Physicists create a device that can 'forget memories'

The brain is the ultimate computing machine, so it's no wonder researchers want to try and simulate it. In the latest research, an interesting step was made to create a device that "forgets" memory, like the human brain.

It is called a `` second-order memory resistor '' designed to mimic a human synapse in the way it remembers information, then loses that information if it is not accessed for a long period of time. .

Although 'memory resistors' do not have many practical applications right away, they could help scientists develop a new type of computer that is the basis of an artificial intelligence system that meets some of the same functions. the brain does.

In so-called analog neural computers , components on the chip (such as memory) can assume the role of each neuron and synapses. That can reduce a computer's power needs and speed up calculations at the same time.

Right now the analog neural computer is hypothetical, because we need to find out how electronics can mimic synaptic flexibility - the way the brain works in strong synchronization over time. and inactive people become weaker. That's why we can cling to some memories while others disappear, scientists say.

Previous efforts to produce 'memory resistors' used nanowire bridges, which then decay over time, in the same way that memories can decay in our minds.

"The problem with this 'memory resistor' solution is that the device tends to change behavior over time and breaks down after prolonged operation. The mechanism we use to implement more powerful synaptic flexibility. In fact, after changing the state of the system 100 billion times, it still works normally, so my colleagues stopped endurance testing , '' said physicist Anastasia Chouprik, from the Institute of Physics and Public Technology. Moscow technology (MIPT) in Russia says.

In this case, the team used a material called hafnium oxide instead of the nanotube, with the electrical polarization changing in response to the external electric field. It means that the low and high resistance state can be set by electrical impulses.

What makes hafnium oxide ideal for this is that it has been used to make chips by companies like Intel. That means it will be easier and cheaper to introduce 'memory resistors' when it's time for a similar neural calculator.

Chouprik said: "The main challenge we encountered was finding the right thickness of the electric iron layer. 4 nanometers were proved to be ideal. Making it just thinner than 1 nanometer, and the ferroelectric properties were no more. , while a thicker layer is too wide a barrier for electrons to pass through. 'Forgetting' is actually done through imperfection which makes hafnium-based processors difficult to grow. Similar defects allow the conductivity of the 'memory resistors' to die over time. It is a promising start, but there is still a long way to go since these memory cells still need to be made more reliable. '

The team also wanted to investigate how their new device could be integrated into the most flexible electronics.

"We will look at the interaction between the different mechanisms that confer resistance in memory. It turns out that the electric iron effect may not be the only relevant factor. To further improve the devices." , we will need to distinguish between mechanisms and learn how to combine them, ' says physicist Vitalii Mikheev, from MIPT.