Researchers at MIT have created artificial neural network synapses that are 1 million times faster than the biological ones found in the human brain.
Just as digital processors need transistors, analog processors need programmable resistors. Once properly configured, these resistors can be used to create a neural network of artificial synapses, according to a press release.
These synapses with artificial neurons are not just extremely fast, they are also extremely efficient. And that's very important, because as digital neural networks become more advanced and powerful, they require more and more energy, significantly increasing their carbon footprint.
See also: Robot player breaks 7-year-old's finger in chess tournament
As detailed in a new paper, the researchers hope their findings will advance the field of “analog deep learning,” a growing field of artificial intelligence (AI).
By abandoning the commonly used organic media and opting for a high-tech glass, known as inorganic phosphosilicate glass (PSG), the researchers were able to reach nanosecond speeds, which were faster than the speeds of biological synapses of neural networks in the human brain.
"The ability of biological cells to act increases and decreases on a millisecond time scale as the voltage difference of about 0.1 volts is limited by the stability of water ," said senior author and professor of nuclear science Ju Li , in the statement. "Here we apply up to ten volts to a special nanoscale-thick solid glass membrane that transmits protons, but without being permanently damaged."
"And the stronger the field, the faster the ion devices are," he added.
Because PSG can withstand high voltages without breaking, it allows protons to travel at incredible speeds between artificial neurons, while also being very energy efficient.
See also: Scientists turn dead spiders into robotic parts
The material is common and easy to manufacture, making it not only the fastest option, but also the most practical.
“Once you get an analog processor, you’re no longer training networks that everyone else is working on,” lead author Murat Onen in the statement. “You’re going to train networks with unprecedented complexity that no one else has the ability to do, and therefore you’re going to outperform them by a wide margin.”
"In other words, this isn't just like a faster car, this is like a spaceship," he added.
Source: futurism.com
