Logo image
Back
CIMulator: Evaluating Neuromorphic Accelerators with HZO FefinFET-Based Synaptic Device
Conference paper

CIMulator: Evaluating Neuromorphic Accelerators with HZO FefinFET-Based Synaptic Device

Hoang-Hiep Le, Hao-Yu Lu, Md. Aftab Baig, Sourav De, Ing-Chao Lin and Darsen Lu
DOI: https://doi.org/10.1109/GTSD62346.2024.10675103

Abstract

Accuracy computing in memory HZO FefinFET MLP MNIST Multilayer perceptrons Neural networks neuromorphic accelerator Neuromorphic engineering Performance evaluation Training Energy Consumption
In this study, we use the CIMulator platform to evaluate the performance of neuromorphic accelerators with novel Hf 0.5 Zr 0.5 O 2 (HZO) ferroelectric fin field-effect transistor (FefinFET) as synaptic device. The MNIST handwritten digit dataset is used for training and inference processes in multilayer perceptron (MLP) neural networks. With the highly optimized synaptic device, the results demonstrate high training and inference accuracy, achieving up to 97% for the MLP, approaching the accuracy of software implementations.

Metrics

1 Record Views

Details

Logo image