Random numbers, strictly defined in Statistics, describe that sequential numbers
contain irregular and no recognizable patterns. Random numbers as intrinsic
concepts of applied sciences also frequently appear in the game theory,
cryptography, chaos theory, quantum physics, metrology, biology, etc. Especially, a
random-number seed to trigger mutation events of genome is a cornerstone of the
molecular evolution in biology. Despite significance as it is, in most cases,
random numbers are generated artificially by mathematical algorithms through
software coding on computer. [1] We call this kind of random numbers the pseudo
random numbers. (PRNs) For instance, nowadays, computing methodology of molecular
evolution is to run evolution algorithms (phylogenetic tree [2]) on personal or
super computers by using the PRNs as seed. However, along with unpredictable
evolution of various viruses, civilization and economics of human-beings face more
lurking diseases. To prevent from invasion of unknown bio-like structures in the
future, it is an urgent necessity for researches to efficiently and accurately
predict paths of molecular evolutions through a high-performance, low-power, and
costefficient application-specific chips based on stable and pure true random
number generators. (TRNGs)
The TRNGs should be developed by noises of nature. For the very-large-scale-
integration (VLSI), the TRNGs are required to generate random bit-strings fast with
energy-and-cost efficiency. Hence, the TRNGs composed of complementary- metal-
oxide-semiconductor ring-oscillators (CMOS-ROs) [3] have been developed, which
takes advantage of signal-propagation fluctuation due to variation of resistance-
and-capacitance (RC) delay. But the multiple-stage CMOS-RO increases cost and
lengthens latency. The latch-type CMOS circuits with compact designs, such as the
static-random-access-memory (SRAM) [4], gradually become mainstream of the TRNGs.
The latched-point of the SRAM can be used as a source of entropy since it is easily
affected by the operating voltage, environmental temperature, and device-mismatch.
Furthermore, passive-element circuits have been applied to produce random numbers
according to the chaos theory [5] The CMOS-based TRNGs improve energy-and-cost
efficiency as the Moores law progresses well, but volatility of random numbers
becomes the major drawback.
Consequently, non-volatile memories (NVMs) have been deeply studied to be
possible non-volatile TRNGs. The resistive-random-access-memory (RRAM) uses
stochastic switches of electrically conductive filaments in the metal-insulator-
metal (MIM) of the RRAM to provide entropy. [6] Spintorque-transfer-magnetic-
random-access-memory (STT-MRAM) uses a border distribution of magneticresistance at
the high-resistance-state (HRS) to offer entropy. [7] Recently, the resistive-gate
NVM (RGNVM) has been developed. [8] In an RG-NVM cell, the bottom electrode of the
RRAM MIM is connected to the gate of a MOS-field-effect-transistor (MOSFET) such
that the resistance of the MIM can modulate the threshold-voltage (Vth) of the
MOSFET. Moreover, through stochastic dynamic properties of filaments in the RRAM
MIM, the RG-NVM has been reported as a qualified TRNG. [9]