Investigating Brain-Inspired Computing Architectures to Enhance Energy Efficiency in AI Models Using Neuromorphic Computing

Brain-inspired computing is a method of enhancing the efficiency of artificial intelligence through clock speed, akin to the human brain. The traditional Deep Neural Network (DNN) had extremely high accuracy but at the expense of enormous computing resources and energy, hence eliminating them from real-time, edge, and embedded systems. Spiking Neural Networks (SNN), coupled with specialty hardware, fuel their motivation from biological neurons to implement event-based and power-conscious processing for Neuromorphic Computing (NC). However, neuromorphic systems are confronted with training complexity, hardware, and unavailability of large datasets, which hinder their widespread use. The question addressed by this research is whether the two can be combined to achieve energy-efficient AI models that are not subject to accuracy loss. Between traditional deep learning architectures, neuromorphic systems, and hybrid approaches, a power consumption analysis, on the one hand, and computational efficiency and scalability analysis, on the other, are conducted. The work proposes a new hybrid DNN-NC structure, using event-driven neuromorphic inference for the first time in energy-restricted edge applications without sacrificing accuracy. They also demonstrate that by doing so, the power reduction obtained is dramatic without sacrificing competitive accuracy compared to traditional AI models. This research closes the gap between neuromorphic computing and high-performance deep learning toward providing scalable, real-time AI systems for deployment in applications ranging from autonomous systems to edge computing and beyond. The outcome proves the viability of implementing hybrid AI structures toward solving computational efficiencies of modern AI with particular emphasis on sustainability in low-energyconstrained regimes.