Spiking Neural Networks from Signals to Silicon:An End-to-End Survey of Dynamics, Learning,Architectures, Toolchains, Neuromorphic Hardware,and Deployment

Spiking neural networks (SNNs) are often intro-duced as neural networks with binary activations and temporalstate. That description omits the engineering contracts that deter-mine whether spikes remain useful after training and deployment.This survey follows an SNN from physical signal to actionableoutput. It connects event sensing and spike encoding to neuronand synapse dynamics; conversion, surrogate-gradient, exact-gradient, and local online learning; convolutional, recurrent,residual, attention, graph, and generative architectures; eventdatasets and evaluation protocols; simulation frameworks, inter-mediate representations, compilers, quantization, placement, mul-ticast routing, calibration, and runtime integration; and digital,mixed-signal, FPGA, compute-in-memory, and sensor-computehardware. We organize these layers through an event contractwith seven fields: event semantics, time, state, precision, locality,adaptation, and evidence boundary. This contract exposes why anominally spiking model can lose its energy advantage throughdense encoding, high firing rates, long time windows, unsupportedoperators, host preprocessing, or routing traffic. The surveyprovides mathematical correspondences among continuous anddiscrete neuron models, derives training and conversion errorpaths, and separates operation-count proxies from accelerator,board, host-inclusive, and sensor-to-decision measurements. Ta-bles compare coding schemes, learning rules, datasets, softwarestacks, deployment transformations, and major neuromorphicsystems. The resulting account treats SNN deployment as a cross-layer optimization problem rather than a choice between isolatednetwork architectures or chips.