Physical Authentication of Electronic Devices using Synthetically Generated 3D Material Signatures

Microelectronics packaging industry possesses complex supply chains. Longer supply chains raise credible security concerns since they increase the probability of malicious entities being inserted/embedded onto a product. This greatly highlights the need for anti-tamper technology that can nondestructively authenticate microelectronic products at any point in a supply chain, ideally in real time. When compared to existing techniques such as electronic RFIDs and hologram stickers, particle filled adhesive films show great promise due to their non-reproducible characteristics. These films when embedded onto a package during the manufacturing process can act as indicators of physical tampering, since the original distribution is lost in damaged or replaced films. In this study, we develop computational tools that synthetically generate 3D packings of particle filled adhesive films and systematically evaluate the material signatures generated from the particles. Plane projections of synthetically generated particle films are considered as material “signatures“ and used to evaluate and classify one pattern from another. To compare particle distributions, we develop Convolutional Neural Networks (CNNs), two-point spatial (statistical) correlation functions, and area as well as density distribution plots. While the developed techniques show tremendous potential for classifying material signatures, high speed accuracy and adaptability of CNN makes it a very good choice for this purpose. Statistical techniques on the other hand, provide important local and global insight into signature. However, their accuracy is degraded as rotational and translational errors are introduced into the signatures.