Neuromorphic Computing Research Areas

Neuromorphic Architectures & Optimization

Focuses on developing scalable, robust, and energy-efficient spiking neural network architectures through diverse methods, spanning from model development (e.g., spiking neural architecture search), learning enhancements, and model optimization.

Paper 1: SpikeNAS: A Fast Memory-Aware Neural Architecture Search Framework for Spiking Neural Network-Based Embedded AI Systems

Paper 2: FSpiNN: An Optimization Framework for Memory-Efficient and Energy-Efficient Spiking Neural Networks

Paper 3: TopSpark: A Timestep Optimization Methodology for Energy-Efficient Spiking Neural Networks on Autonomous Mobile Agents

Neuromorphic Large Models

Neuromorphic Continual Learning & Multi-Task Learning

Fault-Tolerant Neuromorphic Computing

Neuromorphic and Emerging AI Hardware

Neuromorphic Intelligence for Embodied AI Systems

Explores methods to efficiently leverage spiking operations for processing emerging large models (e.g., spiking-based LLMs & ViTs), and systematically optimize them based on their unique characteristics for meeting the given performance (e.g., accuracy), memory, latency, and power/energy constraints.

Paper 1: QSLM: A Performance- and Memory-aware Quantization Framework with Tiered Search Strategy for Spike-driven Language Models

Paper 2: QSViT: A Methodology for Quantizing Spiking Vision Transformers

Paper 3: PrimeSVT: An Automated Memory-aware Pruning Framework with Prioritized Compression Policy for Spiking Vision Transformers

Focuses on enabling continual learning (CL) and multi-task learning (MTL) capabilities for autonomous systems to efficiently adapt to dynamically-changed real-world environments by leveraging bio-inspired spiking neuron, network architecture, learning rule, neural coding, and training mechanism.

Paper 1: Replay4NCL: An Efficient Memory Replay-based Methodology for Neuromorphic Continual Learning in Embedded AI Systems

Paper 2: SpikeDyn: A Framework for Energy-Efficient Spiking Neural Networks with Continual and Unsupervised Learning Capabilities in Dynamic Environments

Paper 3: Scalable Multi-Task Learning through Spiking Neural Networks with Adaptive Task-Switching Policy for Intelligent Autonomous Agents

Develops fault-tolerant neuromorphic computing through hardware- and software-level techniques that mitigate the negative impact of permanent faults and trainsient faults (soft errors) in different parts of spiking neural net work models (e.g., weights) and neuromorphic hardware (e.g., memory and compute engine).

Paper 1: SoftSNN: Low-Cost Fault Tolerance for Spiking Neural Network Accelerators under Soft Errors

Paper 2: ReSpawn: Energy-Efficient Fault-Tolerance for Spiking Neural Networks considering Unreliable Memories

Paper 3: SparkXD: A Framework for Resilient and Energy-Efficient Spiking Neural Network Inference using Approximate DRAM

Develops neuromorphic hardware accelerators that efficiently expedite diverse spiking neural network workloads. This effort explores the use of conventional CMOS, emerging device technologies (e.g., ReRAM, PCM, STT-MRAM), and emerging hardware architectures (e.g., compute-in-memory, 2.5D integration, and photonics).

Paper 1: MorphAtt: A Neuromorphic Accelerator for Efficient Multi-Head Attention Processing in Spiking Vision Transformers

Paper 2: DxPTA: An Architecture Design Space Exploration with Optical Dataflow-guided Strategy for HW/SW Co-Design of Photonic Transformer Accelerators

Focuses on the development of embodied AI systems (e.g., autonomous agents & robotics) that leverage neuromorphic computing, including development of spiking neural network models for processing event-based data stream, efficient sensor & actuation subsystems, and the efficient deployment of spiking neural network models on neuromorphic processors.

Paper 1: SNN4Agents: A Framework for Developing Energy-Efficient Embodied Spiking Neural Networks for Autonomous Agents

Paper 2: A Methodology to Study the Impact of Spiking Neural Network Parameters Considering Event-Based Automotive Data

Paper 3: Enabling Efficient Processing of Spiking Neural Networks with On-Chip Learning on Commodity Neuromorphic Processors for Edge AI Systems

Neuromorphic-Inspired Quantum Machine Learning

Studies the potential of leveraging bio-plausible neuromorphic computing concept (e.g., neural coding & spiking neuron) in enhancing the performance of quantum machine learning (QML).

Paper 1: SPATE: Spiking-Phase Adaptive Temporal Encoding for Quantum Machine Learning

NC Surveys and Roadmaps

Presents comprehensive surveys, roadmaps, and prespective for neuromorphic computing across diverse research directions, including leveraging neuromorphic computing for continual learning and embodied neuromorphic intelligence, such as robotics.

Paper 1: Continual Learning With Neuromorphic Computing: Foundations, Methods, and Emerging Applications

Paper 2: Embodied Neuromorphic Artificial Intelligence for Robotics: Perspectives, Challenges, and Research Development Stack

Team Awards

Best Paper Nomination in the 18th International Conference on Control, Automation, Robotics and Vision (ICARCV) 2024.