“As electronic devices shrink in size, they become increasingly subject to naturally occurring thermal noise,” explains team leader Damien Querlioz at the Centre for Nanoscience and Nanotechnology at Paris-Sud University. “This is usually a problem for circuits, but in our work we actually exploit the effect and leverage the dynamics of superparamagnetic tunnel junctions, which are devices that naturally amplify noise, to produce sequences of random bits without a power supply.

“And that is not all: the smaller, hotter and noisier our random number generator becomes, the better and faster it functions. This is an exciting feature and the exact opposite of traditional systems.”

Bistable magnetic tunnel junctions

Superparamagnetic tunnel junctions are similar to the cells used in magnetic random-access memories (MRAMs). They have two states of magnetization representing a bit 0 or bit 1, each of which have different electrical resistances. However, unlike MRAM cells, which have highly stable states, the energy barrier between the two magnetic states in superparamagnetic tunnel junctions is very low. This means that they spontaneously oscillate between their two states at room temperature because of thermal noise. No write operations are therefore required and simply reading out the device state naturally produces random bits.

“The random number generator we propose in fact periodically reads the state of eight superparamagnetic tunnel junctions and combines them with simple XOR gates to produce high-quality random bitstreams,” team member Damir Vodenicarevic tells nanotechweb.org. “These streams are produced at just 20 ft/bit in less than 2 μm2 of circuit space, which is orders of magnitude more efficient in terms of energy and area than current solutions.”

Better than traditional options

Traditional computers rely on pseudo-random number generators that perform mathematical transformations to produce bit sequences. These sequences appear to be random but are, in fact, fully deterministic.

“Such pseudo-random number generators are big in size and energy-hungry, making them ill-suited to emerging low-power computing applications,’ explains Querlioz. “Although researchers have put forward a number of ‘true’ random number generators that aim to extract random bits from fundamentally random physics phenomena, these approaches either require complex circuits to set the device in unstable states by injecting energy into it, or require large and power-hungry circuits to amplify low-amplitude noise into useable binary sequences.”

Superparamagnetic tunnel junctions act as noise amplifiers

“In our work, superparamagnetic tunnel junctions act as noise amplifiers, turning thermal noise into large binary resistance fluctuations, and we only require minimal amounts of energy to read out their states.

The other good thing is that these junctions are compatible with standard CMOS processes, so that they can be integrated in existing chip fabrication techniques.”

It is not all plain sailing though, he says. “The drawback of our approach is speed: although it generates random bits at a much lower energy than other solutions, it does so at a slower pace (limited to around 10 MHz).”

Internet of Things and wearable devices applications?

That said, the new random number generator might be interesting for use in applications like lower-power Internet of Things (IoT) and wearable devices, he adds.

The team, which includes researchers from the University of Texas at Dallas, AIST Tsukuba and Cornell University, says that it is now looking to integrate its random number generator into probabilistic and brain-inspired computing designs.

The research is detailed in Phys. Rev. Applied 8 054045.