Embedded AI Accelerator Chips

This chapter introduces the implementation of AI acceleration chips in detail. First, the advantages and disadvantages of several AI chip types such as GPU, DSP, FPGA, and ASIC are introduced. Then, several representative embedded AI chips are introduced, including NVIDIA Jetson, Intel Movidius, Google Edge TPU, XILINX DPU, and ARM Ethos NPU, Qualcomm Hexagon, etc. are taken as examples, and their working principles, internal structures, specifications, and application scenarios are introduced in detail. Finally, the above-mentioned main embedded AI accelerators are compared in terms of AI inference performance, power consumption, and inference performance per watt to facilitate embedded system developers to choose the appropriate AI acceleration chip according to their needs.