Fig. 8: ML attack resilience evaluation. | Nature Communications

Fig. 8: ML attack resilience evaluation.

From: Demonstration of high-reconfigurability and low-power strong physical unclonable function empowered by FeFET cycle-to-cycle variation and charge-domain computing

Fig. 8

a XOR PUF structure is introduced to improve resilience against ML attacks. The response bits generated by multiple independent PUF instances are XORed to introduce nonlinearity into the final output. b 8-XOR APUF still shows vulnerability under the investigated attacks with enough CRP samples. c The 8-XOR version of our PUF exhibits almost invulnerable resilience with a maximum accuracy of 52.1% (ideally 50%) even using 108 training samples. d The ML attack resilience of our PUF can be further enhanced by integrating more PUF cells, benefitting from the high scalability of the proposed charge-domain computing. Source data are provided as a Source Data file.

Back to article page