The crisis of car companies lacking cores has been raging for more than a year, when will it be able to pass? There are still many opinions in the industry. But what chips are missing in car companies? Why is there a lack of cores? What are the classifications of car chips? In 2021, when the core shortage crisis is serious, what happened to the automotive chip industry? "Looking Back at 2021" specially planned to focus on chips today, taking you to read all the things about car chips.
Wen 丨 wisdom driving network Huang Huadan
2022 has passed more than a month, and the dilemma of car companies lacking cores is far from being broken.
When will the crisis that has plagued the global auto market for a long time be completely resolved?
The head of the semiconductor business unit of Infineon said that it is not expected to return to normal until the end of 2022, and the shortage should be solved by 2023.
Similarly, in a conference call after Tesla's fourth-quarter earnings release, Musk also said that it expects to solve the chip shortage problem from 2022 to 2023.
2022 may still be unusually tough, but at least, hope is already in sight.
The industry generally believes that the lack of cores is due to the epidemic and supply chain problems, as well as the inability of semiconductor suppliers to increase production capacity.
But industry analyst firm IC Insights believes that in addition to the epidemic and supply chain problems, the real reason for the shortage is the surge in demand for automotive chips in 2021.
On the capacity side, according to IC Insights, semiconductor suppliers' shipments to the automotive industry increased by 30% in 2021 compared to 2020, which is much higher than the 22% increase in total global semiconductor shipments last year.
This aspect stems from the fact that in 2021, the auto market finally ended its reverse growth, and sales began to rise.
On the other hand, due to the fact that most of the current automotive innovations are based on microelectronics technology (semiconductors), especially the acceleration of electrification and autonomous driving, the demand for chips in the industry has increased greatly.
What core is most lacking in car companies?
There are many types of car chips. So, in the wave of core shortages that have been boiling over the past two years, what is the core that car companies lack the most?
Before answering this question, let's get to know a new term: MCU.
The full name of the MCU is Micro Control Unit, that is, the microcontroller, and the list of chips. In layman's terms, it can be understood as a microcomputer.
MCU is to reduce the frequency and specifications of the central processor appropriately, and integrate memory, arithmetic, timers, interfaces, etc. on a single chip to form a chip-level computer and make different combinations of control for different applications.
With the acceleration of the intelligentization of the car, the whole vehicle will be equipped with more functions, and a large number of actuators need to be controlled by the MCU.
From anti-lock braking system, four-wheel drive system, electronically controlled automatic transmission, active suspension system, to the current gradually extended to the body of various safety, network, entertainment control systems and other fields.
Nowadays, a traditional fuel vehicle needs to use dozens to hundreds of MCUs, and the use of new energy vehicles is doubled.
Such a large amount of MCU chips is the protagonist of the lack of cores in car companies in the past two years.
Moreover, the demand for MCUs is constantly increasing.
ACCORDING TO IC Insights, the global automotive grade MCU market is expected to grow from $6.5 billion in 2020 to $8.8 billion by 2023.
From the perspective of the pattern, the automotive MCU market is concentrated in several leading enterprises.
The top five global automotive MCUs market share in 2020 are Renesas, NXP, Infineon, Texas Instruments and Microchip, accounting for a total of 87%.
In the face of the two-year crisis of lack of cores, the head companies are responding quickly to expand production capacity.
Infineon's 12-inch (300mm) fab, which will be built with an investment of 1.6 billion euros in 2021, is officially put into use.
Infineon currently has two large 300 mm thin wafer plants for the production of power semiconductor devices in Dresden and Villach.
TSMC, the foundry overlord, has increased its expected compound annual growth rate of revenue from 10% to 15% to 15% to 20%.
In addition, TSMC plans to build new fabs in Arizona (5nm), Japan (22/28nm), Kaohsiung (7/28nm) and Hsinchu (2nm) in Taiwan, while expanding the existing Nanjing plant, while new construction plans in Germany and Taichung (2nm) in Germany are also under consideration.
