Neural network operation system, method, device and storage medium

Abstract

The invention provides a neural network operation system, a method, a device and a storage medium, which are used for reducing data migration in a neural network calculation process and improving the operation efficiency of a neural network processor. The neural network operation system comprises at least two neural network processing units, a first storage unit and a second storage unit, and the first storage unit is used for storing input data and output data of a neural network and operation parameters required by operation of each layer of neural network; the second storage unit is used for providing input caches and output caches for each neural network unit in the at least two neural network processing units, and each neural network processing unit comprises two input caches and two output caches, two output caches of one neural network processing unit in two adjacent neural network processing units are two input caches of the other neural network processing unit; the at least two neural network processing units are annularly connected.

Description

technical field [0001] The present invention relates to a processor, in particular to a neural network computing system, method, device and storage medium. Background technique [0002] In the process of running the neural network model, the neural network processor (Neural network Processing Unit, NPU) can provide sufficient computing power for the training and reasoning of the deep neural network. Since the deep neural network is calculated by layer, the input and output data of each layer is large, so a considerable amount of memory needs to be provided to store intermediate data and weight data, resulting in a large amount of resources being wasted in data handling between cache and memory. In this process, a storage wall problem occurs, which reduces the computing efficiency of the processor. Contents of the invention [0003] Embodiments of the present application provide a neural network computing system, method, device, and storage medium, which are used to reduce...

AI Technical Summary

Problems solved by technology

Since the deep neural network is calculated by layer, the input and output data of each layer is large, so a considerable amount of memory needs to be provided to store intermediate data and weight data, resulting in a large amount of resources being wasted in data handling between cache and memory. In the process, the storage wall problem occurs, which reduces the computing efficiency of the processor

Images