RECON: Resource-Efficient CORDIC-Based Neuron Architecture

Abstract

Contemporary hardware implementations of artificial neural networks face the burden of excess area requirement due to resource-intensive elements such as multiplier and non-linear activation functions. The present work addresses this challenge by proposing a resource-efficient Coordinate Rotation Digital Computer (CORDIC)-based neuron architecture (RECON) which can be configured to compute both multiply-accumulate (MAC) and non-linear activation function (AF) operations. The CORDIC-based architecture uses linear and trigonometric relationships to realize MAC and AF operations respectively. The proposed design is synthesized and verified at 45nm technology using Cadence Virtuoso for all physical parameters. Implementation of the signed fixed-point 8-bit MAC using our design, shows 60% less area, latency, and power product (ALP) and shows improvement by 38% in area, 27% in power dissipation, and 15% in latency with respect to the state-of-the-art MAC design. Further, Monte-Carlo simulations for process-variations and device-mismatch are performed for both the proposed model and the state-of-the-art to evaluate expectations of functions of randomness in dynamic power variation. The dynamic power variation for our design shows that worst-case mean is 189.73μW which is 63% of the state-of-the-art. INDEX TERMS AF, CORDIC, configurable architecture, MAC, neural network.

CORDIC Hardware Acceleration Using DMA-Based ISA Extension

“…Previous work have shown the benefit of using CORDIC for AI algorithms in terms of power and area. G. Raut et al [7,8] presented an optimized CORDIC-based architecture to enable computations required in the neural networks. The CORDIC engine was employed in hyperbolic rotation mode to realize both tanh and sigmoid activation functions.…”

“…The Coordinate Rotation Digital Computer (CORDIC) algorithm is a well-known algorithm used for computing a wide range of mathematical functions and is applied to computer vision and DSP that require heavy computational functions [6]. CORDIC is also used as an efficient NN computation engine for implementing multiply-and-accumulate (MAC) and nonlinear neuron activation function [7,8]. Efficient implementations of the CORDIC algorithm to calculate the differential equations of the neurons in Spiking Neural Network (SNN) are presented in [9,10].…”

An accurate and noninvasive skin cancer screening based on imaging technique

“…In convolutional layer and fully connected layer, AF played a major role in learning neural nets. 15 In this section, we have modeled a function which has overcome vanishing gradient problem as compared with tanh, sigmoid and ReLU, and so forth AFs. The proposed AF is given by Equations (1a) and (1b).…”

“…Fast and inexpensive detection can effectively be done with artificial intelligence and machine learning capabilities, which is a shining potential to spare time and mitigate errors and thus saving millions of lives in the long run. The efficient neuron computation technique that improve performance have proposed in References 15–17. In particular, convolutional neural networks (CNNs) can automate most of the diagnosis process with equal or more accuracy than the current traditional methods, which is explained in depth later in the paper.…”

An Empirical Evaluation of Enhanced Performance Softmax Function in Deep Learning

“…Research in Multiply-ACcumulate (MAC) and non-linear activation functions (AFs) using the iterative CORDIC algorithm has shown profound results in smaller areas. Still, throughput is a significant concern [14], [17]- [19].…”