Deep-Learning-Based Digitization of Protein-Self-Assembly to Print Biodegradable Physically Unclonable Labels for...

Area of Science:

• Biotechnology
• Materials Science
• Cryptography

Background:

• Counterfeiting poses significant threats to public health and medical device authenticity.
• Physical Unclonable Functions (PUFs) offer a robust solution for secure authentication.
• Existing PUF technologies require further development for broad applicability.

Purpose of the Study:

• To develop a novel biological PUF utilizing protein self-assembly.
• To create a secure and scalable anti-counterfeiting technology.
• To validate the randomness and applicability of the generated cryptographic keys.

Main Methods:

• A facile protein self-assembly process was employed as an entropy source for a biological PUF.
• A deep learning model digitized self-assembly images to extract feature vectors.
• Generated keys were binarized, debiased, and tested for randomness using NIST SP 800-22.
• Images were printed on silk-fibroin biodegradable labels using protein bioinks for physical deployment.

Main Results:

• The protein self-assembly process successfully generated unique fingerprints.
• Deep learning effectively extracted feature vectors, producing sufficiently stochastic cryptographic keys.
• NIST SP 800-22 tests confirmed the high randomness of the generated keys.
• Cellphone imaging of printed labels showed a low error rate compared to source images.

Conclusions:

• The deep-learning-based biological PUF presents a promising low-cost, scalable, and highly randomized anti-counterfeiting strategy.
• Protein self-assembly offers a unique and viable entropy source for biological PUFs.
• Biodegradable, protein bioink-printed labels demonstrate practical deployment potential for authentication.