Scientists Create a Device that Mimics Human Synapses to 'Forget' Memories

New research has taken an intriguing step in that direction to emulate the brain. The scientists had invented a tool that's capable of 'forget' memories, just like our minds do.

The invention is called a second-order memristor - a mix of "memory" and "resistor." The smart design imitates a human brain synapse in a way it remembers information, and gradually drops that data if it's not accessed for a more extended period.

The memristor, while does not have much practical use - for now - could eventually help scientists develop a new kind of artificial intelligence system that fulfills some of the same functions a brain does.

On-chip electronic components like the memristor could take on the role of single neurons and synapses, which could lessen the computer's energy requirements and quickly computations at the same time.

Analog neurocomputers are hypothetical for now as scientists need to work out how electronics can mimic synaptic plasticity - the way that active brain synapses strengthen in time and idle ones get weaker.

Efforts to produce memristors used nanosized conductive bridges that would then decay over time were previously made in the same way that memories might decay in our minds.

Physicist Anastasia Chouprik from the Moscow Institute of Physics and Technology (MIPT) in Russia said the first-order memristor tends to change its behavior over time and breaks down after prolonged operation.

"The [tool] we used to [complete] synaptic plasticity is [sturdier]. [The memistor] was still operating normally after switching the state of the system 100 billion times. So my colleagues ended the endurance test," Chouprik said.

Chouprik and her team used a highly insoluble thermally stable material placed in a bridging device with an electric polarisation which changes in response to an external electric field. It means electrical pulses can set low and high resistance states.

The materials are already being utilized to build microchips by companies such as Intel. In other words, introducing memristors is now easier and cheaper if and when the time comes for an analog neurocomputer.

Chouprik said the main challenge she and her team encountered was figuring out the correct layer thickness. Four nanometers proved to be ideal, she said.

The researcher said the ferroelectric property is gone if they made it one-nanometer thinner while a thicker film is too broad for the electrons to pass through. 

The actual 'forgetfulness' is made via a flaw that makes the microprocessors challenging to develop, Chouprik said. She added the errors would allow memristor conductivity to die down over time.

The innovation is a promising start; however, there is a long way still to go. The memory cells still need to be made more reliable, for example. The researchers would also investigate how their new device could be incorporated into flexible electronics.

Physicist Vitalii Mikheev from MIPT said they are going to investigate the interplay between different mechanisms switching the resistance in their memristor.

"[The] ferroelectric effect [might] not be the only one involved. We will need to distinguish between the mechanisms and learn to combine them to improve the devices further," Mikheev said.

The research has been published in ACS Applied Materials & Interfaces.