A Ultra-Low Power System Design Method of AI Edge Computation

Abstract

With the vigorous development of AI computing chips and the popularity of cloud, edge and end computing modes, the demand of industrial applications for AI edge computing is becoming stronger and stronger, and the demand for AI computing power on the edge side is increasing. However, AI edge computing applications often need battery power, such as patrol robot, 4G IPC intelligent camera and so on. Therefore, the ultra-low power system design method of AI edge computing is becoming more and more important. Under the limited power supply, AI edge computing devices not only need to complete the AI edge reasoning algorithm in real time, but also need to enter the deep sleep state in time to reduce the unnecessary loss of power. Based on Hisilicon hi3559av100 AI edge computing chip, this paper studies the design methods of ultra-low power consumption systems such as clock, power supply network and business logic, so that the whole system can minimize the system energy loss and meet the requirements of the application system while calculating the AI algorithm in real time.