Additively Manufactured RF Electronics With Structurally Integrated Physically Unclonable Functions for Wireless System Security

Physically unclonable functions (PUFs) are alternatives to secret keys stored in non-volatile memory. PUFs derive secret keys on demand based on readings invoked from the complexity of their structures or their physical properties. Conventional PUFs rely on the variations and tolerances involved in their manufacturing processes. On the other hand, traditional manufacturing processes aim to achieve cost-effective device replication by minimizing tolerances often through the use of masks for patterning which in turn limits PUF security by reducing their response diversity across multiple devices. Additionally, the fact that manufacturing tolerances are openly known leads to higher quality adversarial attacks towards these conventional PUFs. In contrast to traditional volume manufacturing, emerging additive manufacturing (AM) techniques are mainly mask-free and therefore cost-effective for creating unique randomizations within devices to act as PUFs. With AM, randomizations can be deliberately introduced by the designers in a way to enhance security and/or complicate the PUF operation mechanisms. This manuscript, for the first time, utilizes laser-enhanced direct print additive manufacturing (LE-DPAM) technology to demonstrate that AM based novel PUF security primitives can be structurally integrated with RF electronics such as antennas. Specifically, we investigate the use of compact I/O expander IC packages and demonstrate a PUF that utilizes 4 such IC packages structurally integrated directly under the ground plane of a 5.8 GHz ISM band patch antenna. Each IC package exhibits 48 input/output (I/O) pins leading to a 192-bit PUF security primitive where the I/O pins are randomly connected to high or low digital states. The NIST statistical test results confirm that the generated PUF outputs exhibit a high degree of statistical randomness, as demonstrated with high confidence p-values across all NIST tests. Furthermore, we evaluate the security of the proposed PUF against widely used machine learning (ML) modeling attacks and demonstrate that the PUF is robust with none of the attacks achieving prediction success that exceeds random guessing.