Printed Electronics-Based Optical Unclonable Label for Artificial Intelligence Authentication

Printed electronics (PE) products have been widely used in diverse applications, ranging from consumer electronics to automotive and healthcare. The mass manufacturability, flexibility, and low cost of lightweight PE products increase their vulnerability to counterfeiting, allowing fake or substandard goods to infiltrate the supply chain. This poses significant risks to both consumers and businesses. In this paper, we propose to print an ultra-low-cost optical unclonable label (OUL) for the authentication of PE or other inexpensive products. The label pattern is easily designed and printed in a negligibly small area. Random fusions among multiple adjacent ink droplets are explored to form unique and unpredictable patterns. The overall fabrication is compatible with the typical PE process. The inherent randomness of production variations in OULs allows for an extremely high encoding capacity of up to $1.59\times 10 ^{3413}$ with a footprint of only $4\times 4$ mm2. We introduce an artificial intelligence (AI) system to authenticate the printed OULs under diverse imaging conditions. The authentication among six categories and 6,000 labels can achieve an ideal accuracy. The authentication accuracy of 300 categories can still achieve an average recognition accuracy of 0.97. Beyond exhibiting robust reliability against bending, thermal fluctuations, humidity, and one-year accelerated aging, the printed OULs are highly resilient to both physical replication and machine-learning-based forgery attempts. To enable practical deployment, the lightweight authentication model is implemented as a smartphone application, allowing users to effortlessly capture OUL images and obtain reliable on-site verification.