Probability-Distribution-Configurable True Random Number Generators Based on Spin-Orbit Torque Magnetic Tunnel...

Area of Science:

• Materials Science
• Computer Engineering
• Physics

Background:

• Stochastic computing requires high-quality random numbers for applications like simulated annealing and Bayesian networks.
• Hardware-level random number generation is preferred for enhanced speed, efficiency, and parallelism.

Purpose of the Study:

• To demonstrate the suitability of spin-orbit torque magnetic tunnel junctions (SOT-MTJs) as true random number generators.
• To showcase the controllable randomness of SOT-MTJs for generating accurate probability distribution functions (PDFs).

Main Methods:

• Utilizing high-barrier SOT-MTJs as the core component for random number generation.
• Controlling the randomness of SOT-MTJs via applied writing voltage to produce Bernoulli bits.
• Configuring SOT-MTJ-based Bernoulli bits to realize various PDFs, including Gaussian, uniform, and exponential.

Main Results:

• SOT-MTJs exhibit excellent speed and endurance characteristics for random number generation.
• The randomness of SOT-MTJs can be precisely controlled by writing voltage, yielding well-performed Bernoulli bits.
• Demonstrated realization of multiple PDFs, such as Gaussian, uniform, exponential, and Chi-square, using SOT-MTJ-based generators.

Conclusions:

• SOT-MTJs are highly suitable candidates for true random number generators in stochastic computing.
• PDF-configurable random number generators based on SOT-MTJs can significantly advance stochastic computing.
• This technology broadens the application scope for SOT-MTJs in computing and beyond.