Morphology-Defined Entropy Streams for Random-Bit Generation and Probabilistic Sampling in Memristor Arrays

Physical entropy sources are essential for hardware security and probabilistic information processing, yet most memristor-based random-number generators rely on stochastic switching outputs without identifying a designable materials-level parameter that controls how intrinsic device stochasticity is expressed. Here, we present a morphology-engineered entropy platform based on Ag/TiO₂/Au nanoisland (NI)/Pt electrochemical memristor arrays, in which the Au NI morphology acts as a tunable structural modulator of the stochastic high-resistance-state transport landscape. Thickness-controlled Volmer-Weber growth produces distinct NI regimes that reshape local-field hotspots and candidate filament pathways through which residual filament configurations, interfacial defects, and stochastic ion migration are translated into HRS-current dispersion. NI-1.5 provides a practical entropy-engineering window where broad hotspot diversity and recoverable HRS states are balanced. The morphology-modulated HRS dispersion is converted into entropy streams in a 32 × 32 array through paired-current comparison and validated using bit probability, binary entropy, min-entropy, restart/reproducibility tests, autocorrelation, NIST SP 800-22 tests, and machine-learning-based predictability evaluation. The validated stream further serves as a stochastic sampling primitive for Monte Carlo probabilistic inference. This work establishes nanoscale morphology as a designable materials-level control parameter linking stochastic memristive transport, array-level randomness, and probabilistic sampling.