Exploiting Read Noise of Filamentary VCM ReRAM for Robust TRNG

Ensuring data security increasingly demands robust true random number generation. At the same time, in the efforts to overcome the von-Neumann bottleneck, a variety of resistive random access memory (RRAM) devices, such as redox-based RRAM (ReRAM), are integrated particularly into memory units. In this paper, the usability of read noise measured on valence change mechanism (VCM) devices for the generation of random numbers is investigated. Especially, the influence of the conduction mechanism on the quality of the generated random numbers is specifically examined. Hence, SrTiO<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sub>- and TaO<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</sub>-based devices are studied as representatives for devices with shallow and deep defect states. Devices with deep oxygen defect states demonstrate superior suitability for the generation of true random numbers, whereas devices with shallow oxygen defect states show a significantly reduced entropy capacity to be used as entropy sources. For TaO<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</sub>-based read noise, true random numbers can be generated in a single-step process. By using a 4-bit Fibonacci Nonlinear Feedback Shift Register (NLFSR), the generation rate is significantly enhanced while introducing additional cryptographic keys. The quality of the respective outputs is checked with the benchmark test provided by the National Institute of Standards and Technology (NIST).