Hardware implementation device and method for activation function

Abstract

The invention discloses a hardware implementation device and method of an activation function, and belongs to the technical field of hardware implementation of function calculation. Aiming at the problem that in the prior art, high performance and hardware resources are difficult to balance when the hardware of a conventional method achieves an activation function, the device comprises an e-exponential calculation module and a CORDIC division module; an input signal calculates an e-exponential e-x or an e-exponential e2x in the e-exponential calculation module through a small index lookup table with 2 as the base and shifting and additive operation. Based on carry-save addition, an adder achieves the calculation function like I1 + I2-I3; based on a CORDIC algorithm, a CORDIC division module calculates a division function result shown in the specification through multiple times of iteration. The circuit structure is simple, no complex control logic exists, circuit layout and wiring areconvenient, the calculation precision is high, sigmoid and tanh activation function calculation in any input range is supported during calculation, and the problem that a traditional hardware implementation method cannot balance performance and hardware resources is solved.

Description

technical field [0001] The present invention relates to the technical field of hardware implementation of function computing, and more specifically, to a hardware implementation device and method for an activation function. Background technique [0002] With the rapid development of artificial intelligence and other fields, neural networks have become a research hotspot in recent years. Artificial neural network is a mathematical model that uses a structure similar to that of brain synaptic connections for information processing. Its main task is to build a practical artificial neural network model based on the principles of biological neural networks and the needs of practical applications, and to design corresponding Learning algorithms simulate certain intelligent activities of the human brain, and then implement them technically to solve practical problems. [0003] As an important part of the neural network algorithm, the hardware implementation of the activation funct...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the problem in the prior art that it is difficult to balance high performance and hardware resources in the traditional hardware implementation of activation functions, the present invention provides a hardware implementation device and method for activation functions, which can better balance hardware performance and hardware resources The use of less hardware resources can ensure the higher performance of the hardware

Images