With the acceleration of automotive electrification, automotive semiconductors have grown rapidly. In 2017, the global market size reached US$28.8 billion (+26%), far exceeding the growth rate of vehicle sales (+3%). The largest share of the market was functional chips (MCUs) (US$6.6 billion, accounting for 23%), followed by power semiconductors (21%), sensors (13%), etc.
Automotive semiconductors can be categorized into functional chips (MCUs, Microcontroller Units), power semiconductors (IGBTs, MOSFETs, etc.), sensors, and others. According to Strategy Analytics, MCUs account for the highest value in traditional gasoline-powered vehicles, at 23%; in pure electric vehicles, MCUs are second only to power semiconductors, at 11%. DIGITIMES predicts that the market size of functional chips (MCUs) is expected to steadily increase from $6.6 billion in 2017 to $7.2 billion in 2020.
▲ Global automobile sales (in ten thousand vehicles)
▲ Global Automotive Semiconductor Market Size (USD Billion)
▲ Classification of semiconductors in gasoline-powered vehicles by type
▲ Classification of semiconductors for pure electric vehicles
▲ Market size of automotive functional chips (USD billion)
Traditional automotive functional chips are only suitable for local functions such as engine control and battery management, and cannot meet the computational demands of high-data-volume intelligent driving.
In recent years, with the increasing penetration rate of intelligent driving, global chip giants have entered the automotive industry, launching main control chips with AI computing capabilities. The main control chip market is expected to grow rapidly, with IHS predicting it could reach $4 billion by 2020.
▲ Automotive chips: Main control chips & functional chips
▲ Automotive main control chip market size (USD billion)
The main control chip giants have strong AI computing advantages, while the functional chip manufacturers have rich experience in the automotive industry chain. Mergers, acquisitions and alliances frequently occur between the two camps.
To date, Nvidia has partnered with over 370 OEMs and Tier 1 suppliers worldwide; Intel acquired Mobileye to enter the automotive industry; and Qualcomm once attempted to acquire NXP.
▲ Automotive chip market landscape
Main control chip: computing power continues to grow
Intelligent driving involves human-computer interaction, visual processing, and intelligent decision-making, with AI algorithms and chips being the core components. According to NXP, a high-end car currently contains over 100 million lines of code, far exceeding that of airplanes, mobile phones, and internet software. In the future, as the penetration rate and level of autonomous driving increase, the number of lines of code in cars will grow exponentially.
The computational demands of autonomous driving software have reached the trillions of TOPS (Tera Operations Per Second). Traditional automotive MCUs struggle to meet the computational requirements of autonomous vehicles, leading to the introduction of AI chips such as GPUs, FPGAs, and ASICs into the automotive market.
▲ Cars already contain over 100 million lines of code
▲ The number of lines of code in automobiles is growing exponentially.
▲ Computing power of a typical automotive MCU
▲ Computing power of NVIDIA GPU SoC
Global leaders in autonomous driving include Google, Baidu, Tesla, and Audi. The SoC chip architecture of their autonomous driving main control modules may offer insights into the development direction of automotive chips.
Google's Waymo uses an Intel CPU + Altera FPGA solution, with Infineon MCUs as the communication interface. Google Waymo's computing platform uses Intel Xeon CPUs with 12 or more cores, paired with Altera's Arria series FPGAs, and Infineon's Aurix series MCUs as the communication interface for CAN or FlexRay networks.
▲ Google Waymo's computing platform architecture
Baidu Apollo uses NXP/Infineon/Rensas MCUs + Xilinx FPGAs/NVIDIA GPUs. Baidu's autonomous driving prototypes use an IPC (industrial control computer) solution, but the size and power consumption of industrial PCs are difficult to meet mass production requirements. Therefore, Baidu has also launched a domain controller embedded solution suitable for mass production. Raw data from various sensors is input into a Sensor Box, where data fusion is performed. The fused data is then transmitted to a computing platform for autonomous driving algorithm processing.
Baidu's dedicated autonomous driving computing platform, ACU (Apollo Computing Unit), defines three product series: MLOC (high-precision positioning, MCU), MLOP (high-precision positioning + environmental perception, MCU + FPGA), and MLOP2 (high-precision positioning + environmental perception + decision planning, MCU + GPU).
