In-memory Implementation of On-chip Trainable and Scalable ANN for AI/ML Applications

TL;DR

This paper introduces an in-memory computing architecture using 6T SRAM for scalable, trainable neural networks that significantly improves energy efficiency and throughput for AI/ML applications by integrating training and inference on-chip.

Contribution

It presents a novel in-memory architecture that enables on-chip training and inference of multi-layer perceptrons, reducing energy consumption and increasing processing speed.

Findings

Achieved approximately 46 times more energy efficiency per MAC operation.

Implemented backpropagation with new in-memory building blocks.

Validated on IRIS dataset with promising results.

Abstract

Traditional von Neumann architecture based processors become inefficient in terms of energy and throughput as they involve separate processing and memory units, also known as~\textit{memory wall}. The memory wall problem is further exacerbated when massive parallelism and frequent data movement are required between processing and memory units for real-time implementation of artificial neural network (ANN) that enables many intelligent applications. One of the most promising approach to address the memory wall problem is to carry out computations inside the memory core itself that enhances the memory bandwidth and energy efficiency for extensive computations. This paper presents an in-memory computing architecture for ANN enabling artificial intelligence (AI) and machine learning (ML) applications. The proposed architecture utilizes deep in-memory architecture based on standard six transistor (6T) static random access memory (SRAM) core for the implementation of a multi-layered perceptron. Our novel on-chip training and inference in-memory architecture reduces energy cost and enhances throughput by simultaneously accessing the multiple rows of SRAM array per precharge cycle and eliminating the frequent access of data. The proposed architecture realizes backpropagation which is the keystone during the network training using newly proposed different building blocks such as weight updation, analog multiplication, error calculation, signed analog to digital conversion, and other necessary signal control units. The proposed architecture was trained and tested on the IRIS dataset which exhibits $\approx46\times$ more energy efficient per MAC (multiply and accumulate) operation compared to earlier classifiers.