Abstract:Crossbar memory arrays have been touted as the workhorse of in-memory computing (IMC)-based acceleration of Deep Neural Networks (DNNs), but the associated hardware non-idealities limit their efficacy. To address this, cross-layer design solutions that reduce the impact of hardware non-idealities on DNN accuracy are needed. In Part 1 of this paper, we established the co-optimization strategies for various memory technologies and their crossbar arrays, and conducted a comparative technology evaluation in the context of IMC robustness. In this part, we analyze various design knobs such as array size and bit-slice (number of bits per device) and their impact on the performance of 8T SRAM, ferroelectric transistor (FeFET), Resistive RAM (ReRAM) and spin-orbit-torque magnetic RAM (SOT-MRAM) in the context of inference accuracy at 7nm technology node. Further, we study the effect of circuit design solutions such as Partial Wordline Activation (PWA) and custom ADC reference levels that reduce the hardware non-idealities and comparatively analyze the response of each technology to such accuracy enhancing techniques. Our results on ResNet-20 (with CIFAR-10) show that PWA increases accuracy by up to 32.56% while custom ADC reference levels yield up to 31.62% accuracy enhancement. We observe that compared to the other technologies, FeFET, by virtue of its small layout height and high distinguishability of its memory states, is best suited for large arrays. For higher bit-slices and a more complex dataset (ResNet-50 with Cifar-100) we found that ReRAM matches the performance of FeFET.

Hiearchical Crossbar Design for ReRAM Based Write Variation Inhibition On-Chip Learning

Bayesian Inference Based Robust Computing on Memristor Crossbar

Circuit Modeling for RRAM-Based Neuromorphic Chip Crossbar Array with and Without Write-Verify Scheme.

Aging Aware Retraining for Memristor-based Neuromorphic Computing

BRoCoM: A Bayesian Framework for Robust Computing on Memristor Crossbar.

Reliable Memristor-based Neuromorphic Design Using Variation- and Defect-Aware Training

Robustness of Neuromorphic Computing with RRAM-based Crossbars and Optical Neural Networks.

Impact of the Interconnection Structure on RRAM-Based Crossbar Array in Neuromorphic Chip

Hardware-aware Training Techniques for Improving Robustness of Ex-Situ Neural Network Transfer onto Passive TiO2 ReRAM Crossbars

Compensation Architecture to Alleviate Noise Effects in RRAM-based Computing-in-memory Chips with Residual Resource

Enhance the Robustness to Time Dependent Variability of ReRAM-Based Neuromorphic Computing Systems with Regularization and 2R Synapse

Compensation Architecture Design Utilizing Residual Resource to Mitigate Impacts of Nonidealities in RRAM-based Computing-in-memory Chips

Modeling and Analysis of a Large-Scale Memristor Crossbar Array for Neuromorphic Chip

An Improved RRAM-Based Binarized Neural Network with High Variation-Tolerated Forward/Backward Propagation Module

A Circuit-Algorithm Codesign Method to Reduce the Accuracy Drop of RRAM Based Computing-in-memory Chip

Mitigate IR-Drop Effect by Modulating Neuron Activation Functions for Implementing Neural Networks on Memristor Crossbar Arrays

An On-chip Layer-wise Training Method for RRAM Based Computing-in-memory Chips.

Bit-Aware Fault-Tolerant Hybrid Retraining and Remapping Schemes for RRAM-Based Computing-in-Memory Systems

Online Soft Error Tolerance in ReRAM Crossbars for Deep Learning Accelerators

Intra-array Non-Idealities Modeling and Algorithm Optimization for RRAM-based Computing-in-Memory Applications

Comparative Evaluation of Memory Technologies for Synaptic Crossbar Arrays- Part 2: Design Knobs and DNN Accuracy Trends