Multi-layer feedforward neural network parallel accelerator

Abstract

A multi-layer feedforward neural network parallel accelerator provided by the invention comprises: a main control module, which controls the entire calculation flow; a coefficient address generation module, which generates an address of the coefficient, outputs the address of the coefficient, receives coefficient data of the storage control module, and outputs the coefficient data after splitting;an operand address generation module, which generates and outputs the storage address of the operand, receives the operand data of the storage control module, and outputs the data after splitting; afeedforward network calculation module, which receives the split coefficient data and the operand data and contains a plurality of parallel computing units; and a neuron address generating module, which receives the neuron data calculated by the feedforward network computing module, generates the storage address of the neuron and neuron data, and outputs the storage address of the neuron and the neuron data. The invention has the beneficial effects of good scalability, adjustable parallelism, high acceleration ratio, support for pipeline operation, and low resource consumption.

Description

technical field [0001] The invention belongs to the field of hardware acceleration, in particular to a multilayer feedforward neural network parallel accelerator. Background technique [0002] Artificial intelligence algorithms make people's lives more and more convenient, but with the continuous change of application requirements, the algorithm complexity is also getting higher and higher. With the increase of algorithm complexity, the requirements of artificial intelligence algorithms for hardware power consumption and speed are also increasing. A neural network is an extensively parallel interconnected network of adaptive, simple units organized to simulate the interaction of biological nervous systems with real-world objects. Multi-layer feed-forward neural network has been more and more widely used in operand retrieval, machine vision, security monitoring and other fields. Note that the input of the neuron in the jth hidden layer of the multi-layer feedforward neural ...

AI Technical Summary

Problems solved by technology

However, both GPU and CPU are general-purpose processors, and both need to perform the process of fetching instructions, decoding instructions, and executing instructions. In this way, the processing of the underlying IO is shielded, so that the software and hardware are decoupled, but it brings data movement and Computing cannot achieve higher efficiency

The difference in the energy consumption ratio between the GPU and the CPU is mainly due to the fact that most of the transistors in the CPU are used in the Cache and the control logic unit. Therefore, compared with the GPU, the CPU has less redundancy for algorithms that are computationally intensive and have low computational complexity. The remaining transistors can't play a role, and the energy consumption is lower than that of CPU than GPU

The above two implementation methods consume a lot of energy and have poor scalability. How to make the multi-layer feedforward neural network run faster, save more energy, and have better scalability has become a hot issue.

Images