Japanese chipmaker Renesas Electronics has also said it will increase its supply capacity of key components for automotive and electronics by more than 50 percent by 2023.
Back to the domestic market, China's MCU market is also the leading overseas market. The 7 leading overseas enterprises occupy about 70% of the market share, and the localization rate is low. For local companies, the alternative market potential is huge.
Under the hot air outlet of the market, domestic semiconductor companies have entered the automotive MCU.
Among them, the limelight is BYD Semiconductor, which just completed its IPO on the ChiNext board on January 27 this year. Not surprisingly, BYD Semiconductor will become the first listed company in China to have an on-board chip.
BYD Semiconductor was originally just a business unit within the BYD Group, and was founded by BYD Microelectronics in 2004. It has made achievements in the field of MCU for many years, from industrial-grade MCUs to products and services such as automotive-grade 8-bit MCU chips and auto-regulatory 32-bit MCU chips.
Officially split out in 2020, it was competed for by more than 100 VCs/PE just after independence, and the valuation soared to 30 billion yuan, and it took only 2 years to achieve an IPO, which is a speed.
NavInfo also has a layout in the MCU field. Its wholly-owned subsidiary, Autochips, focuses on automotive electronic chips and is the first manufacturer of mass-produced vehicle-grade MCUs in China.
▲Autochips 32-bit vehicle-grade MCU-AC7801x
In addition, Chipson Microelectronics, Chipways and other companies have also launched vehicle-grade MCU products, and domestic design companies such as Unigroup Guowei, Zhongying Electronics, Gigabit Innovation, Beijing Junzheng, Lingou Chuangxin (Jingfeng Mingyuan M&A), and Core Technology are also increasing the R&D and certification of vehicle-grade MCUs.
Traditional enterprises expand production capacity, and new players enter the game strongly. Solving the problem of lack of cores in global car companies is a proposition that affects the entire industry.
——02——
What exactly are the car chips?
From the perspective of data processing capability, in addition to the MCU that is essentially controlling the instruction operation, the vehicle chip also includes the main CPU with relatively strong computing power and the AI chip based on intelligent computing (there are also CPUs that are classified as similar to the MCU, which are not expanded here).
AI chips have the highest requirements for computing power.
Broadly speaking, chips that can calculate AI algorithms are called AI chips. However, in a narrow sense, AI chips are generally defined as "chips specifically designed for AI algorithms to accelerate".
At present, general-purpose CPUs, GPUs, FPGAs, etc. can perform AI algorithms, but the execution efficiency varies greatly.
FPGA means Field Programmable Gate Array,000 semiconductor devices based on programmable devices. In contrast to Application Specific Integrated Circuits (ASICs) that are custom manufactured for specific design tasks, FPGAs can be reprogrammed after manufacturing to meet the desired application or functional requirements.
At the same time, some chip companies have begun to design ASIC-specific chips specifically for AI algorithms, such as Google TPU and Horizon BPU.
Before the large-scale rise and batch launch of intelligent driving industry applications, the use of existing general-purpose chips such as GPUs and FPGAs can avoid the high investment and high risk of special research and development of customized chips (ASICs).
However, due to the fact that such general-purpose chips are not designed specifically for deep learning, there are problems such as insufficient performance and excessive power consumption.
Specially customized ASICs can effectively solve these problems, but because of their high costs and risks, the market is not yet mature.
Autonomous driving chips are often referred to as AI chips, and the computing power is measured in TOPS (trillion times per second). The computing power requirements of the smart cockpit chip are relatively lower than those of the automatic driving chip.
From the perspective of chip structure form, with the continuous deepening of automobile intelligence and the improvement of automatic driving level, the traditional CPU has long been unable to meet the computing power needs of intelligent cars.
SoC heterogeneous chips with integrated AI accelerators came into being.
