An Analog-Aware Computing Architecture for Neural Network Algorithms

Abstract

The invention discloses a neural network algorithm-oriented simulation perception computing framework, comprising: a neuron value buffer configured to buffer sample parameters of an object to be analyzed; a synapse weight buffer configured to store parameters corresponding to the sample parameters Synaptic weight: an analog computing processing module configured to perform feature extraction and feature classification on the object to be analyzed in the analog domain according to the synaptic weight and sample parameters. The present invention realizes the feature extraction and feature classification of the object to be analyzed, and the calculation of the sample parameters and synaptic weights is carried out in the analog domain, which has the characteristics of high energy efficiency. At the expense of further reducing energy consumption.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a neural network algorithm-oriented simulation perception computing architecture. Background technique [0002] In recent years, artificial intelligence (AI for short) has developed vigorously. Neural network, as one of the most effective ways to realize artificial intelligence, has attracted more and more attention from academia and industry. Neural networks have been widely used in the fields of image, video and speech recognition. [0003] Among many neural networks, Convolutional Neural Network (CNN for short) and Deep Neural Networks (DNN for short) are the most widely used. Both CNN and DNN are computationally intensive neural networks. The large-scale neural network requires high computing power of the platform. Especially when processing high-dimensional data of images and videos, the data transmission rate may exceed the real-time processing capability of the comp...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem that the ASIC in the prior art can only realize the feature extraction of the convolutional neural network, but cannot realize the feature classification and the high energy consumption of the ASIC, the present invention provides an analog sensing algorithm oriented to the neural network algorithm. computing architecture

Images