Introduction to the composition and standards of automotive SoC chips

This article is from the " 2024 Automotive SoC Chip Industry Analysis Report ". With the improvement of the level of automotive intelligence, the vehicle EE architecture has been upgraded from the previous distributed ECU architecture to the centralized domain controller architecture, and continues to evolve towards the central integrated architecture. In the distributed ECU architecture stage, the MCU is the core of computing and control;

In the centralized domain controller architecture stage, traditional MCU chips can no longer meet the requirements of throughput capacity and faster data processing capabilities for large amounts of heterogeneous data. Therefore, SoC chips with higher data transmission efficiency and greater computing power have become the inevitable choice for domain controller main control chips.

1 Definition of Automotive SoC Chip

1) Basic Definition

Automotive-grade computing chips can be divided into two categories according to the scale of integration: MCU and SoC. Among them, MCU is also called "single-chip microcomputer chip", which integrates processor, memory, input/output interface and other peripherals. It is often used in embedded systems with simple control tasks and high real-time performance. The operating systems commonly run on automotive MCUs are AUTOSAR CP and FreeRTOS, and usually do not support running highly complex operating systems.

SoC chips are system-level chips. Compared with MCUs, they integrate more heterogeneous processing units, have more complex structural designs, and have stronger processing and computing capabilities. They are suitable for multi-tasking and application scenarios with more complex computing tasks. In-vehicle SoCs can run more complex operating systems, including QNX, Linux, Android, and AUTOSAR AP.

2) Hardware composition

The internal structure of an automotive SoC chip usually includes the following major modules: processor, memory, peripheral I/O, etc.

A. Processor - The processor inside the vehicle SoC chip usually includes the following unit modules:

General logic operation unit: usually implemented based on CPU, mainly responsible for some logic operation tasks, used to manage software and hardware resources, complete task scheduling and external resource access, etc., to realize system-level functional logic, diagnostic logic and shadow mode data mining functions, etc. Some typical applications include: decision-making planning algorithms based on optimization, vehicle control algorithms, etc.

AI acceleration unit: It is usually implemented based on a neural network processor such as NPU, and is responsible for large-scale floating-point parallel computing needs; as an accelerator for neural network algorithms, it is mainly responsible for handling AI computing needs.

Image/video processing unit: usually based on DSP, ISP, GPU and other processors. As a visual processing chip, ISP's main function is to adjust the image signal output by the camera, including AE (auto exposure), AF (auto focus), AWB (auto white balance), image denoising, etc.; DSP is a microprocessor with a special structure. Compared with general-purpose CPU, it is more suitable for processing tasks with high computational intensity. Typical applications include: traditional CV image processing, accelerated processing of some custom operators, etc.; GPU has strong floating-point computing capabilities and is mainly used for applications such as 3D rendering and splicing of images.

Hardware Security Module HSM: used to provide encryption and decryption services for applications, manage sensitive information and assets, protect encryption keys, etc.

Satety MCU: Mainly used to monitor the status and communication of each hardware module inside the SoC in real time, and to report errors in time when problems occur, thereby ensuring the functional safety of the entire system.

B. Internal memory: includes two categories: volatile memory and non-volatile memory.

Volatile memory: When the power is off (for example, when the system is shut down normally or unexpectedly), the data will be lost, that is, the stored data cannot be retained. It is mainly used to temporarily store programs and data being processed. Common memory types in automotive SoCs include SRAM and DRAM (DDR, LPDDR, etc.).

Non-volatile memory: It can still save stored data when the power is off. It is mainly used to store fixed data, firmware programs and other data that generally do not need to be changed frequently. Common memory types inside automotive SoCs include NAND Flash (eMMC, UFS, etc.) and Nor Flash.

C Peripheral I/O: including general data interface, camera signal interface, audio interface and display interface, etc.

• Common data interfaces: PCIe, LVDS, USB, SATA, CAN/CAN-FD, Ethernet, etc.
  
  
• Camera signal interface: MIPI-CSI-2, GMSL, FPD Link, etc.
  
  
• Audio interface: I2S, TDM, SPDIP, etc.
  
  
• Display interface: DP, HDMI, etc.
  
  

2. Vehicle SoC chip performance requirements

1) Important parameter indicators

To measure the performance of automotive SoC chips, it is necessary to comprehensively consider multiple dimensions such as AI computing power, CPU computing power, GPU computing power, storage bandwidth, power consumption, and manufacturing process.

a. AI computing power: usually refers to the computing power of MAC instructions (multiply-accumulate). The MAC instruction operation itself is strongly related to the data type. Under different data precision conditions, the measured AI computing power will be quite different. The computing power usually claimed by the company generally refers to the theoretical peak value of the chip's computing power, expressed in TOPS, and generally defaults to Int8 as the computing power quantification standard.

But we can't just look at the theoretical computing power value on the surface. In specific usage scenarios, people are more concerned about the actual effective computing power of the chip, that is, the "computing power utilization" of the chip. Taking smart driving applications as an example, the actual computing power utilization of SoC chips will vary due to differences in image resolution, network structure, etc.

b. Storage bandwidth: During the data processing process, data needs to be continuously "read" from the memory unit to the processor unit, and then "written" back to the memory unit after processing. The frequent migration of data between the memory and the processor will bring serious transmission power consumption problems. Some industry insiders have pointed out that 90% of the power consumption and delay of AI computing are caused by data transfer.

The memory bandwidth of a chip is determined by two factors: the memory itself and the number of memory channels of the chip. The size of the memory bandwidth determines the speed and number of data transfers. Therefore, the size of the storage system bandwidth also determines the actual computing power of the chip to a certain extent.

c. Power consumption: includes dynamic power consumption and static power consumption. Dynamic power consumption is the power consumption loss caused by the change of signal value, which consists of two parts: switching power consumption and internal power consumption. Static power consumption is the power consumed when the device is still powered on but there is no signal value change.

The power consumption of a chip is related to factors such as hardware architecture, layout and wiring, process technology, and computing power. Under the same conditions, the more advanced the process technology, the lower the power consumption of the chip; similarly, the more powerful the chip, the greater the power consumption. Excessive power consumption means greater heat dissipation, and it may be necessary to install a water cooling system, thereby increasing the overall BOM cost.

2) Automotive-grade requirements

According to the application scenarios in daily life, chips can be roughly divided into three categories: consumer grade, industrial grade, and automotive grade. Different application scenarios will lead to differences in the goal setting and implementation methods of chips in design, production, certification, and other links. Compared with consumer and industrial grades, automotive-grade chips have a harsher working environment, lower error tolerance, longer service life requirements, and longer supply life cycle.

Overall, automotive-grade chips are characterized by high reliability, high security, and high stability. Automotive chips need to undergo a series of rigorous testing and certification to ensure that they meet the relevant requirements of automotive standards before they can be put into mass production. Chip automotive certification standards usually include the following three dimensions of control: quality management system certification IATF16949, reliability standard AEC-Q100, and functional safety standard ISO 26262.