研究成果

Physical Unclonable Functions (PUFs) are expected to represent an important solution for secure ID generation and authentication etc. In general, manufactured PUFs are considered to be more secure when the pattern of outputs (the variety of responses) is larger, i.e., the response bit length is longer (e.g., 192-bit response is more secure than 128-bit one). However, the actual bit length is reduced because some response bits are inconsistent (random) for repeated measurements, which are regarded as unnecessary for ID generation and discarded. Latch-based PUFs with N RS latches, for example, generate ideally 2^N responses depending on binary values output from RS latches (0/1). However, some RS latches output random responses which are inconsistent and cannot be used for reliable ID generation, so the variety of responses becomes smaller than 2^N. In this paper, we propose a novel Latch-based PUF structure, which outputs larger variety of responses by utilizing location information of the RS latches outputting the random responses. Differently from random responses themselves, this location information is determined during a manufacturing process, so almost fixed once PUFs are manufactured. The proposed PUF generates 3^N≈ 2^1.58N responses by considering random responses as the third stable value: using ternary values (0/1/random). We estimate the variety of responses generated by the proposed PUFs. According to our experiment with 40 FPGAs, a Latch-based PUF with 128 RS latches can improve it from 2¹¹⁶ to 2^192.7, this being maximized when the 128 latches outputs 0s, 1s, or random outputs with equal probability. We also show the appropriate RS latch structure for satisfying this condition, and validate it using two kinds of different Xilinx FPGAs: Spartan-3E and Spartan-6. The average error rate of responses is only 5.3% when the core voltage is changed within the rated voltage range of the FPGAs. Our proposed PUF using ternary values enhances dramatically the variety of responses while keeping the reliability.