Exploiting Crosstalk Effects in FPGAs for Generating True Random Numbers

This paper presents a new method for implementing TRNGs in FPGA devices, which relies on filling a region or the whole FPGA chip close to its maximal capacity and exploiting the interconnection network as intensely as possible. This way, there are strong chances for the design to exhibit a nondeterministic behavior. Our first design is a computationally intensive core that generates 64-bit numbers, accumulated into a fixed-point accumulator. The bits that exhibit the maximal entropy are then post-processed using the XOR-based bias reduction method. We prove that the resulting TRNG provides high quality random numbers. An explanation of the underlying phenomenon is proposed, based on electromagnetic interferences inside the chip and crosstalk effects. A systematic method for developing TRNG designs based on this approach is proposed and an improved TRNG architecture is then presented.