Mimicking Nerves with Memristors

Researchers are planning to build a computer mimicking the monumental computational power of the human brain. For this, they prefer to use memristors, because these devices vary their electrical resistance on the basis of the memory of their past activity. Memristors are semiconductor devices, and at NIST, the National Institute of Standards and Technology, researchers demonstrate the long and mysterious manner of the inner workings of memristors, explaining their ability to behave as the short-term memory of human nerve cells.

Nerve cells signal one another, but how well they do so depends on the frequency of their recent past communication. In the same way, the resistance of a memristor also depends on the current flow that went through it very recently. The best part is memristors remember even with their electrical power switched off.

Researchers read the memristor with the help of an electron beam. As the beam impinges on various parts of the memristor, it induces currents depending on the resistance value of that part. Traversing the entire device, this yields a complete image of variations of current throughout the device. By noticing the nature of the current variations, it is possible to indicate the places that may fail, as these show overlapping circles within the titanium dioxide filament.

So far, during their study of memristors, scientists have not been able to understand their working, and neither could they develop standard tool-sets for studying them. Now, for the first time, scientists at NIST have been able to create a tool-set that can probe the working of memristors deeply. They envisage their findings will pave the way for operating memristors more efficiently, and minimize current leaks from them.

For exploring the electrical functioning of memristors, the scientists focused a beam of electronics at various locations on the device. The beam was able to knock some of the electronics from the titanium dioxide surface of the device. The free electrons formed an ultra-sharp image of each of the locations. The beam also caused four clear-cut levels of currents to flow through the device. According to the researchers, several interfaces of materials within the memristor were the cause. Typically, a memristor has an insulating layer separating two conducting metal layers. As the researchers could control the position of the electron beam inducing the currents, they were able to know the location of each of the currents.

By imaging the device, researchers located several dark spots on the memristor. They surmised these spots to be regions of enhanced conductivity. These were the places from where there was a greater probability of currents leaking out of the memristor during its normal operations. However, they found the leaking pathways to be beyond the core of the memristor, and at points where it could switch between high and low resistance levels.

Their finding opened up a possibility of reducing the size of the device to eliminate some of the unwanted current leaking pathways. Until now, the researchers were only able to speculate on the current leakages, but had no means of quantifying the size reduction necessary.