Abstract:Spiking neural networks (SNNs) have shown great potential in achieving high energy efficiency and low power consumption compared to artificial neural networks (ANNs). However, there remains a significant accuracy gap between SNNs and ANNs. To address this issue, we present an in-memory neuromorphic computing (IMNC) chip that supports hybrid spiking/artificial neural networks (S/ANNs) and sparsity-aware data flows. With the IMNC chip, we aim to improve inference accuracy while simultaneously achieving high energy efficiency through optimization at the algorithm, architecture, and circuit levels. First, at the algorithm level, we note that SNNs extract temporal features from input spikes using time-domain convolution operations. Based on this insight, we efficiently utilize leaky integrate (LI) neurons to hybridize SNNs and ANNs, thereby improving accuracy while maintaining highly sparse operations. Second, at the architecture level, we design a sparsity-aware architecture that supports a hybrid S/ANN topology with varying sparsity. Finally, at the circuit level, we propose a ring-based in-memory computing (IMC) macro, whose energy consumption is inversely proportional to the input sparsity, making it ideal for performing energy-efficient multiplication and accumulation (MAC) operations in both SNNs and ANNs. We evaluate the proposed hybrid S/ANNs on various classification tasks and demonstrate their stronger classification and generalization ability compared with pure SNNs. Notably, our IMNC chip, fabricated using 22 nm CMOS technology, achieves impressive measured accuracy rates of over 95% for voice activity detection (VAD) and ECG anomaly detection. Additionally, our IMNC chip demonstrates superior dynamic energy efficiency of 0.43 pJ per synaptic operation, outperforming related works.

Uncertainty Quantification of Signal Integrity Analysis for Neuromorphic Chips

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

Hybrid Signal Integrity Modeling and Analysis of Heterogeneous Integrated System with Neuromorphic Darwin Chip

Signal Integrity Analysis of Neuronal Spike Signal in 3-D Packaging

A Training Scheme for the Uncertain Neuromorphic Computing Chips

A Design Methodology for Fault-Tolerant Neuromorphic Computing Using Bayesian Neural Network

Symmetrical uncertainty based neural selection in motor cortex brain-machine interfaces

Analysis of Neural Spike Signal Transmission Characteristics in 3-D Heterogeneous Integration

NEUTRAMS: Neural Network Transformation and Co-Design under Neuromorphic Hardware Constraints

Stochastic spin-orbit-torque synapse and its application in uncertainty quantification

QuantBayes: Weight Optimization for Memristive Neural Networks via Quantization-Aware Bayesian Inference

Sparsity-Aware In-Memory Neuromorphic Computing Unit With Configurable Topology of Hybrid Spiking and Artificial Neural Network

Uncertainty quantification via a memristor Bayesian deep neural network for risk-sensitive reinforcement learning

Neural State-Space Models: Empirical Evaluation of Uncertainty Quantification

Recent Progress on Signal Integrity Modeling of Neuromorphic Chips by the PEEC Method

Scalable and Efficient Methods for Uncertainty Estimation and Reduction in Deep Learning

Two-dimensional materials-based probabilistic synapses and reconfigurable neurons for measuring inference uncertainty using Bayesian neural networks

Toward stochastic neural computing

Enabling High-Quality Uncertainty Quantification in a PIM Designed for Bayesian Neural Network

Neuroscience-inspired Information-Integration System Based on Stochastic Magnetic Tunnel Junctions

An Electromagnetic Perspective of Artificial Intelligence Neuromorphic Chips