Automated design method, device and optimization method applied for neural network processor

Abstract

The invention discloses an automated design method, device and optimization method applied for a neural network processor. The method comprises the steps that neural network model topological structure configuration files and hardware resource constraint files are obtained, wherein the hardware resource constraint files comprise target circuit area consumption, target circuit power consumption and target circuit working frequency; a neural network processor hardware architecture is generated according to the neural network model topological structure configuration files and the hardware resource constraint files, and hardware architecture description files are generated; according to a neural network model topological structure, the hardware resource constraint files and the hardware architecture description files, modes of data scheduling, storage and calculation are optimized, and corresponding control description files are generated; according to the hardware architecture description files and the control description files, cell libraries meet the design requirements are found in constructed reusable neural network cell libraries, corresponding control logic and a corresponding hardware circuit description language are generated, and the hardware circuit description language is transformed into a hardware circuit.

Description

technical field [0001] The invention relates to the technical field of neural network processor architecture, in particular to an automatic design method, device and optimization method applicable to neural network processors. Background technique [0002] With the rapid development of related technologies in the field of artificial intelligence, deep learning, as an interdisciplinary product of computer science and life science, has excellent performance in solving advanced abstract cognitive problems, so it has become a research hotspot in academia and industry. In order to improve the computing performance of the neural network and adapt to more complex application problems, the scale of the neural network is constantly expanding, and the amount of calculation, data volume and computing energy consumption are also increasing. Searching for neural network computing methods and devices with high performance and low energy consumption has become a hot spot for researchers ...

AI Technical Summary

Problems solved by technology

[0003] At present, the real-time task analysis using deep neural networks mostly relies on large-scale high-performance processors or general-purpose graphics processors. These devices have high cost and high power consumption. When they are applied to portable smart devices, they have large circuit scale, high energy consumption and expensive products. Therefore, for the application of energy-efficient real-time processing in embedded devices and small and low-cost data centers, it is more effective to use dedicated neural network processors to accelerate neural network model calculations instead of software. However, the topology and parameter design of the neural network model will change according to different application scenarios. In addition, the development and change of the neural network model is very fast, providing a network that can face various application scenarios and cover various neural network models. It is very difficult to use a general-purpose high-efficiency neural network processor, which brings great changes for high-level application developers to design hardware acceleration solutions for different application requirements

[0004] At present, the existing neural network hardware acceleration technology includes two methods: Application Specific Integrated Circuit (ASIC) chip and Fie

Images