Ultra Low Power 3D-Embedded Convolutional Neural Network Cube Based on $\alpha$-IGZO Nanosheet and Bi-Layer Resistive Memory

TL;DR

This paper introduces a 3D-embedded neuromorphic system using $1$-IGZO nanosheet transistors and bi-layer resistive memory, demonstrating low power consumption and effective CNN performance on standard datasets.

Contribution

It presents a novel 3D-embedded neuromorphic architecture with $1$-IGZO nanosheet transistors and bi-layer RRAM, including device fabrication, modeling, and system-level evaluation for CNN tasks.

Findings

Achieved 92% accuracy on Fashion-MNIST

Achieved 75% accuracy on CIFAR-10

Demonstrated system-level performance with device variation considerations

Abstract

In this paper we propose and evaluate the performance of a 3D-embedded neuromorphic computation block based on indium gallium zinc oxide ($\alpha$-IGZO) based nanosheet transistor and bi-layer resistive memory devices. We have fabricated bi-layer resistive random-access memory (RRAM) devices with Ta$_2$O$_5$ and Al$_2$O$_3$ layers. The device has been characterized and modeled. The compact models of RRAM and $\alpha$-IGZO based embedded nanosheet structures have been used to evaluate the system-level performance of 8 vertically stacked $\alpha$-IGZO based nanosheet layers with RRAM for neuromorphic applications. The model considers the design space with uniform bit line (BL), select line (SL), and word line (WL) resistance. Finally, we have simulated the weighted sum operation with our proposed 8-layer stacked nanosheet-based embedded memory and evaluated the performance for VGG-16 convolutional neural network (CNN) for Fashion-MNIST and CIFAR-10 data recognition, which yielded 92% and 75% accuracy respectively with drop out layers amid device variation.