Cognitive Learning and Neuromorphic Systems Using Resistive Switching Random-Access Memory

Abstract

The exponential growth in data generation and processing demands has exposed the limitations of the traditional von Neumann architecture. The bottleneck caused by the separation of memory and processing units results in significant constraints on computational speed and energy efficiency. Neuromorphic computing, inspired by the structure and function of biological neural networks, has emerged as a promising alternative that enables adaptive and energy-efficient information processing. Among the various technologies advancing neuromorphic systems, Resistive Random Access Memory (RRAM) stands out due to its high density, low power consumption, fast switching speeds, and multilevel data storage capabilities. RRAM operates based on resistive switching (RS), which dynamically switches between the high-resistance state (HRS) and the low-resistance state (LRS) in response to electrical stimuli. This characteristic enables RRAM to effectively mimic synaptic plasticity, a key feature of biological neural networks, including potentiation, depression, and spike-timing dependent plasticity (STDP). Additionally, RRAM-based devices can emulate complex cognitive learning processes such as learning and forgetting, nociceptive behavior, Pavlovian conditioning, and aversion responses. The integration of RRAM with in-memory computing (CIM) architectures eliminates data transfer bottlenecks and further enhances computational efficiency by performing operations such as vector-matrix multiplication within the memory cells. This synergy is particularly advantageous for energy-efficient, miniaturized edge devices and Internet of Things (IoT) applications, enabling real-time learning and decision-making in advanced AI systems. This review provides an in-depth analysis of the role of RRAM technology in neuromorphic computing, discussing resistive switching mechanisms, architectural innovations, and its applicability in cognitive systems. The unique properties of RRAM position it as a core technology for next-generation adaptive computing with the potential to drive innovations in machine learning, AI, and real-time processing systems.