A Lightweight Post-Processing Method for Voltage-Controlled MTJ based True Random Number Generator

Taking into account the influence of process, voltage, temperature (PVT) and environmental variations on the entropy of random bits generated by the VC-MTJ, this work proposes a lightweight post-processing technique named AXprime, which leverages the avalanche effect of non-cryptographic hash functions to effectively mitigate biases and bit correlations in the raw VC-MTJ bit streams. In comparison to recent work based on Add operations or substitution box, our AXprime approach demonstrates its effectiveness in handling raw data with biases of up to 39% and with strong correlations. Furthermore, compared to extensively commercially-used post-processing methods such as Linear Feedback Shift Register (LFSR) and Advanced Encryption Standard (AES), the implementation of AXprime-based post-processing hardware not only meets all NIST 800-22 test standards, but also improves the Figure of Merit (FoM) value by $1.7 \times$ and $18.8 \times$ respectively, achieving a good trade-off between performance and power consumption.