▲ Baidu Apollo's industrial control computer computing platform architecture
▲ Baidu Apollo's domain controller computing platform architecture
Tesla: From Mobileye ASICs to NVIDIA GPUs. In 2014, Tesla released Autopilot 1.0, featuring one front-facing camera, one rear-view camera (not involved in driver assistance), one front-facing radar, and 12 ultrasonic sensors. The vision chip used was the Mobileye EyeQ3, and the main control chip was the NVIDIA Tegra 3.
In late 2016, Tesla released Autopilot 2.0, which features three front-facing cameras (with different perspectives: wide-angle, telephoto, and medium), four side cameras (left front, right front, left rear, and right rear), one rear-facing camera, one enhanced front-facing radar, and twelve ultrasonic sensors (with double the sensing distance). The main control chip is the NVIDIA Drive PX 2, which has a processing speed 40 times faster than Autopilot 1.0.
▲ Mobileye EyeQ3 chip architecture
▲ NVIDIA Drive PX2 chip architecture
Audi: A multi-chip integrated solution combining Mobileye ASIC, NVIDIA GPU, Altera FPGA, and Infineon MCU. The new Audi A8 unveiled its zFAS controller solution. The zFAS comprises four high-performance processors: 1) Mobileye's EyeQ3 for visual information processing, including traffic sign recognition, pedestrian recognition, collision warning, and lane detection; 2) NVIDIA's Tegra K1 SoC for 360° surround view imaging; 3) Altera's Cyclone5 FPGA for sensor fusion, map fusion, and assisted parking; and 4) Infineon's Aurix series MCUs for traffic congestion control and driver assistance.
▲ Audi A8's computing platform architecture
In the field of automotive main control chips, GPUs will remain the mainstream in general automotive main control chips, FPGAs will serve as an effective supplement, and ASICs will become the ultimate direction.
Currently, artificial intelligence and intelligent driving algorithms are not yet finalized. As a general-purpose accelerator, GPUs are expected to maintain their mainstream status as automotive main control chips for a considerable period of time. FPGAs, as hardware accelerators, are expected to become an effective supplement to GPUs. In the future, if all or part of the intelligent driving algorithms are solidified, ASICs will become the ultimate choice with the best cost performance.
▲ Trend chart of automotive main control chips
1. Nvidia: GPU monopoly advantage, from smart cockpits to autonomous driving
Nvidia's revenue and net profit are growing rapidly, with the automotive industry being a long-term driver. Nvidia is the leader in the GPU field, consistently maintaining a market share of over 70%. Nvidia's revenue for fiscal year 2018 (corresponding to calendar year 2017) was $9.71 billion, a year-on-year increase of 40.6%; net profit was $3.05 billion, a year-on-year increase of 82.9%.
▲ Global discrete GPU market share (2009-2017)
▲ Nvidia's Revenue (millions of US dollars)
▲ Nvidia's net profit (millions of US dollars)
NVIDIA's Drive CX digital cockpit computer utilizes advanced 3D navigation, a high-resolution digital instrument cluster, natural voice processing, and image processing to enable driver assistance features. Drive CX is powered by the Maxwell-based Tegra X1 SoC, with an optional Tegra K1 SoC also available.
The main functions of DRIVE CX include: 1) Natural language processing, enabling functions such as address lookup and contact calling through voice recognition; 2) 3D navigation and infotainment, providing high-resolution, high-frame-rate graphics displays for numerous applications; 3) Fully digital instrument cluster, providing rich graphics displays through the instrument cluster or head-up display (HUD); 4) Surround vision, utilizing complex motion recovery structure technology and advanced stitching technology to improve image rendering of fisheye lenses, reduce ghosting, and render a virtual car in a high-definition model to achieve a realistic surround vision effect; 5) Integration with Android Auto, allowing drivers with Android smartphones or iPhones to easily access their mobile devices and interact with applications such as maps, search, and music.
▲ NVIDIA Drive CX Digital Cockpit Computer
NVIDIA's Drive PX autonomous vehicle platform combines deep learning, sensor fusion, and surround vision to transform the driving experience. Key features of Drive PX include: 1) Sensor fusion, which integrates data from 12 cameras, LiDAR, millimeter-wave radar, and ultrasonic sensors; 2) Computer vision and deep neural networks, suitable for running DNN (Deep Neural Network) models for intelligent detection and tracking; 3) End-to-end high-definition mapping, enabling rapid creation and continuous updating of high-definition maps; and 4) The DriveWorks software development kit, which includes applications, tools, and library modules for reference.
