Memristor-Based Architectures for AI at the Edge

Abstract - Memristors, as the fourth fundamental circuit element, have emerged as a transformative technology for enabling efficient in-memory computing and neuromorphic architectures. Unlike conventional CMOS devices, memristors combine storage and processing within a single nanoscale element, allowing massively parallel operations, low leakage, and non-volatile data retention. These properties make them particularly suited for Edge-AI applications, where latency, energy efficiency, and scalability are critical. This paper reviews the role of memristor-based crossbar arrays in implementing synaptic weights for neural networks, highlighting their ability to accelerate matrix–vector multiplications and support adaptive learning mechanisms. Comparative studies against CMOS implementations demonstrate reduced power consumption, higher density, and fault-tolerant performance. Recent research shows applications ranging from healthcare signal analysis to smart sensors and IoT devices, emphasizing memristors as enablers of brain-inspired intelligence at the edge. However, challenges such as variability, limited endurance, and integration with CMOS technology remain significant barriers to commercialization. Future directions include three-dimensional crossbar scaling, hybrid CMOS–memristor systems, and algorithm–hardware co-design to achieve reliable, large-scale deployment. Taken together, memristor-based architectures represent a critical step toward next-generation, low-power, and adaptive edge intelligence. Key Words: memristor, in-memory computing, Edge-AI, crossbar architecture, neuromorphic hardware, energy efficiency.