Designing Neuromorphic Computing Hardware Using Memristive Devices for Energy-Efficient AI at the Edge in Nigerian Technology Applications

Background: Neuromorphic computing architectures using memristive synaptic devices offer orders-of-magnitude improvements in energy efficiency over conventional von Neumann AI processors, with transformative potential for edge AI applications in Nigerian precision agriculture, environmental monitoring, and healthcare. Aim: This study designed neuromorphic computing hardware based on memristive devices and evaluated its performance for energy-efficient edge AI inference tasks relevant to Nigerian technology applications. Methods: HfO2-based resistive switching memristive devices were fabricated and characterised for synaptic weight emulation, endurance, retention, and multi-level conductance states. A 256x256 memristive crossbar array was designed and its matrix-vector multiplication accuracy modelled using Monte Carlo analysis of device variability. SPICE circuit simulations of a spiking neural network implemented on the memristive hardware were evaluated for energy consumption per inference on image classification and acoustic keyword spotting tasks representative of Nigerian field deployment scenarios. Results: Memristive devices achieved 128 distinct conductance states with 8-bit equivalent precision and endurance exceeding 107 switching cycles. The neuromorphic hardware achieved energy consumption of 28 picojoules per multiply-accumulate operation, 420 times lower than GPU baseline. Spiking neural network classification accuracy was 93.1% on agricultural pest image classification. Conclusion: Memristive neuromorphic hardware provides compelling energy efficiency advantages for edge AI in Nigerian field applications. A roadmap for local neuromorphic chip development is proposed for the Nigerian technology sector. Keywords: neuromorphic computing, memristive devices, edge AI, energy efficiency, Nigeria.