▲ NVIDIA's Drive PX autonomous vehicle development platform
2. Intel: Actively pursuing mergers and acquisitions to enter the autonomous driving chip market.
Intel's traditional business is experiencing sluggish growth, but its foray into the automotive sector is creating new growth opportunities. Intel was once the world's largest semiconductor chip manufacturer.
According to PassMark statistics, Intel held an 80% market share in the global CPU industry in Q1 2017. In recent years, with the rise of smartphones and the decline in the personal computer market, the growth rate of its core chip business revenue has slowed significantly, and the company's revenue has been surpassed by Samsung Electronics. The company attempted to produce mobile phone processors but ultimately failed, and was forced to disband the department responsible for that business.
In recent years, Intel has actively expanded into emerging fields such as autonomous driving, the Internet of Things, artificial intelligence, and VR through numerous acquisitions, creating new growth points for its performance and striving to transform from a traditional chip manufacturer into a diversified solutions provider.
▲ Global CPU market share (2004-2017)
▲ Intel's Revenue (millions of US dollars)
▲ Intel's net profit (millions of US dollars)
▲ Intel's revenue breakdown by business category (in millions of US dollars)
▲ Intel's acquisition trends over the past three years
Intel acquires Mobileye: A global leader in vision-based ADAS. Mobileye is one of the world's leading vision-based ADAS providers, holding an 80% market share and possessing a rich portfolio of vision-based ADAS products. Mobileye's proprietary software algorithms and EyeQ chips can perform detailed analysis of visual information and predict potential collisions with other vehicles, pedestrians, bicycles, or other obstacles. It can also detect road markings, traffic signs, and traffic lights.
By the end of 2017, Mobileye's products were used in 313 models from 27 automakers, with 8.7 million units shipped that year. In March 2017, Intel acquired Mobileye for $15.3 billion to create the Intel fleet. This fleet will include various car brands and models to showcase its versatility and adaptability. Level 4 vehicles will be deployed for testing in the United States, Israel, and Europe.
▲ Mobileye EyeQ5 will help cars achieve L4-L5 level autonomous driving
▲ Intel's "Vehicle-to-Cloud" System Solution
Intel's acquisition of Altera: Autonomous driving FPGA chips are now in mass production. Currently, the global FPGA market is mainly divided between Xilinx and Altera, which together hold nearly 90% of the market share and hold over 6,000 patents.
Altera's FPGA products comprise four main series: the top-of-the-line Stratix series (nearly $10,000), the cost-performance balanced Arria series ($2,000-$5,000), the affordable Cyclone series ($10-$20), and the MAX series CPLDs. Intel announced its acquisition of Altera in 2015 to support its rapidly growing data center and IoT businesses.
▲ FPGA market share distribution in 2016
3. Qualcomm: Leveraging its communication advantages, from infotainment to connected vehicles...
Qualcomm's traditional business revenue is declining, prompting it to actively expand into emerging industries. Qualcomm is the world's leading smartphone SoC manufacturer.
In the automotive sector, Qualcomm offers solutions including: 1) In-vehicle infotainment systems, providing cellular network solutions optimized for automobiles; 2) Driving data platforms, which intelligently collect and analyze data from various automotive sensors, enabling vehicles to achieve precise positioning, monitor and learn driving modes, perceive the surrounding environment, and accurately share information from this platform with the outside world; 3) Infotainment, providing 3D navigation, online media playback, parking assistance support, and functions such as voice, face, and terminal recognition; and 4) Wireless charging for electric vehicles, launching the Qualcomm Halo WEVC wireless charging solution.
▲ Global smartphone SoC market share (2016-2017)
▲ Qualcomm's revenue (millions of US dollars)
▲ Qualcomm's net profit (millions of US dollars)
▲ Qualcomm's emerging industry layout
Qualcomm has launched its in-vehicle infotainment system solution. The Snapdragon automotive platform infotainment system is now available in three versions: Select, High, and Premium.
