Scientists have successfully developed electronic synapses, teaching the computer to forget things that must be forgotten

Tram Ho

The human brain is the ultimate computer, even if it still exists what we consider to be disadvantages: forgetfulness. Unlike a computer, just load a new English word and it will remember the word for life or until the hard drive fails, your brain will forget the new word you just learned after a few moments. week. The only way to remember it is to keep repeating it in my mind.

But it turns out, what you consider a weakness, is the advantage in a system called an analog neural computer that scientists are dreaming of developing. Imagine what happens if you create a robot with a brain, and then don’t equip it with “forget” information.

As a result, he will have to remember everything, all the car horns on the road, every face he has met in his life, everything he can read on the internet … There will be no endless memory. Allow the robot to do that. Therefore, it is best to ” forget ” the information itself is not important.

The ability to forget information will help an analog computer save energy. memory, while speeding up the calculation and still having access to important information when needed.

Now, a team in Russia has been able to simulate human “forgetfulness” on a device called a second-order memristor, with a memristor paired from memory and resistor .

The clever design allows it to mimic the nature of a synapse in the human brain when it remembers information, and even when that information fades over time if it is not accessed or recalled.

Although at this time, the memristor does not have many practical applications. In the future, however, scientists can use them to develop a new type of neural computer that underpins artificial intelligence systems. This system can meet some of the same functions that the human brain can perform.

Các nhà khoa học phát triển thành công khớp thần kinh điện tử, dạy máy tính tự quên những thứ cần phải quên - Ảnh 1.

Devices that simulate nerve joints, teach computers to forget things to forget

In an analog neural computer, the electronic components integrated on its chip must take on the role of neurons and synapses as in the human brain. If this could be done, the machine could simultaneously increase its computing speed, while reducing the power demand it consumed.

Unfortunately, at the moment, such analog computers are only available on paper.

One major obstacle is that we have not been able to create an electronic device that can mimic the flexibility of the synapses – in which, the more active synapses become stronger, and the synapses Little activity will gradually weaken.

Plasticity of the synapses is also a mechanism that helps us deeply remember some important memories. The brain, on the other hand, can erase unnecessary information itself to release its activity.

In the past, some scientists have developed memristors to perform the role of synapses, using a nanotube bridge design. This ultra-thin conduction bridge then decays over time, in the same way that memory fades into our minds.

“The device that uses this first-class [memristor] solution has a problem, it’s that it tends to change behavior over time and breaks down after prolonged operation, ” said physicist Anastasia Chouprik. from Moscow Institute of Physics and Technology (MIPT), Russia.

To create a new generation of memristors, Chouprik’s team used hafnium oxide iron material instead of nanotubes. This ferromagnetic material can respond to the external electric field acting on it by changing the state of electric polarization. In other words, when electrical impulses hit the material, it can reset its resistance state.

The forgotten mechanism that the memristor imitates is due to the main defect in the junction between silicon and hafnium oxide. These defects are currently making it difficult for chips to be developed from hafnium. But as it turned out, it was the perfect mechanism for a memristor to cause its conductivity to degrade over time, much like the way memory is forgotten.

The main challenge we encountered was finding the right thickness for the electric iron layer , Chouprik said. ” The 4 nm figure proves to be ideal. If that layer of iron is thinner than 1 nm, its iron properties will disappear. Whereas, a thicker layer becomes a barrier too large for allow electrons to pass through . ”

In comparison with the old memristor generation, ” the mechanism we use to simulate synaptic plasticity is now more accurate,” Chouprik said. “In fact, after changing the state of 100 billion times, the system was still functioning properly, so my colleagues stopped testing the endurance [before even reporting equipment failures] “.

Các nhà khoa học phát triển thành công khớp thần kinh điện tử, dạy máy tính tự quên những thứ cần phải quên - Ảnh 2.

A memristor (right) simulates synaptic plasticity (left).

Another thing that makes hafnium oxide an ideal material for making synaptic memristors is that it’s also being used by companies like Intel to make chips.

Scientists calculate that, if in the future they make a true neurological machine, the mass industrialization of memristors will also be made easier on existing lines.

Still, we have to admit that the successful creation of neural computers still has a long way to go. Scientists will need to make memristor memory more reliable. They are also investigating how the memristor can be integrated into other electronic devices for flexibility.

We will look at the interaction between the different resistor conversion mechanisms in our memristor, ” said physicist Vitalii Mikheev from MIPT. ” Turning out electric iron may not be the only effect involved here. To improve the devices even further, we will need to distinguish the mechanisms and learn how to put them together.”

The study was published in the journal ACS Applied Materials & Interfaces.

Refer to Sciencealert

Share the news now

Source : Trí Thức Trẻ