A CMOS rectified linear unit operating in weak inversion for memristive neuromorphic circuits

Article Computer Science, Hardware & Architecture

A Native SPICE Implementation of Memristor Models for Simulation of Neuromorphic Analog Signal Processing Circuits

Bo Li et al.

Summary: This article introduces the application of memristors in analog/mixed-signal circuit design and presents a native SPICE implementation of memristor models. The effectiveness of this implementation in terms of simulation accuracy, efficiency, and convergence for large-scale simulation tasks is validated through case studies and functional simulations.

ACM TRANSACTIONS ON DESIGN AUTOMATION OF ELECTRONIC SYSTEMS (2022)

添加到收藏夹

Article Computer Science, Hardware & Architecture

AccuReD: High Accuracy Training of CNNs on ReRAM/GPU Heterogeneous 3-D Architecture

Biresh Kumar Joardar et al.

Summary: The paper introduces an M3D-enabled heterogeneous architecture, AccuReD, that combines ReRAM arrays with GPU cores to address challenges in CNN training and achieve high accuracy. AccuReD supports all types of CNN layers and achieves near-GPU accuracy even with the low-precision and nonideal behavior of ReRAMs, accelerating CNN training while maintaining accuracy.

IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS (2021)

添加到收藏夹

Article Computer Science, Hardware & Architecture

Diagonal Matrix Regression Layer: Training Neural Networks on Resistive Crossbars With Interconnect Resistance Effect

Yan Liao et al.

Summary: Resistive crossbars implement parallel vector-matrix multiplication in analog fashion, but can suffer from output distortions due to interconnect resistance and sneak path issues. This article proposes an accurate and computationally efficient model called diagonal matrix regression (DMR), which is integrated into neural networks as DMRL. By using this technique, accuracy in MNIST and fashion-MNIST classification tasks is significantly improved.

IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS (2021)

添加到收藏夹

Article Computer Science, Hardware & Architecture

Memristor-Based Edge Computing of ShuffleNetV2 for Image Classification

Huanhuan Ran et al.

Summary: In this article, a memristor-based ShuffleNetV2 for image classification is proposed, which is suitable for edge computing due to low power consumption and high integration. By optimizing the design of memristor crossbars and using inverters, the required memristors and power consumption in MCNN have been significantly reduced, achieving effectiveness in image classification.

IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS (2021)

添加到收藏夹

Review Computer Science, Artificial Intelligence

A survey on modern trainable activation functions

Andrea Apicella et al.

Summary: In recent years, there has been a renewed interest in trainable activation functions, which can be trained during the learning process to improve neural network performance. Various models of trainable activation functions have been proposed in the literature, many of which are equivalent to adding neuron layers with fixed activation functions and simple local rules.

NEURAL NETWORKS (2021)

添加到收藏夹

Article Computer Science, Hardware & Architecture

RxNN: A Framework for Evaluating Deep Neural Networks on Resistive Crossbars

Shubham Jain et al.

Summary: Resistive crossbars are a promising component for DNN hardware, but suffer from various nonidealities that can lead to accuracy degradation. It is crucial to study the impact of these nonidealities on large-scale DNNs, although existing models are too slow for application-level evaluations. RxNN is introduced as a fast and accurate simulation framework to evaluate large-scale DNNs on resistive crossbar systems, showing significant accuracy degradations (9.6%-32%) and enabling model-in-the-loop retraining to mitigate the degradation.

IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS (2021)

添加到收藏夹

Article Multidisciplinary Sciences