Correlated Vanadium Oxides as Physical Encoders for AI-Enabled Intelligent Sensing

Conventional sensors are typically designed to generate simple, nearly linear responses that are subsequently analyzed using computational algorithms. An emerging alternative is intelligent sensing, in which the sensing element itself participates in information processing by producing rich multidimensional responses that are decoded using machine learning [1]. Rather than using AI simply to analyze experimental data after the measurement, the physical device itself becomes part of the information-processing architecture. In this work, we propose that correlated vanadium oxide thin films, such as V₃O₅ and V₄O₇, could provide a fundamentally different physical platform for intelligent sensing without requiring complex gated structures. The strong electron correlations, phase competition, and history-dependent transport of correlated vanadium oxides provide a rich physical state space that may significantly expand the information content available to AI compared with conventional sensing materials [2-4]. Instead of producing a single scalar output (for example, a resistance), the device could be intentionally operated to generate multidimensional electrical responses—for example, —or other response landscapes acquired under controlled optical excitation, allowing the correlated material itself to function as a physical encoder that transforms the incident optical state into a multidimensional electrical signature. Optical sensing provides a natural first demonstration because correlated vanadium oxides are known to exhibit pronounced photoinduced changes in their electronic state [5–6]. Machine-learning algorithms would be trained using a library of known response landscapes to identify and classify previously unseen optical excitation conditions. While optical sensing serves as the initial demonstration, the same framework could ultimately be generalized to multimodal sensing by combining optical excitation with electrical or thermal control. In this framework, the complexity of the correlated material becomes an advantage rather than a limitation: the material performs the nonlinear physical encoding, while AI performs the decoding. As the database matures, AI could also guide the measurement process itself by identifying the most informative regions of the response landscape to interrogate, thereby reducing acquisition time and enabling adaptive sensing strategies. This work aims to establish the foundation for a new research direction at the intersection of correlated quantum materials, intelligent sensing, and AI-enabled physical information processing. An initial proof-of-concept will establish whether correlated vanadium oxide films can serve as AI-enabled intelligent sensors and determine how their intrinsic nonlinear dynamics can be harnessed to realize adaptive, information-rich sensing architectures beyond conventional device paradigms. 1. Wang Y et al. Prog. Quantum Electron. (2025) 100-101, 100563. 2. Rúa A et al. Appl. Phys. Lett. (2026) 129, 021906. 3. Camino FE et al. J. Appl. Phys. (2026) 140, 015105. 4. Rúa A et al. J. Appl. Phys. (2026) 139, 185102. 5. Bartenev A et al. Adv. Electron. Mater. (2025) 11, 2400539 6. Bartenev A et al. J. Appl. Phys. (2024) 136, 125109.