A Memristor-Based SNN Hardware Architecture with AHaH Plasticity

The rapid expansion of artificial intelligence has exposed limitations in conventional von Neumann architectures, particularly in energy efficiency and data movement. Neuromorphic computing, especially memristor-based spiking neural networks (SNNs), offers a promising alternative by integrating memory and processing. This paper presents a hardware architecture for a memristor-based SNN that incorporates compute-in-memory (CIM) and Anti-Hebbian and Hebbian (AHaH) plasticity. The design utilizes differential memristor synapses within a scalable modular framework, controlled by an STM32 microcontroller with improved on-off logic to mitigate voltage fluctuations and hardware risks. Experimental validation using the Wisconsin Breast Cancer dataset demonstrates a classification accuracy of 91.26%, highlighting the effectiveness for low-power, event-driven binary classification of our system. The proposed architecture advances the deployment of robust and energy-efficient neuromorphic systems, addressing key challenges in real-time memristor control and scalability.