Processor based Intrinsic PUF Design for Approximate Computing: Faith or Reality?

Approximate computing has recently emerged as an advanced computing paradigm for energy efficient and lightweight system design. However, several potential vulnerabilities and threats show a negative impact on the security and integrity of approximate computing systems. To address this, a processor datapath based intrinsic PUF design methodology is demonstrated for approximate computing. This work investigates and verifies the effectiveness of processor based PUF design for approximate computing. First, an 8-bit ripple carry adder is analyzed by applying a popular approximate computing technique namely voltage over-scaling. Due to the voltage over-scaling and process variations, the ripple carry adder shows stochastic timing errors. Exploiting this phenomenon, CMOS based datapath is designed and operated in approximate computing region to generate PUF response. It is observed that with the reduction in supply voltage, harvested PUF bits show increased uniqueness. At a 45nm technology node and 0.5V supply voltage, the extracted PUF bits exhibit a maximum uniqueness of 40.31%. Moreover, the reliability of PUF reduces with scaling supply voltage, exhibiting a reliability of 85.67 (at 0.5V) with a temperature range of -30 degrees C to 80 degrees C. Interestingly, the uniqueness of PUF bits increases with the reduction in power consumption of datapath for approximate computing. Thus, designing intrinsic processor based PUFs shows large benefits and scope due to the increased vulnerabilities and lower resource utilization of approximate computing.