The New Superconducting Synapse Is 20 Million Times Faster Than Human Synapses

  • Scientists have created an artificial synapse that works like a human brain. 
  • It’s a 10 micrometers (diameter) squat metallic cylinder that can process incoming electrical spikes to configure outgoing signal. 
  • Strengthening the connection between input and output enables the artificial synapse to ‘learn’, just like a neural synapse. 

Researchers at the NIST (National Institute of Standards and Technology) have developed a superconducting switch that operates like a human brain. It could connect different processors and store information in future artificial biological systems.

The switch is named synapse, a biological counterpart, and it promises to supply the crucial part of neuromorphic computers. Such computers, powered by artificial intelligence, could drastically improve perception levels and decision-making features for applications like cancer diagnosis and self-driving vehicles.

Like other deep learning technologies, it is designed to learn through experience, or just from the surrounding. But how different it is from all other existing systems and future candidates? Let’s find out.

Artificial Synapses Vs Human Synapses

When it comes to context recognition, there is nothing like human brain, which processes the information in both sequential and parallel way, and stores this information (memories) in synapses all over the system.

Getty Images

Synapse is a small gap at the end of a neuron in the central nervous system. It allows a signal to pass from one neuron to the next. The artificial synapse built by NIST can process incoming electrical spikes to configure outgoing signal. It’s a 10 micrometers (in diameter) squat metallic cylinder.

The processing of artificial synapse is based on a flexible internal design, which can be tweaked either by surrounding or by experience (or previous inputs). The more firing between processors or cells, the more cohesive/stronger the connection.

Both artificial and real synapses can maintain old circuits and make new ones. However, the artificial synapse is much faster than the real one – it can fire 1 billion times per second compared to brain cell’s 50 times per seconds, making it 20 million times faster.

To transmit signals, artificial synapses use a whiff of energy, around one ten-thousandth as much as a real synapses. More specifically, the spiking energy is lower than one attojoule (1018 joules) – lesser than the background energy at room temperature. This means, the new artificial synapse requires far less energy compared to human synapse.

These artificial synapses would be utilized in neuromorphic computers, mostly build of superconducting materials that can carry electric charge without any resistance. Hence, they would be more efficient compared to other ordinary designs based on software and semiconductors. Information would be stored, sent and processed in magnetic flux units.

Read: The Mind-Reading AI Can Now Visualize Your Thoughts

How Does It Work?

The new artificial synapse is a Josephson junction – voltage standards used by NSIT. These junctions are nothing but an insulator sandwiched between superconducting materials. Voltage spikes occur when the current passing through the junction exceeds a threshold value called critical current.

The synapse consists of niobium electrodes and nano-scale manganese clusters (approximately 20,000 per square micrometer) in a silicon matrix. These nanoclusters behave like small bar magnets whose spins can be controlled. Multiple nanoclusters can be made to point in the same direction (magnetic ordering), affecting junction’s superconducting properties.

Disordered and ordered nanoclusters

In order to increase the magnetic ordering, researchers applied electrical pulses in a magnetic field. This effect decreases the level of critical current and makes it quite easier to produce standard conductor and voltage spikes. When entire nanoclusters are aligned, the critical current is at lowest level.

Reference: Sciencemag | doi: 10.1126/sciadv.1701329 | NIST

The process can also be reversed by applying pluses (with no magnetic field) to decrease the magnetic ordering and increase the crucial current.

By altering the design and its operating temperature, one could configure the behavior of the synapse. The pulse energy can be decreased by making the nanoclusters smaller. For instance, raising the operating temperature by 2 Kelvin (from 2K to 4K) results in higher voltage spikes.

The overall design, where different inputs change alignment of spin and final output signals, is very similar to how the human brain works.

In order to utilize this technology in real world computing, synapses can be stacked in 3D to make bigger systems. So far, researchers have developed a circuit model for simulating how the system would operate.

Read: Yet Another Artificial Synapse To Put Brain On A Chip

They have demonstrated synaptic weight training with electrical pulses as low as 3 attojoules. The Josephson plasma frequencies of the instruments, which regulate dynamical time scales, all exceed 100 gigahertz. According to the research team, these new artificial synapses provide a crucial step towards a faster, energy-efficient and far more complex neuromorphic platform that has been demonstrated with other techniques.

Written by
Varun Kumar

Varun Kumar is an experienced science and technology journalist interested in machines, AI, and space exploration. He received a Master's degree in computer science from Indraprastha University. To find out what his latest project is, feel free to directly email him at [email protected] 

View all articles
Leave a reply