Biomimetic SNN simulator in pure Rust (zero-deps, lock-free parallelism) (github.com)

A new Spiking Neural Network (SNN) simulator developed in pure Rust has been released, marking a significant advancement in the field of AI and machine learning. This simulator operates with zero dependencies and leverages multi-threaded lock-free parallelism, making it highly efficient for simulating biophysically accurate neuron models. Designed as a personal project, it implements critical features such as the Leaky Integrate-and-Fire model with adaptive thresholding, spike-timing-dependent plasticity (STDP) for long-term weight changes, and Tsodyks-Markram short-term plasticity for simulating dynamic synaptic responses. The technical implications of this simulator are noteworthy; it maintains network activity stability through a synaptic scaling mechanism and encodes signals using time-to-first-spike (TTFS) latency. Notably, the simulator achieves a simulation pass execution time of approximately 2.63 milliseconds for 100 neurons across four CPU threads. With its innovative design focusing on lock-free concurrency and Data-Oriented Design, this project not only provides a valuable tool for researchers experimenting with SNNs but also enhances our understanding of neuronal behaviors and adaptations in neural networks, aligning with biological principles like Dale's Principle.