Reconfigurable neural network computing chip

Abstract

The invention relates to the field of neural network chip architecture, in particular to a reconfigurable neural network computing chip. According to the invention, the intersection structure of the word lines and the bit lines forming the neural network is replaced by a structure C capable of switching between the neuron function and the synaptic function, and the function diversification is realized through flexible configuration by switching the neuron function and the synaptic function at different intersections; and the utilization rate of the calculation unit PE array can be maximized. And meanwhile, the architecture is suitable for application, the system scale can be expanded instead of a rigid architecture adopting the traditional chip design, and more flexible chip design is realized. And the function of the structure C can be adjusted, and configuration and connection are performed according to the requirements of the neural network to be mapped to form different shapes, so that the function of secondarily constructing a large neural network by taking a plurality of reconfigurable neural network computing chips as subunits is realized.

Description

technical field [0001] The invention relates to the field of neural network chip architecture, in particular to a reconfigurable neural network computing chip. Background technique [0002] The artificial neural network is a nonlinear and self-adaptive information processing system composed of a large number of interconnected processing units, and is composed of a large number of neurons, that is, nodes connected to each other. Each node represents a specific output function, called the activation function. Each connection between two nodes represents a weighted value for the signal passing through the connection, called weight, which is equivalent to the memory of the neural network. The output varies depending on how the network is connected, weight values ​​and activation functions. Neural network is a non-linear statistical data modeling tool, which is often used to model the complex relationship between input and output, or to explore the pattern of data, and finally ...

AI Technical Summary

Problems solved by technology

[0005] In view of the above-mentioned problems or deficiencies, in order to solve the problems that the existing neural network technology cannot take into account universality and relatively low energy efficiency, the present invention proposes a reconfigurable neural network computing chip, by combining the word lines and The intersection structure (neuron or synapse) of the bit line is replaced by a structure that can switch between neuron or synapse functions. Through the switching of neuron and synapse functions at different intersections, various functions can be realized with flexible configuration. , and can maximize the utilization of the computing unit PE array

Images