Brief

Researchers have developed a new framework to make large language models more compatible with neuromorphic hardware. The method focuses on creating spike-friendly approximations for the nonlinear operators within Transformers, which are typically challenging for standard spiking neuron dynamics. By decomposing these nonlinearities into recurring primitives and using population computation with neuron groups, the framework can approximate common nonlinearities like Softmax and SiLU with minimal accuracy loss. AI

Researchers have introduced FiTS, a novel spiking neuron model designed to enhance the interpretability of Spiking Neural Networks (SNNs). FiTS achieves this by separating temporal computation into Frequency Selectivity (FS) and Temporal Shaping (TS) modules. The FS module identifies a neuron's optimal frequency, while the TS module modulates how frequency components influence membrane voltage accumulation. This approach has demonstrated improved performance on auditory benchmarks compared to standard LIF neurons, offering clearer insights into the network's learned temporal and frequency organizations. AI

Researchers have developed a new neuromorphic circuit architecture that leverages the inherent non-equilibrium dynamics of electrochemical random-access memory (ECRAM) devices to implement short-term plasticity (STP). This co-design framework integrates ECRAM synapses with a delay-feedback leaky integrate-and-fire neuron, enabling transient conductance modulation to directly influence neuron excitability and synaptic efficacy. Simulations show this approach can achieve STP behaviors with low energy consumption and enables tunable temporal filtering for spiking neural networks. AI