Jungyeop Oh1,Sungkyu Kim2,Junhwan Choi1,Sung Gap Im1,Byung Chul Jang3,Sung-Yool Choi1
Korea Advanced Institute of Science and Technology (KAIST)1,Sejong University2,Kyungpook National University3
Jungyeop Oh1,Sungkyu Kim2,Junhwan Choi1,Sung Gap Im1,Byung Chul Jang3,Sung-Yool Choi1
Korea Advanced Institute of Science and Technology (KAIST)1,Sejong University2,Kyungpook National University3
Recent advances in neuromorphic edge devices and Internet of Things (IoT) technologies have brought the unprecedented expansion of interconnected networks and devices, enabling artificial intelligence (AI) based services to the public. This AI service based on smart IoT devices is an attractive target for cybercriminals because it utilizes remotely connected networks to process large amounts of personal and security-related data. However, it is challenging for the software-based security system, which is subject to physical attack, to combat alone the threat of cybercriminals, therefore a built-in hardware-based security system is essential.<br/>Here, we report that hardware-based security primitives implemented using a physically unclonable function (PUF) that exhibits the high entropy achieved via the random switching nature of a poly(1,3,5-trivinyl-1,3,5-trimethyl cyclotrisiloxane) (pV3D3)-based memristor. The excellent insulating property of pV3D3 induces the thermal dominating reset by localizing the Joule heat generated by the reset high current generated to the conductive filament and enhances the stochasticity of the tunneling distance for randomly ruptured Cu filaments. With high bandgap energy (<i>E</i><sub>g</sub>) (8.25 eV) and low-<i>k</i> (2.2) of the pV3D3, the pV3D3-PUFs exhibit near-ideal entropy, uniqueness, and uniformity as well as reconfigurability. Thanks to the outstanding chemical stability of pV3D3, pV3D3-PUFs show reliable operation under mechanical stress and the water immersion condition for IoT applications in outdoor environments. We demonstrate a 1kbit pV3D3 memristor crossbar array enabling the strong pV3D3-PUF via stochastic in-memory computing, inducing robustness to machine learning attacks of the multi-layer perceptron and the generative adversarial networks. Furthermore, the strong pV3D3-PUF passed the NIST randomness test and showed outstanding PUF performances compared to other emerging PUFs. Finally, we have proposed a protocol for highly secured smart IoT devices with pV3D3-PUF, which is expected to advance the era of hyperconnectivity.