Hardware Authentication Enhancement of Resistive Random Access Memory Physical Unclonable Functions

Abstract

Advancements in microprocessors and sensor technologies have led to many innovations in the Internet of Things (IoT). These developments have both improved the quality of life for individuals and led to a need for securing users' information. This is especially true in devices such as pacemakers, cars, and credit cards, which can provide information that can harm users. To protect users from hackers who want this information, Physical Unclonable Functions (PUFs) can be used. Memory-based PUF are especially useful, as they can be readily implemented on most systems without much effort or additional hardware. This device is also unique in that it is very difficult to clone and hackers will have a hard time reading the contents of the device. Resistive Random Access Memory (ReRAM) PUFs in particular provide a similar manufacturing process to current Flash technologies, making them easily integrated into current technologies. On top of being similar to manufacture, ReRAM devices are also lower power than flash, allowing them to be used in low power devices such as Radio Frequency Identification Tags. While this is an advantage, ReRAM devices are currently limited in use since they vary greatly in different operating conditions. In this paper, a statistical model is proposed to account for shifts that occur at different temperatures. To generate the model, a mean square error linear regression analysis was performed, and found that these devices can be loosely represented as mean shifted Gaussian distributions at different temperatures. This model allows for a better understanding of how the system will perform during the challenge response pair authentication process. It was also found that the error rate can be reduced to near zero using this method, but may need improvement due to the limitations of this data-set. These limitations can be seen with the bit error rate, however these were improved using multi-state soft decoding. This process compared a ternary and eight state grouping, which allows for a better understanding of how each cell affects the array. Along with the statistical model the system will have minimal burden on the servers during the challenge response process, as it is computationally simple. Future works will include an implementation of this system to further allow ReRAM to become a more powerful technology, and help innovate the IoT.