The application of artificial intelligence (AI) has spread, and the corresponding computing needs have become a hot topic. AI computing can be divided into two main applications. One is server/cloud computing with performance as the primary consideration, and the other is terminal applications with lower computing power requirements, which require low power consumption and high efficiency. IP manufacturer Imagination’s products include GPU, Neural Network Accelerator (NNA) and Ethernet (EPP). The GPU architecture focuses on power saving, and has long been invested in the four major fields of mobile devices, consumer electronics, automobiles and data centers.
The importance of GPUs for cars is rising
Imagination’s GPU architecture has two major features: multiprocessing and independent cutting. During multitasking, up to 8 jobs can be executed at a time, or it can be cut into up to 8 independent GPUs to complete the work when the operating system is independent. For example, the capacity of the memory can be cut for different tasks and will not occupy each other, ensuring safety and flexibility in use.
Automotive is one of the markets that Imagination is currently actively investing in. Renesas and Texas Instruments are both early customers that have cooperated with Imagination since 2004. At present, Imagination’s GPU architecture accounts for about 40-50% of the automotive IC market. Starting in 2019, the partners have expanded to Telechip in China and SemiDrive in South Korea. Affected by the rising trend of electric vehicles, and the consensus of many countries around the world to ban the sale of fuel vehicles in 2025, the increase in the amount of automotive electronics, and the application trend of driving assistance, autonomous driving and automotive entertainment systems, GPUs are used in automotive ICs. The role is more important.
Lin Huanxiang, business director of Imagination Taiwan, mentioned that the current self-driving level is between Level 2 and 3, and it is estimated that it will reach Level 5 in 2024. By then, the number of sensors in self-driving and electric vehicle applications will double, and automotive electronics will increase With a growth of at least four times, the car of the future can be said to be a walking data center.
Lin Huanxiang, business director of Imagination Taiwan, believes that the trend of self-driving and electrification will increase the consumption of automotive electronics.
XS GPU IP has passed ISO 26262 certification. In actual use, the dashboard will distinguish areas that need priority processing. For example, GPS navigation and speedometer will affect driving safety, so safety and stability must be ensured, which is the area that GPU prioritizes processing. . Other functions such as entertainment will not affect security, so the GPU will process data in a random order. In addition, dashboards with multiple screens have their own independent operating systems, so if a screen that is not in the priority processing area has a functional problem, it will be restarted in accordance with the general procedure and will not affect the important functions of the dashboard.
NNA enhanced image recognition
Lin Huanxiang said that in the past there have been GPU vendors with computing power advantages in server/vehicle use, but now the GPU field has opened a new battle for high-performance computing (HPC). Faced with emerging computing power competition, ICs based on the Imagination NNA architecture With an 8-core, 5nm process, it can reach a computing power of 30 watts per second.
Series4 NNA further strengthens the image integration of GPUs for cars. Taking night driving as an example, multiple cameras on the car receive image data at the same time. After GPU processing and integration, NNA performs image recognition to assist driving decision-making. In addition to road condition analysis, NNA can also detect driving behaviors, such as issuing warnings when driving fatigue and not looking directly at the road ahead. Automatic parking is also one of the applications of NNA. In order to meet the ever-expanding computing power demand of AI, the Series4NX-MC product series can provide up to 100 TOPS of computing power. NNA is used with GPU cutting work, and the work is divided and completed one by one according to the sequence, which effectively improves the efficiency utilization rate.
NNA can warn when driving behavior is abnormal