The minimalist solution supports three displays, including the infotainment system, instrument cluster, and head-up display (HUD); the high-end tier supports up to four displays, with separate screens for the front passenger or rear seat entertainment, and also supports premium audio, low-latency wireless transmission of high-definition video, surround view processing, and deep learning and computer vision processing to identify nearby obstacles and pedestrians; the top-of-the-line solution supports up to six displays, including the instrument cluster, infotainment system, HUD, front passenger, and two separate screens for the rear seats.
At CES 2017, the Maserati showcased hardware featuring a custom Snapdragon automotive solution, including a Snapdragon automotive processor, Gobi 3G/4G LTE wireless modem, Wi-Fi, and Bluetooth modules. Another exhibited vehicle, the Chrysler Portal, featured a Panasonic in-vehicle entertainment concept system, which would operate based on the latest version of Android Auto and a Qualcomm Snapdragon chip.
▲ Snapdragon 602A Automotive Processor
Qualcomm has launched a new chipset for connected vehicles, supporting LTE and DSRC (Digital Switched Car-to-Everything) connectivity. The Snapdragon X5 LTE supports LTE connected vehicles at speeds up to Category 4, with downlink speeds of 150Mbps and uplink speeds of 50Mbps. The Snapdragon X12 LTE supports speeds up to Category 10, with downlink speeds up to 60 MHz 3x CA (450Mbps) and uplink speeds up to 40 MHz 2x CA (100Mbps).
The Snapdragon X16 LTE modem supports peak download speeds of up to 1Gbps, helping to meet the connectivity needs and use cases of next-generation connected vehicles, including high-definition map updates, connected navigation with real-time traffic and road condition information, software upgrades, Wi-Fi hotspots, and multimedia streaming.
In addition, in September 2017, Qualcomm launched the world's first commercial cellular vehicle-to-vehicle (C-V2X) solution based on the 3GPP Release 14 specification: the Qualcomm 9150 C-V2X chipset. This chipset includes an application processor running the Intelligent Transportation Systems (ITS) V2X stack and a Hardware Security Module (HSM). It was expected to be available in the second half of 2018, with mass production and supply to automakers as early as 2019. C-V2X simultaneously supports DSRC and LTE communication, providing vehicles with information about their surroundings and information in non-line-of-sight (NLOS) scenarios.
Functional chips: The technology is relatively mature, and the market landscape is stable with some changes.
The functional chip market is relatively mature and its structure is relatively stable. According to Strategy Analytics, the global installation volume of automotive MCUs exceeded 2.5 billion in 2016, with an average of 25-30 MCUs installed per car. The top 5 global automotive MCU manufacturers in 2016 were NXP (14%), Infineon (11%), Renesas Electronics (10%), STMicroelectronics (8%), and Texas Instruments (7%).
Compared to consumer chips and general industrial chips, automotive chips operate in much harsher environments: temperatures can range from -40 to 155°C, they experience high vibration, dust, and electromagnetic interference. Due to safety concerns, automotive chips have even higher requirements for reliability and safety, typically designed for a lifespan of 15 years or 200,000 kilometers. "Automotive-grade" chips require stringent certification processes, including reliability standard AEC-Q100, quality management standard ISO/TS 16949, and functional safety standard ISO26262.
It typically takes 2-3 years for a chip to complete automotive-grade certification and enter the supply chain of OEMs; once in the supply chain, it can generally have a supply cycle of 5-10 years. High safety and high reliability standards, long supply cycles, and long-term cooperative relationships with midstream and downstream component manufacturers and OEMs are the main reasons for the current stable automotive chip landscape.
▲ Automotive-grade chips vs. consumer and industrial-grade chips
▲ Overview of Major Global Automotive MCU Companies
▲ Computing power of a typical automotive MCU
▲ Global automotive MCU market share in 2016
The landscape of the functional chip market is also subject to change: 1) Traditional functional chip manufacturers are actively expanding into main control chips while maintaining their existing market share, such as NXP Bluebox, Infineon Aurix, and Renesas R-Car; 2) Functional chip manufacturers are consolidating their advantages through mergers and acquisitions, such as NXP's acquisition of Freescale and Infineon's intention to acquire STMicroelectronics; 3) Semiconductor giants also hope to acquire automotive technology and channel experience by acquiring functional chip manufacturers, such as Intel's acquisition of Mobileye and Qualcomm's attempt to acquire NXP.
