Data processing device and artificial intelligence processor

Abstract

The invention relates to a data processing device and an artificial intelligence processor, the data processing device is applied to a processing core of the artificial intelligence processor, and thedata processing device is connected with a storage module and a calculation module, and comprises: an address generation module used for generating a first address of first data to be read accordingto a set address generation mode; the data conversion module that is used for reading the first data from the storage module according to the first address and converting the read first data to obtainconverted second data, and sending the second data to a calculation module; and the control module that is used for determining a corresponding address generation mode according to a preset processing category and a data category and controlling the address generation module to generate an address. By generating the address corresponding to the data according to the processing category and the data category and performing data conversion, the access speed of the data can be increased, and the calculation efficiency of the neuromorphic chip is improved.

Description

technical field [0001] The present disclosure relates to the field of computer technology, in particular to a data processing device and an artificial intelligence processor. Background technique [0002] In recent years, the field of neuromorphic computing has developed rapidly, and hardware circuits can be used to directly construct neural networks to simulate the functions of the brain. Specifically, neuromorphic chips can be used to build a large-scale, parallel, low-energy computing platform that can support complex pattern learning. On this basis, there are more and more types of neural networks represented by artificial neural networks (ANNs, Artificial Neural Network) and spiking neural networks (SNNs, Spiking Neural Network), and the scale is getting larger and larger. Computing power puts forward higher requirements, so neuromorphic chips should have higher data processing and organization capabilities. [0003] However, the types of data that traditional neuromo...

AI Technical Summary

Problems solved by technology

[0003] However, the types of data that traditional neuromorphic chips can process are limited, and some are limited to matrices or low-dimensional tensors

In addition, the data precision that traditional neuromorphic chips can process is limited, making it impossible to apply to more types of networks.

For example, the spiking neural network as the third generation of neural network, its data is encoded by the spatiotemporal information of neuron spike signal, which can be described by three-valued data (for example, 0, 1 and -1), while most of the current neuromorphic While the chip supports ternary data operations, it cannot efficiently support multi-precision (for example, 8-bit or more) data calculations, nor can it convert with other types of data according to the calculation needs of neural networks

In addition, the modules used for data processing in traditional neuromorphic chips have a complex structure, and different processing methods are used for different computing tasks and data types (such as vectors, matrices, and tensors), which greatly limits the use of neuromorphic chips. Computational efficiency and inference efficiency of neural networks, and poor scalability

Images