Research Projects

Conventional computing architectures are increasingly limited by energy inefficiency and data movement bottlenecks. In-memory computing offers a promising alternative by integrating processing directly within memory units. Memristor crossbar arrays, due to their nonvolatile nature, high density, and analog computing capabilities, are key enablers of this paradigm and are particularly suitable for artificial intelligence and edge computing tasks. Neuromorphic computing, modeled after biological neural systems, also benefits from memristor technology. These arrays can perform vector-matrix operations efficiently while storing weights locally. However, as memristor-based architectures move toward real-world deployment, concerns about their security and reliability become critical. Attacks such as WriteHammer, NeuroHammer, and adversarial perturbations pose significant risks, especially in safety-critical applications. This project focuses on building a security-aware and fault-tolerant framework for memristor-based in-memory and neuromorphic systems. The research aims to understand device-level vulnerabilities, design robust architectures, and develop system-level defenses that account for hardware variability and adversarial behavior. The central hypothesis is that integrating knowledge across device modeling, system architecture, and machine learning can significantly improve resilience against both intentional attacks and natural hardware faults. Simulations using compact memristor models, fault injection tools, and behavioral frameworks will support this exploration. Experiments will include modeling physical attack mechanisms, developing runtime anomaly detection, evaluating adversarial robustness in neural networks, and testing fault-tolerant techniques like error correction and reconfiguration. The outcomes are expected to advance understanding of attack-resilient hardware behavior, improve reliability in neuromorphic inference, and contribute to secure system design for edge applications. The project will also generate validated models and open-source tools for the broader research community.