A Spatial Majority Voting Technique to Reduce Error Rate of Physically Unclonable Functions

The Physically Unclonable Function (PUF) is a promising hardware security primitive with a wide range of applications, such as secure key generation, device authentication, IP protection, and hardware entangled cryptography. Due to their physical construction, PUF responses are inherently noisy. Error correction codes can be used to turn noisy PUF responses into keys or static values for these applications. However, a general construction of error correction is expensive and could introduce high entropy loss for PUFs with high error rates. Some PUF pre-processing techniques have been proposed, such as temporal majority voting and dark bit schemes, applied before error correction. In this paper, we introduce a simple and yet effective method to reduce PUF error rate called Spatial Majority Voting (SMV). The idea is to group PUF bits together to produce a single, more stable bit from the group. Experimental data show that SMV works very well, reducing the mean error rate from 6.5% to 0.3% with a group size of 9 on SRAM PUFs implemented in 65 nm CMOS. We also show that SMV can be combined with the dark bits method to further reduce the error rate to less than 0.01%, thus avoiding the need for expensive error correction schemes.