<p>With the rapid advancement of edge intelligence, ensuring the security of edge devices and protecting their communication has become critical. Physical unclonable functions, known as <i>hardware fingerprints</i>, are an emerging hardware security solution enabled with the physical variations inherent in the hardware systems. To facilitate a widespread edge deployment, here we present chip-scale reconfigurable physical unclonable functions built with carbon nanotube charge-trapping transistors, where the charge-trapping memory and physical variations of the transistors are harnessed to render over 10¹³ reconfigurable states and the demonstrated ideal physical unclonability. Arising from this, the physical unclonable functions prove robust resilience against advanced machine learning and artificial intelligence attacking (limiting success to ~50–60%) as well as brute force cracking (requesting an estimated 10¹⁶ years to crack). This performance, along with their scalability and low-power operation as well as cryogenic temperature robustness, position the physical unclonable functions a promising hardware security solution for edge intelligence. As a practical demonstration, we model self-driving vehicular network in Central Hong Kong and prove secure vehicle communication using the physical unclonable functions.</p>