As the vane of the semiconductor industry, the memory industry as a whole is under pressure in the third quarter of this year. Recently, in addition to Samsung, the original head memory manufacturers SK Hynix, Micron, and Kioxia have all proposed to reduce capital expenditures for next year based on current performance and market performance, and colleagues have slowed down their recommendations for advanced technology. This means that in the face of today's high inventory pressure, the original upstream memory factory is actively taking the lead in reducing inventory.
But Yole predicts that the revenue of automotive memory will triple by 2027, accounting for 17% of the automotive semiconductor market, with a compound annual growth rate of 20% from 2021 to 2027, exceeding the global memory market of 8% and the automotive semiconductor market in the same period 10% of the same period.
According to Yole's latest report, the independent memory market is as high as $167 billion, accounting for 28% of the entire semiconductor market. In contrast, the automotive memory market ($4.3 billion in 2021) accounts for 2.6% of global memory market revenue and 10% of automotive semiconductors, indicating the prevalence of non-memory electronic components in current vehicles.
Automotive Semiconductors: Coming Soon
According to Yole, the automotive semiconductor market continues to grow.
This is to be expected as the penetration rate of semiconductor-based applications such as higher levels of advanced driver assistance systems (ADAS) and electrification increases. While the light vehicle market is relatively flat, the semiconductor chip market is expected to grow from $44 billion in 2021 to $80.7 billion in 2027, at a CAGR of 11.1%. This represents about $550 worth of semiconductor chips per vehicle, growing to about $912 by 2027. This is because of the growth in the number of chips implemented in cars, from about 820 chips today to about 1,100 chips per vehicle in 2027.
The rapid development of vehicle electrification requires new types of substrates, such as SiC for power electronics. It is expected to reach 11,300 substrates by 2027. Although still low compared to silicon substrates projected in 2027, SiC will grow faster than silicon and GaAs/Sapphire.
ADAS is also an important driver, and MCUs with cutting-edge technology down to 16nm/10nm will enter ADAS, including radar and other sensor controls. Level 4 and Level 5 autonomous driving will drive growing demand for more memory (DRAM) and computing power.
For electrification, vertical integration is becoming increasingly popular among OEMs and can be done in a variety of ways: full integration down to the component level, system integration and subcontracting of on-demand printed parts, strategic partnerships/direct investments with key component suppliers, etc. The traditional automotive supply chain needs to thoroughly review its positioning and transform through joint ventures, mergers and acquisitions, as well as new investments and divestments, to maintain its competitive advantage.
Although semiconductors are critical to an automotive industry undergoing a disruptive transformation, most players, including OEMs and T1 suppliers, do not yet have a clear semiconductor strategy. Preparing for the future calls for specific expertise in internal and external semiconductor technologies and their supply chains.
Automotive memory: The Big Upgrade
Automotive memory is a market with great potential, and a variety of memory products will have great growth, including NOR/NAND FLASH, DRAM, EEPROM, SRAM, etc.
Led by Tesla, a growing number of automakers are converting front fascias to touchscreen displays, sparking demand for high-capacity memory. In addition, the commercialization of 5G and the exploration of autonomous driving have produced a large number of high-speed data interaction applications in automobiles, and also made high-speed and high-capacity memory a necessity.
In 2017, the hardware cost of each car memory device was only around $20 (not including the memory unit integrated in the MCU). When the intelligent driving reaches L4/L5, the hardware cost of the memory device is about 300-500 US dollars. Some people estimate that a bicycle needs to be equipped with 32GB DRAM and 208GB NAND to match its functions, so as to realize the functions of a smart car.
NAND and DRAM will be the main growth points of on-board memory, driven by the upgrading of communication and entertainment systems and the implementation of automatic driving brought about by the rapid development of 5G and AI.
Advanced Driver Assistance Systems (ADAS) and driverless vehicle systems require powerful processors with memory capacity and memory requirements that only DRAM can match.
Similarly, large-capacity NAND also plays an irreplaceable role. ADAS reminds the driver to pay attention to potential dangers such as the car being too close to other obstacles, automatically turns on the headlights, adjusts the driving speed, initiates emergency braking, reminds the driver to pay attention to Friday vehicles, keeps the vehicle driving normally in the lane, and even monitors Driver's blind spot. In addition, infotainment system settings must be saved instantaneously in case information is lost due to a power outage. All these functions require high-performance NAND to support.
