Researchers at the University of Minnesota say that they have taken a big step in a strange but growing field of computer science. Called stochastic computation, the method uses random bits to compute via simpler circuits, at lower power, and with greater error tolerance. Although it was first conceived in the 1960s, one of the factors that led to stochastic calculus was the lack of appropriate devices for make it practical.
At the  2017 IEEE International Electron Devices Meeting last month in San Francisco, Professor of Electrical Engineering Jian Ping Wang and PhD student Yang Lv reported that They had built such a device. Their device, similar to a memory cell MRAM can perform stochastic computer versions of addition and multiplication on four logic inputs.
An MRAM cell is essentially a two-terminal device at the nanoscale called Magnetic Tunnel Junction . Like many devices, it basically presents a sandwich structure: ferromagnetic layer, non-ferromagnetic layer, ferromagnetic layer. The orientation of the magnetizations in the upper and lower layers is the key. If they have the same orientation, the current can flow from one to the other with little resistance. If their magnetizations point in opposite directions, the resistance becomes enormous. MRAM can store data because one of the magnetic layers can have its field returned using some type of current. However, when writing data to a cell, it is possible that the cell does not switch to the desired orientation. This is a problem when these devices are used as memory.
Instead of removing this randomness from the nature of the junction of the magnetic tunnel, Wang and Lv have used it. They designed a cell that produces random strings of bits that carry and compute information.
The previous incarnations of the stochastic calculation would use such a device as a random number generator whose bits would then be introduced into a set of stochastic logic circuits. In such a system, the values would be represented as probabilities – 4 would be represented by a seemingly random string of 100 bits, of which about 40 would be 1. It is easy to see that stochastic computing would be relatively insensitive to simple errors where a few or two are reversed, as it would make little difference. But it also makes some calculations less complex. The multiplication of two numbers, for example, can simply be done with a single AND gate.
“Until now, most stochastic computing propositions involving new spintronic devices use spintronic devices as random generators but still use conventional logic gates as part of computation,” Yang explains. “We continued to implement random generation and compute functions in a single MTJ cell.”
They demonstrated that the randomness of the tunnel junction could be regulated by four independent values: the amplitude and width of a current pulse feeding the junction, a bias current flowing through the junction and a magnetic field of polarization. Stochastic data can easily be converted to any of them with simple circuits. When Yang and Lv did this, they found that the device added all the values entered as pulse amplitude, bias current, and bias magnetic field. The result of this triple summation was then multiplied by the value represented by the pulse width to produce a response in the stochastic calculation form – a string of random bits with a particular probability of occurring.
They have built and tested stochastic magnetic tunnel junction calculators in their lab, but to really see what they can do, they are partnering with the GlobalFoundries chip maker. The company has a commercial process for manufacturing MRAM embedded in microprocessors and other chips, and Wang hopes to find a convenient way for this memory to also make calculations. “What we have done here, in principle, GlobalFoundries could manufacture in the near future,” he says.
Minnesota ‘s demonstration of a single magnetic tunnel junction stochastic computing unit is part of a wave of recent research aimed at using the inherent random activity of devices at the same time. nanoscale to calculate new ways. Engineers from Purdue University and the University of California at Berkeley have proposed using a thermally unstable tunnel junction combined with a transistor to form what is called a p-bit , which has the ability to allow logic circuits to operate and back . Osaka scientists used tunable randomness in connections between carbon nanotubes to improve detection and believe that it could also help computer science. And researchers at Hewlett-Packard Laboratories have demonstrated that a particular type of memristor demonstrates a kind of controlled chaotic behavior . When they simulated a network of such so-called analog computation engines, chaos helped speed up the solution of a problem traveler .