SoC (System on Chip), generally referred to as system-on-chip, is also known as a system-on-chip, that is, an entire system capable of performing a function is integrated on a single chip. SoC is often composed of CPU + GPU + DSP + NPU + various peripheral interfaces, storage types and other electronic components.
Autonomous driving chips and smart cockpit chips are generally SoC system-level chips.
Compared with MCUs, SoCs have higher somatic complexity and richer integration features, enabling the operation of multi-task complex systems.
The MCU has reduced the frequency and specifications of the CPU, and the complexity is lower.
The MCU is chip-level and is used to control instruction calculations. SoCs are system-level and are used for intelligent computing.
An autonomous driving AI chip that provides the brain for the whole vehicle
As the level of autonomous driving increases from L0 to L5, the computing power requirements for the chips that provide them with brains are also rising.
How much computing power does it take to achieve autonomous driving at L4 and above? It is generally believed that the AI computing power required by L4 > 100TOPS, and the AI computing power required by L5 is as high as 500-1000TOPS.
But there is no clear conclusion in the industry. Among the existing products, there are Mobileye's 176TOPS to support L4/L5 level automatic driving, and there are also Nvidia's large-scale 1000TOPS hashrate redundancy scheme.
The competition in computing power has obviously become a silent tacit understanding between various AI chip manufacturers.
Relying on its strong technical strength, NVIDIA is far ahead in the field of high-level automatic driving, and is currently the deserved king of automatic driving chips.
On April 12, 2021, Nvidia unveiled A NEW GENERATION AI processor NVIDIA DRIVEAtlan for autonomous vehicles at the GTC conference, with a single SoC hash rate of a staggering 1,000TOPS, which is stronger than the computing power of most L4 autonomous vehicles, and it will be used in the 2025 models of multiple automakers.
▲ NVIDIA DRIVEAtlan
NVIDIA's previous NVIDIA DRIVE Orin SoC can provide 254 TOPS of computing power, which can fully cover the needs of ADAS to L5 autonomous driving.
NVIDIA DRIVE AGX Xavier provides 30 TOPS operations for L2+ and L3 autonomous driving. At its core is NVIDIA's first automotive-grade Xavier System-on-Chip.
Orin and Xavier are the mainstream of the current car companies.
Weilai ET7, Zhiji L7, WM M7, Xiaopeng P5 and other models that will be mass-produced and listed have all chosen Orin X.
Under pressure from Nvidia, Mobileye unveiled three new chips at CES 2022, namely EyeQ 6L, EyeQ 6H and EyeQ Ultra.
The EyeQ Ultra has a total of 64-core accelerators, 5nm process. It also supports 176 TOPS's 8-bit deep learning operations. Compared to the competitors' numbers, 176 is not large, but Professor Shashua, CEO of Mobileye, believes that the key is not only computing power, but also efficiency.
Judging from Mobileye's product data, 176TOPS is about the sum of the computing power of 10 EyeQ 5 chips. Mobileye, on the other hand, has Robotaxis powered by 8 EyeQ 5 drivers.
It can be seen that the single-chip EyeQ Ultra, equivalent to 10 EyeQ 5 integrated, is fully capable of supporting the computing power required by RobotTaxi. The price is only a few hundred dollars, and the energy consumption of the entire system is very low, less than 10W.
The hash rate of EyeQ 6H and EyeQ 6L is about 45TOPS and 5TOPS respectively, which is very limited compared to the previous generation of products in the case of a significant increase in computing power.
It can be seen that the highlight of Mobileye is to reduce energy consumption and cost as much as possible under the premise of ensuring sufficient computing power.
For chips that need to be mass-produced on the car, energy consumption and cost are also very important indicators in addition to computing power.
In addition, in addition to NVIDIA and Mobileye, in early 2022, the autopilot chip track has entered a new heavyweight player.
Also at ces 2022, american chip company Ambarco launched the CV3 series of AI domain controller chips based on the 5nm process, whose AI equivalent computing power reached 500eTOPS (e in eTOPS represents equivalent (equivalent)), which can be used for L2+ to L4 autonomous driving.