Due to the surge in the amount of data generated and processed by electronic devices in cars, car memory will undergo an unprecedented round of upgrades in terms of capacity and specifications.
Market
Currently, the overall automotive memory market is dominated by Micron with an estimated market share of 45% (by revenue), followed by Samsung with 13%. Infineon, KIOXIA, SK Hynix and ISSI are all ranked behind Samsung, with a market share of ≤7%.
With the continuous evolution of automotive computing, the largest incremental space for DRAM still comes from autonomous driving.
The core of autonomous driving is AI technology, which requires extremely high computing power, innovative memory and memory systems to process and save the large amount of data that computers need to make decisions that mimic human beings.
Self-driving cars will contain more lines of code than any other software platform created to date. In the future, it is expected that a typical vehicle will contain more than 300 million lines of code, contain more than 1 terabyte of memory, and require memory bandwidth of more than 1 terabyte per second to support the computing performance required by autonomous driving platforms. Traditional DRAM is obviously not up to the task.
Future Trends of Automotive Memory
The development of electrification, informatization, intelligence, and networking promotes the automotive memory revolution. In the future, automotive memory will move from GB level to TB level. From the current point of view, the ADAS system, the new generation of central control system, the introduction of 5G connection technology to realize the Internet of Vehicles, the terminal edge cloud and OTA, etc. are the carriers of basic code, data and parameter memory. In the future, richer entertainment systems, stronger central control computers and digital cockpits, more complete event recording systems, more sensors and assisted driving decision-making will put forward "TB-level" requirements for memory space.
The latest generation of LPDDR5 and the upcoming DDR5 are given high hopes. LPDDR5 is particularly well suited for the larger displays used in the latest models of vehicles, capable of managing the increasingly complex navigation images and control areas of cockpit units. In addition, digital clusters utilizing in-vehicle cameras, front and rear sensors, and driver monitoring systems, derived from traditional IVI, all require the use of LPDDR5’s high-end features. The more powerful DDR5 will appear in the next generation of autonomous driving SoC.
Memory manufacturers have already taken action. Recently, memory giant Micron announced that the first automotive LPDDR5 memory has begun to deliver samples. The solution is designed based on the international standard ISO 26262 and meets the most stringent requirements of the American Automobile Association.
Despite the bright future, industry insiders are still cautious about whether DRAM can break out in the near future. The increase in the amount of DRAM needs to consider the progress of large-scale application of technologies such as autonomous driving and intelligent transportation. The electrification and intelligence of automobiles cannot be completely equated with the widespread application of autonomous driving. It is believed that DRAM will only explode after autonomous driving matures. sexual growth.
Smart functions require faster interfaces and greater density
The automotive market has specific requirements in terms of quality, qualification, reliability, functional safety, and longevity of supply that require extra effort and dedication from the players addressing this market.
For a long time, the memory technologies used were limited to robust solutions such as EEPROM and NOR flash. This has changed with the proliferation of digital cockpits and ADAS smart sensors and autonomous driving features.
From the analog dashboard with the main infotainment unit, vehicles are now adopting fully centralized digital cockpit electronics. For example, DRAM has evolved from DDR2 and DDR3L to LPDDR4(x), and in some cases even GDDRx. The associated increase in data memory requirements has led to the adoption of larger, faster-manageable NAND solutions. Starting with eMMC, more and more designs are incorporating UFS and potentially PCIe solid-state drives (SSDs) in the most luxurious vehicles.
ADAS capabilities are driving the proliferation of smart sensors such as front-facing cameras, imaging radar, and even lidar. This will require high-density NOR flash ((Q)SPI to xSPI) and DRAM (DDR3L or LPDDR4), depending on the application processing unit used.
Autonomous driving requires central processing and artificial intelligence capabilities. The latter requires high-bandwidth DRAM. While most designs use LPDDR4(x) or DDR4, some use GDDRx, and we might think of HBM in the future. Autonomous driving needs to store a lot of code and data, so eMMC or UFS devices are required. Event data recorders (black boxes) for autonomous vehicles are expected to require very high density and speed in the future, which may lead to the adoption of PCIe SSDs.
As an indispensable member of the automotive electronic system, the prospect of automotive memory is limitless. Huge business opportunities At present, the three major memory giants have made various layouts for the automotive market.
