A Lightweight Neighbor-Averaging Technique for Reducing Systematic Variations in Physically Unclonable Functions

Andres Martinez-Sanchez, Deva Borah, Wenjie Che
New Mexico State University


Physically Unclonable Functions (PUFs) are emerging hardware security primitives that leverage stochastic random process variations during chip manufacturing to generate unique secrets. However, the biased systematic variations that exist in the process variations will cause non-random spatial correlations among PUF elements in the layout, which significantly degrades randomness of PUF generated secrets. Existing methods of reducing systematic variations involve operations with high computational complexity and therefore require high implementation overheads. In this paper, we propose a lightweight method based on averaging neighboring PUF values to derive spatial bias and hence reduce spatial variations. Experimental results using RO PUF data from 192 Spartan 3E FPGAs show that the proposed method achieves comparable or even better randomness improvement compared to existing methods. The proposed method also demonstrates advantages of parameter diversity. The proposed method implemented on Xilinx FPGAs shows up to more than 10x lower implementation overhead compared with the existing method.