Research on photonic neural network and high-precision image recognition technology

In the era of big data, the amount of data is increasing day by day. The existing processors based on electronic chips are confronted with the problems of limited speed and excessive energy consumption when dealing with massive data processing. Low scalability of the current high-speed photonic neural network computing chips limits its further development. This paper proposes an ultra-high-speed scalable photonic neural network computing chip and integration technology for artificial intelligence. Dividing the photonic computing chip into light source module, electro-optical conversion module, routing module, computational unit module, etc., can significantly improve the computing performance in different scenarios. The chip's digit recognition accuracy can reach 96.74% and 97.67% for tested and calculated values, respectively. The computing density is as high as 1.523 TOPS/mm2 at 40 computational units. It provides a new solution for ultra-high-speed optoelectronic information processing