Enhancement of cross-modal associative memory through memristive neural networks utilizing parallel programmable...

Cross-modal memories play a pivotal role in retrieving and understanding multifaceted information associated with identical entities. However, hardware implementations of parallel-programmable cross-modal associative memory networks have remained elusive. Memristors, nanoscale devices that emulate synaptic connectivity in the brain, offer a compelling platform for circuit-level neural-inspired architectures. In this work, we present a parallel-programmable cross-modal associative memory network circuit that leverages reverse-complementary memristor crossbar arrays. This design enables parallel programming of the weight matrix, dramatically reducing update times and enhancing overall efficiency and scalability. We validate the proposed circuit through comprehensive PSPICE simulations, demonstrating robust recall performance for distorted or noisy grayscale images with high accuracy and resilience to noise. Furthermore, we extend the architecture to realize bidirectional cross-modal memory interactions between audio and image modalities, highlighting its potential for advanced multimedia content analysis and retrieval. Our results address a critical hardware gap in cross-modal associative memories and underscore the promise of memristor-based technologies for scalable, neuromorphic computing.