Compared with the previous generation of CV2, CV3's AI performance has increased by 42 times, an amazing increase.
Electric truck maker Rivian, self-driving technology company Motional and electric vehicle manufacturer Arrival have all adopted Amba CV2-based solutions.
Tesla, which uses a "full-link" route, has also developed its own FSD (Full Self Driving) chip.
The design and planning of the FSD chip began in 2016, when Tesla claimed not to have found other solutions that worked for them. FSD chips are currently used on the Model 3.
In the domestic market, horizon, black sesame intelligence, xinchi technology and other enterprises have also launched self-driving chip products that have been widely recognized by the market.
Journey 5 is Horizon's third-generation vehicle-grade product with an equivalent computing power of 128TOPS. It can be applied to both autonomous driving and intelligent cockpit platforms with high computing power requirements.
Back in 2017, Horizon had already received nearly $100 million in investment from Intel.
Huashan No. 2 A1000 Pro is the latest generation of large-scale automotive standard-level automatic driving computing chip under Black Sesame Intelligence, released in April 2021, and successfully snippet in July of the same year, which is currently the most powerful domestic automotive standard-level automatic driving chip.
Black Sesame Intelligence has previously launched the Huashan A1000, which has an 8-core CPU and a computing power from 58TOPS (INT8)-116TOPS (INT4). Huashan A1000L is its simple version, 6-core CPU, hash rate 16TOPS.
A single A1000L is suitable for ADAS assisted driving, a single A1000 is suitable for L2+ autonomous driving, and four A1000 can support the autonomous driving needs of L4 and above.
And The core chi technology, which was only established in 2018, launched the V9 series of automatic driving chips.
During the Shanghai Auto Show in April 2021, XCHI Technology launched the latest generation of V9T autopilot chips, which use two completely independent quad-core Cortex-A55 application processor clusters that can be operated independently to provide higher performance or can be redundant backups to improve security.
As a well-deserved leader in the field of consumer electronics, Huawei is determined to migrate its technology accumulation to the automotive field, and naturally it is not willing to fall behind in automatic driving.
Based on its MDC Pro 610 platform customized for BAIC ARCFOX Alpha S, Huawei has created the latest MDC 810 platform standardized product for all car company customers, with a computing power of 400+ TOPS.
Huawei claims to be the largest computing intelligent driving platform that has been mass-produced. It is said that the platform uses the Ascend series chip Ascend 610, but Huawei has not officially announced it.
Zero-run cars have taken the same self-research route as Tesla, cooperating with dahua in the field of security to launch Lingxin 01. Using 28nm process technology, computing power of 4.2Tops, power consumption of 4 W, equipped with zero-run C11, has been mass production and listing.
Zero-run definition of Lingxin 01 is China's first vehicle-grade AI intelligent driving chip with completely independent intellectual property rights.
Superstar Future, which was founded in early 2019, also released the NOVA30P, a next-generation high-level autonomous vehicle computing platform, at the International Intelligent Connected Vehicle Technology Annual Conference in May 2021, using a high-performance processing chip and a heterogeneous hardware solution of a functional safety ASIL-D-rated MCU.
In addition, the cloud smart chip manufacturer Cambrian also said on the investor interaction platform a few days ago that its subsidiary Xingge Technology is designing and developing in-vehicle smart chips for high-level intelligent driving application scenarios.
With the development of smart cars, AI chips have great potential.
According to soochow Securities Research Institute, the value of AI chip bicycles will increase from $100 in 2019 to $1,000+ in 2025.
The scale of the mainland automotive AI chip market will also increase from US$900 million in 2019 to US$9.1 billion in 2025, with a compound growth rate of 46.4%; by 2030, it will reach US$17.7 billion, with a ten-year compound growth rate of 28.1%.
