The chip of vehicle specification level is divided into control class, power class, sensor class and other classes by function; At present, most of the chip giants come from foreign manufacturers. The control chips include AI chips and MCU (single chip computer) chips. We often say that the car gauge chip refers more to AI chips, which are system-level SOC chips with the most powerful performance.
1、 Vehicle specification chip standard
The chip standard of car specification level is much higher than that of consumer level, and the certification process is long.
1. The working environment is even worse: compared with consumer chips and general industrial chips, the working environment temperature of automotive chips ranges from minus 40 to minus 155 degrees Celsius, which is vulnerable to wide light, high vibration, dust and electromagnetic interference.
2. High reliability and safety requirements: most cars are designed to last for 15 years or 200000 kilometers, which is longer than the service life of consumer electronics. Under the same reliability requirements, the more components and links of the system, the higher the reliability requirements of the components.
3. The certification process of car specification chip is long: it takes about 2 years for a chip to complete the car specification certification, and generally has a supply cycle of 5-10 years after entering the car enterprise supply chain.
2、 What are the chip of vehicle specification?
1. FPGA, Field Programmable Gate Array, programmable logic gate array, high computational power, small-scale customized development and test scenarios are more applicable. Users can define the wiring of their internal structure by burning the configuration file, so as to achieve customized circuits. Its chip mass production cost is high, and its energy efficiency is poor, which is not comparable with ASIC special chip; It is more suitable for scientific research and enterprise development. Once the scheme is determined, its cost advantage will no longer be prominent. Programmable logic, high computing efficiency, closer to the bottom IO, and logic programmable through redundant transistors and wiring. The representative manufacturers include Celine, Altra (acquired by Intel) and Shenjian Technology.
2. ASIC, Application-Specific Integrated Circuit, is an integrated circuit designed for special purposes, with the highest computational power and excellent energy efficiency ratio. It is mainly aimed at the needs of specific users, suitable for relatively single large-scale application scenarios, and runs faster than FPGA under the same conditions. However, at the architecture level, specific intelligent algorithms are hardened and supported. The instruction set is simple or the instruction is completely solidified. If the scene changes, this type of AI chip will no longer be applicable and has the requirement of upgrading. At present, AI algorithms are constantly changing, and a large number of algorithms are developed every year, which is not applicable. So there is no real ASIC chip at this stage. The transistor is customized according to the algorithm, with low power consumption, high computational efficiency and high computational efficiency. For chips specially customized for specific needs, the programming framework is fixed, and the replacement algorithm needs to be redesigned
3. CPU: central processing unit. 70% of the transistors are used to build the cache, and some control units are also used. The number of calculation units is small, which is suitable for scenes with complex operations and logic, but the amount is small. It is irreplaceable; The calculation power is the lowest and the energy efficiency ratio is poor; But it is universal in all fields.
4. GPU: Most transistors build computing units with low computational complexity and are suitable for large-scale parallel computing. Support various programming frameworks, more general than FPGA and ASIC; High computational power and medium energy efficiency ratio; It is widely used in various fields of graphic processing, numerical simulation and machine learning algorithms.