Feature extraction and encoding of electrospun nanofibers for unclonable anti-counterfeiting and authentication

Abstract

Random nanopatterns have emerged as potential physically unclonable functions (PUFs) due to their high-dimensional features and high unclonability at the nanoscale. Herein, a random pattern of electrospun nanofibers (ESNFs) is proposed as a novel PUF candidate in the merit of facility, low cost, and scalability. Both auto- and cross-correlation algorithms evaluate the uniqueness of the nanopatterns. The ESNF-PUF device has good mechanical strength, and the fiber pattern is very robust to bending, stretching, and torsion. The tape and vibration test have also verified the stability of the device against external deformation. Then the features of fiber amount of the patterns are quantitatively extracted and used to build an identifier of 15-bit with a reconstruction success rate over 80%. Finally, a deep learning strategy based on convolutional neural network is developed to achieve fast authentication of anti-counterfeiting tags under various imaging conditions.