The smart cockpit SoC that provides the brain for the car machine
Smart cockpit chips are another type of highly complex SoC chips in in-vehicle chips.
The intelligent cockpit SoC is mainly responsible for the operation and processing of massive data in the cockpit.
With the rapid development of the intelligent cockpit, the cockpit SoC not only needs to deal with the needs of multi-screen scenarios such as instruments, cockpit screens, AR-HUDs, etc., but also needs to perform operations such as speech recognition and vehicle control, and the performance and computing power demand of smart cars for cockpit SoCs continue to rise.
Qualcomm is the absolute hegemon in the field of intelligent cockpit chips, and the mainstream intelligent cockpit car chips on the market are basically from Qualcomm.
Qualcomm's mass-produced SA8155P chip AI hash rate of about 8TOPS, and its fourth-generation cockpit SoC integrated NPU hash rate of up to 30TOPS, which is the cockpit SoC product with the highest AI hash rate released so far, and is scheduled to be put into production in 2023.
WEY Mocha, Xiaopeng p5, WEILAI ET7, WM W6 and Zero Run C11 and other models all use Qualcomm's latest 8155 processor.
Tesla, which has always played by the norm, has replaced its performance version of modelY's cockpit SoC from Intel A3950 to AMD Ryzen from November 2021.
It is reported that compared with the Qualcomm 8155 processor, AMD Ryzen's CPU performance is 2 times stronger, and the GPU performance is about 1.5 times stronger.
The industry generally expects that Tesla's upgrade will change the situation of a dominant smart cockpit chip and set off a new round of "arms race" for smart cockpit chips.
In addition, traditional automotive SoC manufacturers such as Samsung, Renesas, NXP, and Texas Instruments have also launched smart cockpit SoC products.
For example, in July 2021, Renesas launched a new series of R-Car Gen3e products, covering six new products, and plans to mass-produce in 2022.
In November 2021, Samsung released the new Exynos Auto V7 for mid-to-high-end intelligent cockpit systems, which has been applied to Volkswagen's in-vehicle computer ICAS3.1.
In the domestic market, Huawei HiSilicon developed the Kirin 990A based on the mobile phone processor Kirin 990, which uses an 8-core CPU, has 3.5TOPS computing power, and supports 5G networks.
X9U smart cockpit SoC of Xchi Technology has a CPU final strength of 100KDMIPS and AI computing performance of 1.2TOPS.
The X9U supports up to 10 independent Full HD displays with a single chip, including front-row meters, a central control screen, a HUD, and multiple entertainment screens for multi-screen sharing and interaction.
MediaTek's AutosI20 (MT2712) is a high-performance six-core infotainment solution with flexible interfaces that support multiple displays, including four ARM Cortex-A35 processors and two Cortex-A72 processors.
In addition, Geely's core engine technology for the first time in December 2021 publicly released a new generation of 7-nanometer vehicle specification-level intelligent cockpit multimedia chip "Dragon Eagle One", and its AI computing power reached about 8TOPS.
As the only 7-nanometer chip produced by TSMC in China, Longying-1 is one of the very few high-end intelligent cockpit chips on the market that uses a 7-nanometer process.
Core Engine Technology was established by a joint venture between EGATONG and Anmou China.
In addition to the MCU chips, autonomous driving AI chips and smart cockpit SoCs mentioned above, in-vehicle semiconductors also include power semiconductor devices such as IGBTs and MOSFETs.
BYD Semiconductor's performance in the IGBT module is particularly prominent, in 2019, 2020 for two consecutive years, BYD Semiconductor ranked second in the world among new energy passenger car motor drive manufacturers, domestic manufacturers ranked first, the market share reached 19%, second only to Infineon.
M&A and investment, buying, buying and buying to improve business
Overall, the current market for in-vehicle chips is still monopolized by a few giants. However, with the development of the industry, a number of emerging companies are rising, and capital and technology have entered the game.