NXP: Provides complete automotive semiconductor solutions; its Bluebox platform supports Level 4 autonomous driving.
Automotive electronics portfolio: NXP's automotive semiconductor products cover MCUs and MPUs, in-vehicle networking, media and audio processing, intelligent power drivers, energy and power management, sensors, system base chips, driver assistance transceivers, automotive safety, etc.
Autonomous Driving Platform: NXP BlueBox is an autonomous driving development platform that integrates the S32V234 automotive vision and sensor fusion processor, the S2084A embedded computing processor, and the S32R27 radar microcontroller. BlueBox can perform multi-sensor fusion (millimeter-wave radar, vision, LiDAR, vehicle-to-everything), supports Level 4 autonomous driving, consumes less than 40W of power, and has a computing power of 90,000 DMIPS (Dhrystone Million Instructions executed Per Second).
Vision chip: S32V234 vision processor, featuring a CPU (4 ARM Cortex A53 cores and 1 M4 core), a 3D GPU (Vivante GC3000), and a vision acceleration unit (2 APEX-2vision accelerators), supporting 4-channel cameras. It can be used in front-view cameras, rear-view cameras, surround-view systems, sensor fusion systems, etc., enabling real-time 3D modeling with a computing power of 50 GFLOPs. Simultaneously, the S32V234 chip reserves interfaces to support millimeter-wave radar, LiDAR, and ultrasonic sensors, enabling multi-sensor data fusion and supporting standards up to ISO26262ASIL-C.
Radar chip: S32R27 radar processor, using two e200z7 32-bit CPUs and two 32-bit lockstep mode e200z4s. It can support functions such as adaptive cruise control, intelligent headlight control, lane departure warning and blind spot detection.
▲ NXP's Revenue (millions of US dollars)
▲ NXP's net profit (millions of US dollars)
▲ NXP Bluebox Autonomous Driving Development Platform
Infineon: Covers integrated circuits and power semiconductors, vision and radar chips supporting ADAS functions.
Automotive electronics portfolio: Infineon's automotive semiconductor products cover body semiconductors, automotive safety, chassis assemblies, powertrains, hybrid and electric vehicles, active antennas, etc.
Autonomous Driving Platform: Infineon introduces the Aurix Autonomous Driving Domain Controller, which can perform sensor signal fusion (radar, camera, ultrasonic, and lidar), calculate the best driving strategy, and trigger actuators in the car to support enhanced ADAS functions such as traffic assistance and autonomous obstacle avoidance.
Vision chip: Enables ADAS functions such as lane departure warning, forward collision warning, traffic sign recognition, and pedestrian recognition.
Radar chips: 1) 77GHz long-range radar system, using SiGe (silicon germanium) technology to ensure high-frequency functionality and durability, can be used in collision avoidance systems; 2) 24GHz short/medium-range radar system, also using SiGe (silicon germanium) technology, can be used in blind spot monitoring systems.
In-vehicle 3D camera chip: Infineon launches the Real3 series of 3D image sensor chips, which use a time-of-flight (ToF) camera to measure the 3D environment, recognize driver behavior and transmit this information to ADAS, and can also improve the HMI experience such as gesture recognition.
▲ Infineon's revenue (millions of euros)
▲ Infineon's net profit (millions of euros)
▲ Infineon Aurix Autonomous Driving Controller Architecture Diagram
Renesas: A wide range of automotive MCUs and SoCs, with the R-Car platform supporting Level 4 autonomous driving.
Automotive electronics portfolio: Renesas automotive semiconductor products cover system-on-a-chip (SoC), power management, battery management, power devices, communication devices, video and display, etc.
Autonomous Driving Platform: Renesas introduces its R-Car autonomous driving SoC, featuring an ARM CPU and PowerVR GPU. The scalable hardware platform covers entry-level (R-Car E series), mid-range (R-Car M series), and high-end (R-Car H series), supporting various open-source software (Android, QNX, Linux, Windows, Genivi, etc.). In addition, there are also external camera chips (R-Car V series), internal camera chips (R-Car T series), smart cockpit chips (R-Car D series), and vehicle connectivity chips (R-Car W series).
