Method and device for realizing brain-like calculation based on vector instruction set

Abstract

The embodiment of the invention discloses a method and device for realizing brain-like calculation based on a vector instruction set, and the method comprises the steps: compiling a pulse neural network model according to a self-defined vector instruction set, and generating a vector instruction and a configuration instruction; a memory access unit of the brain-like computational vector processor performs memory access operation according to the configuration instruction; meanwhile, the calculation unit performs calculation according to the vector instruction. Through the above mode, the embodiment of the invention adopts the mode of the self-defined vector instruction set and the vector processor to optimize sparse calculation involved in spiking neural network simulation; a front end convenient to use is provided for a user, and the user can conveniently transplant codes from other pulse neural network simulators supporting Python; the memory access step and the calculation step in the calculation core are separated, the memory access step and the calculation step can be carried out at the same time, and efficient time overlapping of memory access and calculation is achieved.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of brain-like computing, and in particular, to a method and apparatus for implementing brain-like computing based on a vector instruction set. Background technique [0002] In the field of neuroscience, brain-like computing, as the main means of numerical simulation, is regarded as the most important research method besides theoretical and experimental research. From the perspective of computer architecture, the event-driven, storage-computing integration and other characteristics of brain-like computing have the potential to inspire the design of new processor architectures; brain-like computing is also regarded as one of the important evolution paths for the next generation of artificial intelligence. As the main computing model in the field of brain-like computing, Spiking Neural Network (SNN) introduces more biological reality, such as nerve impulses, membrane potential, etc. Featu...

AI Technical Summary

Problems solved by technology

This existing design method is efficient, but it also lacks programmability and has some design limitations; second, the implementation based on the general instruction set, such as the spinnaker chip of the University of Manchester, which implements SNN based on the ARM core The main body of computing, hardware modification is mainly concentrated on the interconnection level; the third is the implementation of general-purpose processors + custom acceleration hardware, such as Intel's loihi

However, there is no brain-inspired computing instruction set based on vector computing and the corresponding brain-inspired chip design

The main limitations of the three types of brain-like chips mentioned above are as follows: First, custom-structured chips lose programmability and design flexibility while pursuing performance, and have some design restrictions. Only simple neuron models are supported, or some complex neuron models are implemented using fixed-point numbers and low-precision floating-point numbers, but precise and loaded neuron models are not supported because they require a large number of floating-point number calculations and complex calculations Second, the implementation based on the general-purpose instruction set is not optimized for the SNN calculation process, and the performance is poor; third, the general-purpose processor and customized acceleration hardware are combined, and its optimization idea is mainly for the SNN cycle calculation process The hardware pipeline design, but the current design only supports fixed-point calculations, the accuracy is poor, and it cannot accurately simulate complex neuron models

Images