over to Samsung, but it is not HW4.0;

Ø In 2022, the supply chain of Taiwan, China, reported that the outsourcing of Tesla's HW4.0 intelligent driving chip will be transferred to TSMC 7nm; and the latest news shows that HW4.0 hardware will be built using TSMC's 4nm/5nm process.

It is reported that in order to work closely with the supply chain in Taiwan, Tesla plans to expand the size of its design team in Taiwan.

For Tesla, the competition on the autonomous driving track is intense, and it may be imperative for HW4.0 chips to adopt more advanced processes; and for TSMC, new orders are a big cake, but also have special significance.

In 2021, Tesla delivered more than 936,000 vehicles worldwide, a year-on-year increase of about 87.4%. In the first three quarters of this year, Tesla has delivered more than 900,000 vehicles. According to internal planning, Tesla's annual delivery target is raised by 50%, and the delivery volume is expected to be about 1.4 million vehicles this year and may exceed 2 million next year.

Although it is not clear whether TSMC will replace Samsung 100%, HW4.0 chip shipments are as high as millions. Therefore, some analysts pointed out that Tesla is expected to become TSMC's top seven customers next year.

In 2021, TSMC's head customer information was exposed, which were Apple, MediaTek, AMD, Qualcomm, Broadcom, NVIDIA, Sony and other companies. Taking last year's order amount as an example, Apple's proportion accounted for 20%, about 11.3 billion US dollars. The seventh-largest customer is expected to account for about 2% of revenue, involving up to $1.1 billion.

More importantly, TSMC's previous large customers are generally concentrated in technology giants such as Apple, chip design companies and IDM manufacturers, and the emergence of Tesla represents that its head customers have begun to appear in car companies, more precisely, new energy vehicle companies.

At present, the growth rate of consumer electronics market represented by smart phones has slowed down, the development of electric vehicles and autonomous driving industries has pressed the accelerator button, and chip design companies and even foundries are looking forward to this new red sea.

Intel CEO Henry Kissinger previously predicted that by 2030, the automotive chip market is expected to reach $115 billion, accounting for about 11% of the entire chip market. Combined with Tesla's goal of producing 20 million vehicles a year in 2030, TSMC's foundry contract for HW4.0 chips may become a prelude to the influx of orders from other car companies.

And Tesla is not unreasonable to find TSMC.

At the NVIDIA GTC 2022 conference held in September, Huang Jenxun casually dropped a lethal "bomb" - Thor. Compared with the earliest announced product route Atlan, Thor's single computing power has doubled to 2000TOPS, close to 8 times that of the existing Orin. In addition, Thor is almost born for a central computing architecture, which can control the cockpit and the intelligent driving domain at the same time.

At that time, Huang explained that the three key components of Atlan, CPU, GPU and Tensor Core have all received huge updates in the past two years. Therefore, NVIDIA decided to skip Atlan and release Thor directly. Having said that, it has to be admitted that the autonomous driving track has become more and more volatile, and both consumers and car companies are no longer satisfied with the current demand.

According to the "China New Car Purchase Intention Research Report" released by J.D. Power last year, nearly a quarter of consumers who have the intention to buy a car within six months regard intelligent experience as the most important factor in car purchase decision. IDC also pointed out in a report released in May this year that with the increasing maturity of L2 autonomous driving in the Chinese market, consumers' acceptance of autonomous driving has increased with the use rate. Autonomous driving, or driver assistance functions, is becoming an important consideration for consumers when buying a car.

From the perspective of car companies, meeting the needs of consumers and complying with market trends can be long-lasting.

Although there are various indications that Tesla's HW4.0 hardware will use 7nm process, Tesla chose to keep up with NVIDIA in just one year. The latter did not announce the process node chosen by Thor, and it is speculated that it could be TSMC's 4/5nm. In addition, Qualcomm also announced as early as the beginning of last year that the core SoC of the Snapdragon Ride autonomous driving platform will use a 5nm process. Even if Tesla has stronger algorithm software, 7nm still pales in comparison to 5nm.

As for the choice of partners, Tesla continues to find Samsung foundry is nothing more than familiarity with each other, lower foundry prices and other reasons, on the contrary, 5nm is more critical to the yield problem. Earlier, Samsung was exposed to 4/5nm yield fraud, causing companies such as NVIDIA to switch to a single TSMC. Perhaps, HW4.0 upgraded to 4/5nm, Tesla bet on TSMC also has this consideration.

Musk has repeatedly stated that autonomous driving, while not perfect, will be 10 times safer than human driving. This is the key to why the market expects HW4.0 or fully autonomous driving computer version 2.0. But as autonomous SoCs move toward more advanced processes, who will be the winner? We'll see.