Survey of Embedded Neural Network Accelerator for Intelligent Aerospace Applications

With the increasing complexity of space missions, the demand for intelligent technologies in aerospace is rapidly growing. Neural network accelerators play an indispensable role in intelligent aerospace applications as dedicated hardware devices for accelerating neural network inference. In this paper, we have explored the challenges associated with selecting the appropriate Neural Processing Unit (NPU) for aerospace intelligent applications. We have shown that the peak specifications of an NPU may not accurately represent its actual performance. To overcome this challenge, we conducted a survey of neural network accelerators that have the potential to be used in our aerospace intelligent field. In addition, we conducted practical tests on specific algorithms to gain a deeper understanding of their performance. The results of our research provide valuable insights for the selection of embedded neural network accelerators for future aerospace intelligent applications. In conclusion, we summarized our findings and looked forward to the future development directions of neural network accelerators for space intelligent applications.