Outcomes
Click here for video presentation
Fig: Multi-Layer Fully Connected Neural Network
Fig: The neuron architecture having multiply-accumulate unit followed by multiple activation functions (path-A and path-B).
Fig: The efficient recursive sign 8-bit precision RECON architecture configured by select and ‘ctr’ line for MAC [2] and AF [1] computation.
Fig: Proposed Block-level architecture for RECON. The red dotted line shows blocks required for MAC computation. The blue and green dotted lines represent the blocks used to compute tanh and sigmoid function [2].
Work Design Features:
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 Co-ordinate 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.
Work Contribution:
We propose a CORDIC-based design of an unsigned/signed computational unit which can compute both MAC and non-linear activation function.
We demonstrate how CORDIC is configured within RECON to operate in linear or hyperbolic rotation mode to solve arithmetic (for MAC) and trigonometric operations (for AF in [1]) respectively.
Optimization of the proposed CORDIC-based architecture in terms of area, power, and delay. We further employ power-gating approach and evaluate it with 45nm technology node to demonstrate power-savings. We analyse and discuss the impact of technology scaling on circuit’s physical parameters like area, power and delay
Publications:
Raut, Gopal, Shubham Rai, Santosh Kumar Vishvakarma, and Akash Kumar. “A CORDIC Based Configurable Activation Function for ANN Applications”. IEEE Computer Society Annual Symposium on VLSI (ISVLSI) 2020. DOI: 10.1109/ISVLSI49217.2020.00024. [Best paper nominee]
Raut, Gopal, Shubham Rai, Santosh Kumar Vishvakarma, and Akash Kumar. "RECON: Resource-Efficient CORDIC-based Neuron architecture." IEEE Open Journal of Circuits and Systems (OJCAS), 2 (2021): 170-181. DOI: 10.1109/OJCAS.2020.3042743.