Random fractal-enabled physical unclonable functions with dynamic AI authentication

Counterfeiting causes significant economic losses and safety risks across various sectors, including electronics and pharmaceuticals. Current anti-counterfeiting labels often fall short due to deterministic fabrication, making them susceptible to duplication. Physical Unclonable Functions (PUFs) offer a promising solution, providing unique, fingerprint-like features generated through stochastic processes. Existing PUFs include graphical PUFs based on randomly distributed micro/nanostructures (e.g., wrinkles, buckling, evaporative patterns), spectral PUFs utilizing analytical tools for readout (e.g., Raman scattering, luminescence), and complex electronic PUFs leveraging material imperfections. While graphically encoded tags are convenient and robust for direct optical identification, material limitations often compromise their environmental stability. This research aims to overcome this limitation by developing a new type of graphical PUF with superior stability and flexibility. Random fractal structures, ubiquitous in nature, exhibit self-similarity and inherent randomness, making them ideal candidates for PUFs. The study utilizes the fractal-guided depercolation of gold films during thermal annealing to create random Au networks. Gold's physicochemical properties ensure the label's durability. To enhance security, the authors explore integrating chemical encoding using Raman spectroscopy with the plasmonic properties of the Au network. Finally, a dynamic deep learning system is designed for efficient authentication, addressing the challenge of managing large PUF databases.

The paper reviews various existing PUF technologies, categorizing them into graphical, spectral, and electronic PUFs. It highlights the advantages and limitations of each category, particularly focusing on the trade-off between the flexibility and complexity of graphical PUFs and their environmental stability. Existing graphical PUFs, while offering flexible controllability in code complexity, often suffer from low physical robustness due to the use of flexible materials. The review emphasizes the need for PUF carriers that combine high environmental stability with code flexibility. The authors also discuss the integration of chemical encoding with PUFs to achieve multi-level security and the limitations of current authentication systems, pointing out the challenges of using conventional image processing algorithms and the need for more efficient and reliable AI-based authentication methods.

The authors develop a fabrication process for random fractal-enabled Au network PUFs. First, a laser direct writing (LDW) technique and magnetron sputtering create Au films of defined thicknesses on a Si/SiO2 substrate. Thermal annealing induces a fractal-guided depercolation process, forming random Au networks. The process involves the formation of voids due to thermal expansion mismatch, followed by the stochastic retraction of void edges to create irregular mesh-like structures. The complexity of the network, characterized by its fractal dimension, can be adjusted by controlling the film thickness and annealing parameters. The authors use a fractal branching model to describe the network structure. The encoding capacity is determined through image analysis. To enhance security, a surface roughening technique is employed to generate plasmonic hotspots, enabling surface-enhanced Raman scattering (SERS). Rhodamine 6G is used as a Raman probe molecule to demonstrate chemical encoding. For authentication, the authors develop a dynamic deep learning-based system. A ResNet50-based convolutional neural network (CNN) is trained on a large dataset of PUF images. A dynamic key database strategy is implemented to improve efficiency and scalability. The system utilizes a smartphone with a mini-microscope for convenient readout, with the images subsequently processed by the deep learning model. The Feature Similarity Index (FSIM) is used to compare images and ensure accuracy. The performance of the PUF tags, including uniqueness, encoding capacity, and environmental stability, is rigorously evaluated. Various tests demonstrate the PUFs’ robustness under extreme temperatures, aqueous conditions, and mechanical abrasion. The deep learning model's performance is evaluated using various metrics, including validation accuracy and false positive/negative rates.

The study successfully demonstrates the fabrication of random fractal Au network PUFs with high encoding capacity (10^34). The complexity of the network is shown to be highly controllable by adjusting the film thickness, leading to configurable encoding capacities. The cross-correlation analysis reveals a high degree of uniqueness among the fabricated PUFs. The surface roughening process significantly enhances the SERS signal, enabling multi-level security through chemical encoding. The dynamic deep learning authentication system achieves 0% false positives and a low false negative rate (5%). The authentication process is fast (6.36s) and convenient using a smartphone-based readout. The fabricated PUF tags exhibit excellent environmental stability, enduring extreme temperatures and harsh conditions. This enhanced stability is further improved by encapsulating the PUFs with protective coatings, such as PMMA or SiO2, tailored to the specific application environment. The proposed fabrication and authentication methods show high compatibility with mass production, which is particularly relevant for real-world commercial applications. The results show that the fractal dimension increases as the film thickness decreases. They also demonstrate that the encoding capacity increases exponentially with the size of the PUF pattern. The integration of SERS enhances the security, creating a multi-dimensional PUF tag. The study also highlights the effectiveness of the dynamic database strategy in reducing the training time of the deep learning model.

The findings address the research question by presenting a novel approach to create highly secure and stable PUFs for anti-counterfeiting applications. The combination of fractal-guided film annealing, plasmonic enhancement, and dynamic AI authentication addresses the limitations of existing PUF technologies. The high encoding capacity, coupled with zero false positives in authentication, ensures exceptional security. The configurable complexity and demonstrated environmental stability make these PUFs suitable for a wide range of applications. The dynamic database strategy significantly improves the efficiency of the authentication system, making it scalable for large-scale deployment. The results suggest that fractal-guided self-organization processes can be a powerful tool for designing PUFs in various material systems. The versatility of this approach opens opportunities for integrating functional elements, such as flexible packaging or invisible displays. Future research can focus on exploring different materials and fabrication methods to further expand the applicability of this technology.

This work successfully demonstrates a novel anti-counterfeiting system based on random fractal Au network PUFs and a dynamic deep learning authentication system. The system exhibits high security, scalability, and environmental stability. Future work could explore integrating additional security layers (e.g., more complex chemical encoding) and adapting the system to different applications.

The current study primarily focuses on the visual and Raman spectral properties of the PUF tags. Exploring other characterization techniques and their integration into the authentication system could further improve security. The dynamic deep learning model's performance relies on the initial dataset quality. Further refinement of the data preprocessing and model training could enhance accuracy and robustness. The long-term stability of the chemical encoding over extended periods needs further investigation.