Photonic computer chips mimic the way the human brain operates, but at 1000x faster speeds.
As artificial intelligence continues making machines smarter, many in the tech sector believe the “singularity” is right around the corner — the point of technological progress where machines are exponentially smarter than humans.
But when it comes to computing, the human brain remains much more powerful (and more energy-efficient) than any technological processing system on the planet.
In fact, developing microchips capable of imitating the way brain synapses work — particularly their ability to process and store information using almost no energy — has long been the holy grail of computing.
With a new innovation, scientists have moved one step closer to that goal. A team led by researchers from Exeter, Oxford, and Münster universities has developed specialized photonic microchips, which could one day help to create computers that can store and process huge amounts of information at faster speeds than the human brain can.
Because the microchips are powered by light, they could also execute high-speed computing at a lower power supply than could ever be possible with any electronic processing system.
The human brain’s powerful hardware
In the brain, synapses connect nerve cells (or neurons) to one another.
The brain signals that form memories and thoughts move through these nerve cells as tiny electrical charges; when a charge reaches a synapse, it triggers the release of neurotransmitters, the chemical messengers that communicate information throughout the brain.
Synapses are essentially the “circuits” that facilitate our thinking, feeling, and activation of movement. And there are plenty of them: with around 100 billion neurons existing in a healthy brain — and each one connected to thousands of others — estimates put the number of synapses in the brain between 100 trillion and 1,000 trillion.
More amazing than their number, though, is their speed: the many trillions of “synaptic connections” in the brain operate at a speed akin to that of a computer with a 1 trillion bit per second processor. (For reference, that’s about 10,000x faster than the 100 megabits-per-second Ethernet at your office.)
Creating a more human-Like microchip
Mindful of the brain’s innate processing power, researchers across the field of “neuromorphic computing” are developing brain-
To date, even the world’s fastest processors and CPUs have been unable to come close to the processing speed of brain synapses. In one simulation, it took 40 minutes of work by over 80,000 processors in K computer — one of the fastest supercomputers on the planet — to achieve just 1 second of biological brain processing time.
Making computers faster is possible, but also demands more energy. That gives the human brain essentially two main components where it has the upper hand over machine. Our immense network of neurons and synapses can:
1) rapidly process and store vast amounts of information simultaneously (this is known as “parallel processing”), and
2) carry out parallel processing using very little energy totaling just a few tens of watts of power.
“Since synapses outnumber neurons in the brain by around 10,000 to 1, any brain-like computer needs to be able to replicate some form of synaptic mimic,” said Professor Wolfram Pernice, a co-author of the paper from the University of Münster. “That is what we have done here.”
With their light-powered microchip, the team of European researchers tackled both issues — creating a “hardware synapse” capable of incredible speed, that operates with incredibly little power.
The team created the chip by combining “phase-change” materials commonly found in household items, such as rewriteable CDs and DVDs, with photonic integrated circuits.
Phase-change materials are a class of substances capable of storing and releasing large amounts of energy in response to heat stimuli. Photonic integrated circuits, meanwhile, use light — rather than electrons — to manipulate atoms and perform other functions.
Using light in place of more expensive, inefficient energy sources (such as electricity) is a goal of many scientists. In fact, an entire discipline known as “integrated photonics” is focused on developing fast-processing microchips based on light signals for use in faster, greener electronics.
The photonic microchips developed by the Exeter, Oxford, and Münster researchers are a leading example. According to the team, their photonic synapses can operate at speeds a thousand times faster than those of the human brain — representing a crucial step toward unlocking the “holy grail” of brain-mimicking computers.
While the research team has so far only conducted tests to prove the microchip’s programmability and effectiveness in “synaptic mimic,” they say their brain-inspired innovation displays the essential requirements of a solution for a neuromorphic computing tool. And that is what potentially makes the photonic microchip, in the words of the researchers, such a “pioneering breakthrough.”
It comes back to AI: if a high-speed, light-based neurochip is developed into our electronics in the not-so-distant future, those devices will be able to process artificial intelligence algorithms at a faster, lower-energy clip than ever. If devices can support AI and conduct advanced machine learning faster, the human brain begins to lose some of its supremacy over the machine, perhaps putting us another step closer to the singularity.
The original research “On-chip photonic synapse” was published in the journal Science Advances on September 27, 2017. Full information is available here.