Abstract: Neuromorphic computing is a computer-system design approach inspired by the structure and information-processing principles of biological neural systems. Unlike conventional von Neumann architectures, neuromorphic systems integrate computation and memory more closely and exploit event-driven communication, sparse activity, parallelism, and, in some implementations, on-chip learning. This paper presents a structured, source-audited review of neuromorphic computing, focusing on architecture, spiking neural networks, learning mechanisms, hardware implementation, scalability, energy efficiency, edge intelligence, and engineering challenges. Rather than combining heterogeneous measurements through unsupported statistical meta-analysis, the review distinguishes documented hardware specifications, workload-specific experimental measurements, and reported performance claims. Representative platforms, including IBM TrueNorth, Intel Loihi, Loihi 2, and Hala Point, are examined in terms of neuron and synaptic capacity, architectural organization, learning capabilities, and deployment context. Evidence indicates that neuromorphic systems can provide energy and latency advantages for selected sparse, event-driven workloads, although performance depends strongly on workload characteristics, algorithms, implementation, and measurement methods. TrueNorth demonstrated large-scale digital neuromorphic integration, while Loihi introduced programmable on-chip learning and Loihi 2 expanded neuron-model flexibility and capacity. Hala Point demonstrates further multi-chip scaling potential, but its specifications should not be interpreted as biological equivalence. Recent initiatives such as NeuroBench improve evaluation consistency by distinguishing algorithm-level and deployed-system measurements. Key challenges include training large spiking networks, hardware variability, software interoperability, benchmarking, memory and communication constraints, and integration with established AI workflows. The review concludes that neuromorphic computing is a complementary paradigm with strong potential for sparse temporal workloads, low-latency applications, and power-constrained environments.

Keywords: Brain-Inspired Computing Neuromorphic Architecture Spiking Neural Networks Synaptic Plasticity

Pages: 141-151

WSEAS Transactions on