After losing Huawei, TSMC teamed up with Tesla to specialize in self-driving chips

Although TSMC lost Huawei as a major customer, it received a large order from Tesla.

According to Taiwan’s “Business Times” report, the new high-performance computing (HPC) chip jointly developed by the global IC design leader Broadcom and Tesla will be produced with TSMC’s 7nm advanced process, and TSMC’s integrated fan-out (InFO) chip will be used for the first time. ) system single wafer advanced packaging technology (SoW), each 12-inch wafer can cut about 25 chips. The chip will start production in the fourth quarter of this year, with an initial production of about 2,000 wafers, and is expected to enter the full mass production stage in the fourth quarter of next year.

At the same time, the report pointed out that the HPC chip built by Broadcom for Tesla will become the core computing special application chip (ASIC) of future Tesla electric vehicles, which can be used to control and support advanced driver assistance systems, electric vehicle power transmission , car entertainment system, and car body Electronic components and other four major application areas of automotive electronics, and further support the real-time computing required for self-driving, so as to achieve the goal of “completely autonomous driving” in the true sense.

In this regard, 21ic guesses that the ASIC chip produced by TSMC this time should be the Hardware 4.0 version!

ASIC chips are expected to become the darling of the industry

When it comes to ASIC chips, many people may feel unfamiliar. By definition, an ASIC chip refers to an integrated circuit designed and manufactured in response to specific user requirements or specific electronic system needs.

From the current stage, the mainstream chips in the field of autonomous driving mainly include CPU, GPU, FPGA, ASIC and so on. Among them, GPU, CPU, and FPGA belong to general-purpose chips; while ASIC belongs to special-purpose chips.

In order to let everyone understand the characteristics of these chips more intuitively, 21ic has briefly sorted them out, as follows:

◆ GPU: Single-instruction, multi-data processing, using a large number of computing units and super-long pipelines, just like the name, graphics processor, GPU is good at processing the computing acceleration in the image field. But the GPU cannot work alone, it must be controlled by the CPU to make it work.

◆ CPU: can act independently to handle complex logic operations and different data types. However, when a large amount of data of a uniform type is required to be processed, the GPU can be called for parallel computing.

◆ FPGA: Contrary to GPU, FPGA is suitable for analysis of multiple instructions and single data stream, so it is often used in the prediction stage, such as the cloud. FPGA uses hardware to implement software algorithms, so it is difficult to implement complex algorithms. The disadvantage is that the price is relatively high.

◆ ASIC: A dedicated custom chip, a chip customized to achieve specific requirements. Compared with general-purpose chips such as GPUs, CPUs, and FPGAs, ASICs are smaller in size, lighter in weight, lower in power consumption, more reliable, and have higher computing performance and efficiency. But the disadvantage is that the development cycle is long and the initial cost is high.

In view of the above comparison, 21ic experts believe that with the rapid rise of artificial intelligence applications in recent years, people’s requirements for information processing capabilities are getting higher and higher. Therefore, it is foreseeable that heterogeneous computing will occupy a larger share in future data centers. In the future, more and more car companies will be more inclined to choose ASIC-specific chips.

The future may redefine autonomous driving

After introducing ASIC chips, let’s talk about autonomous driving technology.

In the capital market, Tesla, which has been established for only 17 years, has subverted a number of traditional car companies for a century, and sat on the throne of the world’s largest car company by market value, relying on autonomous driving technology.

From “buying” chips to “making” chips, in fact, Tesla has accumulated many years in the field of autonomous driving – as early as 2014, it launched the first generation of autonomous driving products with Mobileye; The team of chip architects led by legendary chip designer Jim Keller is working on developing its own AI chip, with the goal of designing a powerful and highly efficient self-driving chip.

After three years of silence, Tesla finally officially launched the chip, Hardware 3.0, in April 2019, making it the world’s first vehicle company that can develop chips.

It is reported that this Hardware 3.0 chip uses a 14nm process technology, and the foundry is South Korea’s Samsung. Although compared with the previous generation of autonomous driving hardware driven by NVIDIA hardware, the number of frames per second has increased by 21 times, but the power consumption The amount has hardly increased. At that time, Tesla CEO Elon Musk publicly stated that although Hardware 3.0 will be revolutionary, Tesla is already developing the next-generation chip, and I hope it is 3 times better than this chip!

It is worth mentioning that at the opening ceremony of this year’s World Artificial Intelligence Conference, Musk said: “We have developed a special chip for Tesla, which is equipped with a dual-system fully autonomous driving computer, with 8-bit and accelerator functions. In click computing, artificial intelligence involves a lot of click operations, and if you know what click operations are, you know that the huge amount of click operations means that our computers have to do a lot of click operations. In fact, we carefully turned on the chip a few months ago. The second system, taking full advantage of Tesla’s full self-driving system, may be at least a year away.”

At the same time, he said that Tesla’s current research and development progress is very close to Level 5 autonomous driving technology, which means that in the future Model 3, Y, S, and X can drive themselves without the driver’s attention at all. In the future, Tesla will also develop a Dojo training system that can quickly process large amounts of video data to improve the training of artificial intelligence systems…

Although Musk’s remarks are very “attractive”, Tesla has also been accused of false propaganda due to the frequent occurrence of self-driving safety accidents.

So, here’s the question: Is Tesla’s “full self-driving” really going to happen? What kind of autopilot chip can Tesla develop this time with TSMC, the “king of chips”? Let’s wait and see!

The Links:   SKKT330-14E   2MBI200TA-060