Physical Unclonable Function with 3D Stacked Memristor Crossbar Array Using Self-Differential Pair

In this work, we demonstrate a 3D stacked memristor crossbar array capable of self-differential pairing, utilizing it as a random entropy source for a physical unclonable function (PUF). A 32 × 32 memristor crossbar array is fabricated, and an identical array is integrated on top of the first layer, forming a 2 × 32 × 32 3D stacked memristor crossbar array. To isolate the two crossbar arrays, a conventional back-end-of-line process is employed, using a SiO₂ passivation layer applied via plasma-enhanced chemical vapor deposition and contact holes. The electrical switching characteristics of all 2048 devices between the low-resistance state and high-resistance state are experimentally verified. Additionally, it is confirmed that the current difference between the two devices at corresponding positions can be sensed through the middle electrode using Kirchhoff's current law, enabling a self-differential pairing mechanism. This self-differential approach is used to extract device-to-device variations from the fabricated 3D crossbar array, serving as a random entropy source for the PUF chip. Based on this method, a challenge-response pair (CRP) extraction process is proposed, and the performance metrics of the PUF chip are evaluated. Thanks to the 3D stacked structure, the system achieves a large CRP space of approximately 6 × 10¹⁹, with reconfigurability demonstrated via cycle-to-cycle variations enabled by reprogramming the fabricated 3D crossbar array. Furthermore, robustness and randomness are validated through machine learning attack simulations and the National Institute of Standards and Technology (NIST) test suites. This approach is expected to enhance cryptographic security by providing a robust entropy source through the self-differential pairing scheme of the 3D stacked memristor crossbar array.