US team develops superconducting electronic component that acts like human synapse
Engineers and physicists from the US National Institute of Standards and Technology (NIST) in Colorado have now developed a component that they claim behaves in the same way as synapses, the biological connections and switches between nerve cells in the human brain, and may even be able to outperform real synapses in some respects. Publishing their research in Science Advances, the NIST team, led by Mike Schneider, used superconducting technology in their artificial synapse, which is key to its performance.
Computers that function like biological brains, known as neuromorphic systems, are believed to have potential for running artificial intelligence programmes that have to make decisions based on incoming data, such as those used for medical diagnosis or operating autonomous vehicles, because they would have the ability to process data both in sequence and simultaneously, and store data in a distributed way. This would allow them to mimic human perception more closely and recognise patterns, improving decision-making. Their construction depends on being able to copy the characteristic way that neurons in the brain communicate with each other: through electrical pulses that rise rapidly from no current to a signal current. This is known as a spiking signal, and in the brain is seen in synapses.
Artificial synapses are a missing link in the construction of neuromorphic systems, as chips that act like brain cells have already been developed, using superconducting materials that transmit, process and store information in units of magnetic flux rather than electrical current, as in conventional computers.
Housed within a metallic cylinder about ten microns in diameter, the NIST component comprises a Josephson junction: a sandwich of superconducting materials with an insulator as a filling. These are generally used to calibrate voltage standards because when a current passes through them, the insulator breaks down at a characteristic level to produce a voltage spike. In the artificial synapse, the “bread” in the sandwich is composed of niobium electrodes, while the filling is a silicon barrier containing nanoclusters of manganese. The distribution of these nanoclusters is around 20,000 per square micron, and their behaviour gives the component its properties. “These are customised Josephson junctions,” Schneider said. “We can control the number of nanoclusters pointing in the same direction, which affects the superconducting properties of the junction.”
The clusters basically behave like bar magnets, whose quantum mechanical spin can be distributed randomly or aligned. The degree of alignment dictates the critical current of the junction— the point at which the insulating nature of the silicon barrier breaks down. In their paper, the team explains that this ordering can be controlled by applying a magnetic field to the component: the more nanoclusters are aligned, the lower the critical current. This design, in which different inputs alter the spin alignment and resulting output signals, is similar to how the brain operates, the team explains.
Further tuning of the device is possible by changing the size of the clusters. Smaller clusters need lower magnetic energy to affect the order of alignment, which has the effect of reducing the device’s operating temperature. Raising the operating temperature slightly from -271.15° C to -269.15° C, for example, results in more and higher voltage spikes. These devices fire even faster than the synapses in a human brain: a billion times per second, as opposed to the natural synapse’s rate of 50 times per second, and need only an attojoule of energy to make them spike.
“The NIST synapse has lower energy needs than the human synapse, and we don’t know of any other artificial synapse that uses less energy,” Schneider said.
Moreover, the team adds, the synapses can be ‘stacked’ in a three-dimensional arrangement to form a larger system linking devices acting as neurons, which the term says can be made by conventional electronic component construction methodology.
See how it works in this animated video.