Improving PUF security with regression-based distiller

Abstract

Silicon physical unclonable functions (PUF) utilize fabrication variation to extract information that will be unique for each chip. However, fabrication variation has a very strong spatial correlation and thus the PUF information will not be statistically random, which causes security threats to silicon PUF. We propose to decouple the unwanted systematic variation from the desired random variation through a regression-based distiller. In our experiments, we show that information generated by existing PUF schemes fail to pass NIST randomness test. However, our proposed method can provide statistically random PUF information and thus bolster the security characteristics of existing PUF schemes.