During the 2022 World Artificial Intelligence Conference (WAIC) on September 1, Chen Tianshi, the chairman of Chinese AI chip firm Cambricon, disclosed that the company is developing three autonomous driving chips spanning L2-L4 functions, which could meet differing computing power requirements from 10 to 1000 TOPS.
The three chips under development are “Si Yuan 590,” a new-generation cloud intelligent training chip, “SD5223,” an integrated chip for L2+ automated driving and parking, and “SD5226,” a SoC (system-on-chip) for L4 advanced automated driving multi-domain convergence platforms.
Si Yuan 590 adopts the brand-new architecture of MLUarch05, and its measured training performance has been greatly improved compared with the company’s flagship products on sale. It provides a larger memory capacity and higher memory bandwidth, and its input and output and inter-chip interconnection interface have also been greatly upgraded compared with previous generations.
SD5223, an integrated automated driving chip for the L2+ market, provides 16 TOPS computing power and higher DDR bandwidth. It supports auto-grade visual processing and low-power natural heat dissipation, and is suitable for 8M IFC/5V5R/10V5R and other multi-product forms.
In addition, according to changing demand in the market, Cambricon upgraded the specifications of its SoC chip SD5226, a high-level autonomous driving multi-domain convergence platform for the L4 market, and its computing power will reach more than 400 TOPS. It adopts 7nm process technology and an advanced AI architecture to adapt to algorithm evolution.
At the conference, Chen Tianshi said in his speech: “It is foreseeable that in the next five years, we will see four major trends of automated driving. First, L2+ automated driving systems will be rapidly popularized and will exist for a long time. In the next five years, the overall penetration rate of L2+ and above automated driving systems may exceed 50%. L4 automated driving in restricted scenarios will begin to be implemented, and L2+ to L4 automated driving will exist in parallel. The second trend is that the algorithms of automated driving are more complex, the amount of data processed increases exponentially, and the demand for computing power keeps rising. The third trend is that the vehicles, roads and cloud computing cooperate to realize a closed loop of big data, and the driving experience continues to upgrade. The fourth trend is that in order to meet the individualized needs of consumers and enhance the differentiated competitiveness of manufacturers, the demand for self-learning at the car end is constantly increasing, and cars will be very different from each other in terms of systems.”
At the conference, Cambricon regarded “collaboration of vehicles, roads and cloud computing” as the theme, displaying intelligent processors and chip products covering full scenarios, numerous industry solutions based on Cambricon’s chip development, and diversified AI hardware products. It demonstrated core technical strength and implementation abilities in the field of intelligent chip design.