In 2021, a number of start-ups will receive a new round of financing. Under the pressure of competition, traditional chip manufacturers have also begun to seek to consolidate their own moats. In addition to further developing new products, improving technology and business through mergers and acquisitions is also a common model adopted by major manufacturers.
For example, the NVIDIA acquisition of ARM, which has lasted for more than a year.
Just this past February 7, 2022, Nvidia officially announced that it would abandon the acquisition of ARM.
Nvidia's announcement that it will buy ARM for $40 billion is still in September 2020.
As we all know, NVIDIA is a king-like existence in the field of GPU and AI chips. ARM is the world's largest provider of semiconductor intellectual property (IP). The combination of the two is bound to affect the pattern of the entire industrial chain.
Therefore, as soon as the acquisition was announced, it was unanimously opposed by various industry giants and even regulators in various countries.
On December 2, 2021, the U.S. Federal Trade Commission (FTC) issued a statement saying that it would formally file an administrative lawsuit against Nvidia's acquisition of ARM on August 9, 2022.
On February 7, Nvidia officially announced that it would abandon the acquisition of ARM. To that end, Nvidia also lost $1.25 billion that had been prepaid to SoftBank.
Coincidentally, in 2018, Qualcomm proposed to acquire Dutch chip giant NXP, and finally had to announce its abandonment for similar reasons.
But failure hasn't stopped the giants from making mergers and acquisitions.
Just january 28, China's State Administration for Market Regulation (SAMR) just announced conditional approval of U.S. chip company AMD's plan to buy FPGA maker Xilinx for $35 billion.
Previously, the transaction has been approved by market regulators in many countries and economies, including the European Union, the United Kingdom and the United States.
Founded in 1984, Xilinx, along with Altera, is the inventor of FPGA, programmable SoC and ACAP. In 2015, Altera was acquired by AMD's old rival Intel for $16.7 billion.
In addition, following the news in early 2021, foreign media reported in January this year that Samsung intends to acquire Infineon or NXP.
Samsung Electronics is currently the world's largest memory chip manufacturer and the world's second largest chip foundry after TSMC. Foreign media analysis believes that it will acquire companies in the field of automotive semiconductors to enhance its strength in this field.
Semiconductor giants are spending money on buying and buying, and startups are busy attracting investment.
In June 2021, Horizon completed the C7 round of financing, with a financing amount of up to US$1.5 billion, and investment institutions including Weihao Chuangxin, BOE, etc., with a post-investment valuation of up to US$5 billion.
In July 2021, Xinchi Technology announced the completion of nearly 1 billion yuan of B round financing, in addition to a number of funds and investment and financing companies, CATL also through The Heavy Position of Morning Road Capital.
Black Sesame Intelligence completed the multi-hundred-million dollar strategic round and Series C financing led by Xiaomi Yangtze River Industry Fund in September 2021, and then completed the C+ round of financing in January 2022, with the investor being Boyuan Capital, a subsidiary of Bosch Group.
2021 for the automotive chip industry is both tough and full of life.
Car companies that lack cores have to stop work and reduce production, while chip manufacturers have made great efforts to expand production capacity, and new players have also entered the game.
At the policy level, on February 8, the official website of the European Commission officially released the European Chip Act. Under the bill, the EU will invest more than €43 billion in public and private funding to support chip production, pilot projects and start-ups.
At the same time, Intel also announced on February 8 that it will set up a new $1 billion fund to support early-stage startups and mature companies to build disruptive technologies for the foundry ecosystem.
When will the crisis of lack of cores pass?
There are relative optimists in the industry, such as Infineon and Tesla, who believe that it will be solved in 2022 to 2023. However, according to the Financial Associated Press, Intel CEO recently said that the chip supply shortage will continue throughout 2023, and the supply situation will improve in 2025-2030.
No matter when this crisis can really end, we believe that under the positive stimulation of market demand, the industry will move forward at the fastest speed, and the future pattern will change.