▲ Renesas Operating Revenue (100 Million Yen)
▲ Renesas Net Profit (100 Million Yen)
▲ Renesas R-Car hardware and software platform
STMicroelectronics: A safety-focused semiconductor manufacturer whose ADAS products cover vision, radar, and connected vehicles.
Automotive electronics portfolio: STMicroelectronics' automotive semiconductor products cover advanced driver assistance systems (ADAS), vehicle comfort systems, chassis and safety systems, new energy vehicles, entertainment systems, mobility services, powertrain systems, communications and networks, etc.
Vision chips: These can be used for signal processing in front-view, rear-view, side-view, and in-vehicle camera systems. Furthermore, STMicroelectronics collaborated with Mobileye to develop the EyeQ series chips, responsible for chip manufacturing technology, dedicated memory, high-speed interface circuitry and system packaging design, as well as overall security architecture design.
Radar chips: 1) 77GHz long-range radar system, STRADA770 single-chip transceiver, covering 76-81GHz, suitable for adaptive cruise control (ACC), automatic emergency braking (AEB), forward collision warning (FCW), lane change assist (LCA), pedestrian detection (PD), etc.; 2) 24GHz short-range radar system, STRADA431 chip, including one transmitter and three receivers, suitable for blind spot detection (BSD), lane change assist (LCA), parking assist (PA), rear cross traffic alert (RCTA), collision mitigation braking (CMB), etc. For more details, please follow the WeChat official account 【车端】 (Car End).
Vehicle-to-Everything (V2X) Chips: Based on DSRC, STMicroelectronics and Israeli V2X manufacturer Autotalks began collaborating on V2X chipset development in 2014. The V2X solution showcased at CES 2018 integrates STMicroelectronics' Telemaco3 in-vehicle infotainment platform and Autotalks' CRATON2 chipset.
▲ STMicroelectronics Revenue (millions of US dollars)
▲ STMicroelectronics Net Profit (millions of US dollars)
▲ STMicroelectronics ADAS System
Texas Instruments: Provides open ADAS SoC solutions.
Automotive electronics portfolio: Texas Instruments' automotive semiconductor products cover advanced driver assistance systems (ADAS), infotainment systems and instrument clusters, body electronics and lighting, HEV/EV and powertrain systems.
Autonomous Driving Platform: Texas Instruments' main ADAS products are the TDAx series, including three SoCs: TDA2x, TDA3x, and TDA2Eco. Based on heterogeneous hardware and a general software architecture, they provide scalable and open ADA solutions. The TDA2x, released in October 2013, is primarily targeted at the mid-to-high-end market. It features two ARM Cortex-A15 cores and four Cortex-M4 cores, two TI C66x DSP cores, four EVE vision accelerator cores, and an Imagination SGX544 GPU. It is mainly used for processing information from front-facing cameras, including lane departure warning, collision detection, adaptive cruise control, and automatic parking systems.
Released in October 2014, the TDA3x is primarily aimed at the mid-to-low-end market. It features a reduced dual-core A15 processor and SGX544 GPU, and is mainly used for rear cameras, 2D or 2.5D surround view, etc. It supports a variety of ADAS algorithms such as lane line assist, adaptive cruise control, traffic sign recognition, pedestrian and object detection, forward collision warning and reversing collision warning.
Sensor chips include camera chips (front-view, rear-view, side-view, surround-view), radar chips (long-range, short-range, multi-mode), scanning lidar chips, ultrasonic chips, and sensor fusion chips, etc.
▲ Texas Instruments Revenue (millions of US dollars)
▲ Texas Instruments Net Profit (millions of US dollars)
▲ Texas Instruments TDAx Product Comparison
summary
The automobile has evolved from a "feature phone" to a "smartphone," and from "automotive electronics" to "autonomous driving." Strategically, we are optimistic about automotive chips as a core component within the intelligent driving industry chain.
Globally, major companies investing in the automotive chip industry include Nvidia, Intel, and Qualcomm; potential merger and acquisition targets include Infineon. Domestic companies are starting with products with lower safety requirements, such as in-vehicle entertainment systems, and are expected to gradually penetrate from aftermarket to OEM and from domestic OEMs to joint venture OEMs.
