コロイドナノワイヤのランダム性に基づく人工知能攻撃耐性物理的複製不可能性関数

Countless tiny needles scatter across a surface, each landing in a unique, unpredictable pattern. This principle is extended to the microscale by utilizing colloidal nanowires, which are deposited into randomly structured networks via spin-coating, forming a physical unclonable function (PUF) resistant to artificial intelligence (AI)-based modeling attacks. The intrinsic randomness of the nanowire arrangement generates distinct, irreproducible cryptographic fingerprints for secure authentication. To enhance security, a triple-key authentication scheme is implemented, integrating morphological feature extraction, end/cross-point detection, and co-occurrence analysis. This approach achieves an optimal balance of uniformity and uniqueness, with both metrics approximating 50% and exhibiting minimal variance. This multilayered authentication approach introduces significant nonlinearity, ensuring strong resistance to predictive modeling. Even with advanced AI-based modeling trained over 1000 iterations, the maximum prediction accuracy remains below 62%, demonstrating robust resistance against AI-driven attacks. To evaluate the environmental robustness, stress tests under temperature fluctuation, humidity variation, and thermal exposure were performed, with results confirming that the encapsulated nanowires effectively preserves the PUF device's structural stability across diverse environmental conditions. Compared with existing PUF technologies, our approach provides greater scalability, streamlined fabrication, and strengthened security against malicious AI attacks. By leveraging the randomness of colloidal nanowires and utilizing standard imaging techniques, this approach provides a highly secure, cost-effective solution for authentication security applications, paving the way for next-generation hardware security systems in the era of AI and the Internet of Things.