Recent Advances in Efficient Spiking Neural Networks: Architectures, Learning Rules, and Hardware Realizations

Spiking Neural Networks (SNNs) have emerged as a promising paradigm for energy-efficient and event-driven computing, drawing inspiration from biological neurons. Recent research has introduced novel methods to enhance the performance, robustness, and hardware compatibility of SNNs. This review synthesizes key advances across five major areas: pulse width modulation (PWM)-based spike generation to eliminate timing errors, spike-timing-dependent plasticity (STDP) acceleration via early termination and spike count strategies, event-driven spike detection for neuromorphic implantable brain–machine interfaces (iBMIs), one-spike phase coding with base manipulation to minimize ANN-to-SNN conversion loss, and memristor-based radial-basis spiking neuron circuits for adversarial attack resilience. By analyzing over 45 recent IEEE studies, we highlight trade-offs between accuracy, latency, and power consumption while benchmarking hardware implementations. Furthermore, we discuss open challenges, including the need for improved conversion techniques, adaptive coding schemes, and scalable hardware platforms. This review aims to provide a comprehensive foundation for researchers and engineers seeking to advance SNN technologies for next-generation neuromorphic systems.