Bit selection algorithm suitable for high-volume production of SRAM-PUF

Physically Unclonable Functions (PUFs) are impacted by environmental variations and aging which can reduce their acceptance in identification and authentication applications. Prior approaches to improve PUF reliability include bit analysis across environmental conditions, better design, and post-processing error correction, but these are of high cost in terms of test time and design overheads, making them unsuitable for high volume production. In this paper, we aim to address this issue for SRAM PUFs with novel bit analysis and bit selection algorithms. Our analysis of real SRAM PUFs reveals (i) critical conditions on which to select stable SRAM cells for PUF at low-cost (ii) unexplored spatial correlation between stable bits, i.e., cells that are the most stable tend to be surrounded by stable cells determined during enrollment. We develop a bit selection procedure around these observations that produces very stable bits for the PUF generated ID/key. Experimental data from real SRAM PUFs show that our approaches can effectively reduce number of errors in PUF IDs/keys with fewer enrollment steps.