ABSTRACT As artificial intelligence transitions from high‐level cognitive function toward complex combinatorial optimization, compact nano‐oscillators have emerged as pivotal computing units applicable to deterministic and probabilistic paradigms. Unlike conventional architectures, these oscillators enable systems to navigate physical energy landscapes to identify global minima without encountering the local minima entrapment of traditional algorithms. This study explores the underlying physical mechanisms and engineering strategies related to binary transition‐metal oxides, NbO x and SiO x , to realize these fundamental oscillators. These materials exhibit abrupt threshold switching (TS) and rapid recovery, facilitating self‐sustained oscillations. Energy‐efficient oscillatory neural networks for deterministic inference can be realized by leveraging mutual interaction in coupled NbO x TS devices. Conversely, stochastic oscillations are triggered by deliberately engineering SiO x TS variability by interfacial control of oxygen vacancies. This stochasticity allows the SiO x TS device to serve as a robust entropy source for probabilistic bits (p‐bits), offering precise sigmoidal probability control. These p‐bits are employed to solve optimization problems with rapid convergence within a few physical cycles and provide hardware‐level randomness for cryptographic security. Overall, this dual‐functional framework demonstrates how tailoring material‐level physics can bridge the gap between fundamental oscillator dynamics and architectural requirements to address the energy and throughput bottlenecks of current Si‐based computing.
Kim et al. (Mon,) studied this question.