Synopsis: Mimicking the Brain with Superconductors and LEDs

A proposed computer made of superconductors communicating via light could carry out more operations than a human brain while using less energy.  
Synopsis figure
J. M. Shainline et al., Phys. Rev. Applied (2017)

The human brain is a highly efficient processor, using less energy than a light bulb to perform quadrillions of operations. The world’s fastest supercomputer, Sunway TaihuLight in China, performs a few times more calculations every second than a brain but consumes about 800,000 times more energy. Scientists from the National Institute of Standards and Technology, Colorado, have proposed a new computing system—superconductor devices communicating using light—that mimics the brain’s neural architecture. The team’s calculations suggest their system could operate on less energy and perform more calculations than a human brain.

In most computers, each semiconductor component interacts with just a few others. But if, as in the brain, each component needs to be linked to thousands of others, the circuit’s architecture quickly becomes chaotic. Jeffrey Shainline and colleagues suggest that instead of electrons, photons could act as the information carriers, eliminating this problem—photons allow multiple elements to communicate with each other without the need for wired connections.

The team’s artificial neuron consists of a superconducting wire connected to a LED. The two elements act as the signal detector and the signal sender, respectively. In the absence of incoming photons, the LED is off and the neuron is inactive. When the superconductor absorbs photons, its temperature increases, triggering a transition from a superconducting to a metallic phase. This alters the current flow in the LED, switching it on and making the neuron active. Since the transition requires the absorption of multiple photons, the circuit can mimic real neurons, which “fire” only if the incoming signal is above some threshold. Branched waveguides then funnel the emitted photons to thousands of other superconducting neurons. According to the team’s calculations, such a system could perform 10 times more operations than the human brain and require just 20 W of energy.

This research is published in Physical Review Applied.

–Katherine Wright

Katherine Wright is a Contributing Editor for Physics.


Features

More Features »

Subject Areas

OptoelectronicsElectronicsPhotonics

Previous Synopsis

Quantum Information

Traveling with a Quantum Salesman

Read More »

Next Synopsis

Fluid Dynamics

Superfluid Storm at a Surface

Read More »

Related Articles

Viewpoint: Lost Photons Won’t Derail Quantum Sampling
Photonics

Viewpoint: Lost Photons Won’t Derail Quantum Sampling

A photon-based method for demonstrating the advantage of quantum over classical machines can handle photon loss, facilitating experiments. Read More »

Focus: Wave Trick May Lead to Wireless Charging at a Distance
Energy Research

Focus: Wave Trick May Lead to Wireless Charging at a Distance

A technique involving wave interference may improve the practicality of charging a phone wirelessly at some distance from the power source. Read More »

Synopsis: Two-Face Dipole
Optics

Synopsis: Two-Face Dipole

A proposed dipole source of electromagnetic waves can selectively couple its emission into either of two neighboring waveguides. Read More »

More Articles