18 June 2022 in Nature Communications
Real-world sensory-processing applications require compact, low-latency, and low-power computing systems. Enabled by their in-memory event-driven computing abilities, hybrid memristive-Complementary Metal-Oxide Semiconductor neuromorphic architectures provide an ideal hardware substrate for such tasks. To demonstrate the full potential of such systems, we propose and experimentally demonstrate an end-to-end sensory processing solution for a real-world object localization application. Drawing inspiration from the barn owl’s neuroanatomy, we developed a bio-inspired, event-driven object localization system that couples state-of-the-art piezoelectric micromachined ultrasound transducer sensors to a neuromorphic resistive memories-based computational map. We present measurement results from the fabricated system comprising resistive memories-based coincidence detectors, delay line circuits, and a full-custom ultrasound sensor. We use these experimental results to calibrate our system-level simulations. These simulations are then used to estimate the angular resolution and energy efficiency of the object localization model. The results reveal the potential of our approach, evaluated in orders of magnitude greater energy efficiency than a microcontroller performing the same task.
1 June 2026 in Neuromorphic Computing and Engineering
Taking inspiration from the brain to design efficient computational systems remains challenging due to its complexity. This work investigates a key biological feature observed across mammals: a hierarchy of time scales in cortical areas. Experimental evidence shows that intrinsic neural dynamics slow down across the cortical hierarchy, forming a temporal hierarchy. We examine whether this property benefits artificial systems by introducing temporal hierarchy into spiking neural networks (SNNs), which inherently process information over time. We implement hierarchical time scales across neuronal, synaptic, and recurrent dynamics, and evaluate their effect under two settings: (1) as an inductive bias, and (2) as an emergent property through optimization. On temporal benchmarks such as multi-timescale-XOR and keyword spotting, hierarchical SNNs consistently outperform non-hierarchical ones, achieving 2%–6% higher accuracy and up to 5 parameter reduction under iso-accuracy conditions. Moreover, when trained freely, temporal hierarchy emerges spontaneously through gradient descent. Finally, our theoretical analysis shows that hierarchical time constants enable processing of multi-frequency temporal signals with only log N layers—compared to N layers for non-hierarchical systems—highlighting hierarchy as a key organizational principle for efficient temporal computation.
25 June 2026 in Neuromorphic Computing and Engineering
Neuromorphic computing promises low-latency and energy-efficient control, yet featuring neuromorphic principles in the training of control policies with Reinforcement Learning (RL) remains costly. We present a training framework that efficiently brings neuromorphic principles into control RL in MuJoCo environments. Our approach centers on a Policy Distillation (PD) pipeline: a high-performing teacher is trained with Soft Actor Critic (SAC), then distilled into constrained student policies via supervised imitation and a short RL finetuning stage, yielding a per-student speedup of more than 7.5x for Artificial Neural Network (ANN) policies and 8.1x for Spiking Neural Networks (SNNs) compared with end-to-end RL training. For SNNs, the runtime advantage remains favorable even when accounting for teacher training, reducing total training time by 2.6x. Using this pipeline, we systematically study three pillars of neuromorphic control policies: sparsity, stateful dynamics, and spike-based computation. We show that distilled policies tolerate substantial architectural compression, low-precision weights, and high levels of weight and activation zero-suppression with limited performance loss. To address reduced sensing and event-inspired inputs, we evaluate recurrent (stateful) policies under masked observations of the environment and differential/delta encodings, where memory enables robust performance in partially observable settings. Finally, we train stateful SNN policies with surrogate gradients and introduce DiffPOP, a differential population coding interface that improves spike-based encoding/decoding efficiency and performance over prior population-coding baselines. Together, these results establish an efficient pipeline for training neuromorphic control policies under deployment-relevant constraints, lowering the barrier for practitioners to integrate sparsity, statefulness, and spike-based computation into RL workflows targeting resource-constrained neuromorphic hardware.