Hybrid CMOS LIF Neuron and Memristor Array System for Visual Pattern Recognition

This paper presents a hybrid neuromorphic system that integrates a CMOS Leaky Integrate-and-Fire (LIF) neuron with a memristor array, designed for visual pattern recognition. Unlike traditional spike-timing-dependent plasticity (STDP) methods, the system employs a Python-based external program to pre-train neural network weights, which are applied to the memristor array through controlled voltage pulses. Simulations on the MNIST dataset demonstrate a classification accuracy of $\mathbf{9 2. 5 \%}$. With a power consumption of just $\mathbf{5. 6} \mu \mathrm{W}$, the system is ideal for energy-constrained applications requiring moderate accuracy. These results highlight its potential for low-power, real-time neuromorphic applications, offering flexibility for consistent recognition in predefined pattern detection tasks.