Filamentary Random Telegraph Noise-Based Multiple-Resistive-State True Random Number Generator for Probabilistic Hot/Cold Bits

In this article, we present a true random number generator (TRNG) design based on a 40 nm multi-resistance-state one-transistor-one-resistive-random access-memory (1T1RRAM) macro comprising one million cells, integrating both random number generation and multi-level storage functionalities. By employing a gradual-FORMing scheme, a thinner filamentary conducting path is developed within the RRAM cell, enabling continuous multi-resistance tuning through incremental SET and RESET operations. Consequently, 12 distinct resistance states are established: four corresponding to low-resistance states (LRSs) and eight to high-resistance states (HRSs). While the distributions of the LRS states exhibit significant overlap, the HRS states demonstrate well-defined separations. To leverage these characteristics, we propose a cold-hot data operation scheme, wherein LRSs serve as “hot” data for frequent random number generation, while HRSs function as “cold” data for storage. To ensure high-quality randomness from the LRSs, a TRNG is designed, with entropy primarily derived from thermal noise and random telegraph noise (RTN), as identified through noise distribution analysis. The TRNG exhibits strong performance in key metrics, achieving an Intra-Hamming distance (Intra-HD) of 49.82%, an Inter-Hamming distance (Inter-HD) of 50.26%, and a Hamming weight (HW) of 49.6%. The autocorrelation function (ACF) analysis confirms a 95% confidence. The proposed TRNG successfully passes the National Institute of Standards and Technology (NIST) randomness test at both $25~^{\circ }$ C and $75~^{\circ }$ C, underscoring the potential of RRAM-based TRNGs for secure communication and encryption in AIoT applications.