Imprimir Republish


The secret of the memoristor

Researchers try to demonstrate the mechanism behind memoristors’ ability to process and store data

Illustration of neural network made up of memoristors: the nanocomponents behave similarly to synapses

IBM Illustration of neural network made up of memoristors: the nanocomponents behave similarly to synapsesIBM

In 1971, a professor of electrical engineering and computer science at the University of California, Berkeley, published an article in which he proposed the existence of a new, basic electronic component. Leon Chua argued that there could be a resistor with memory, called a memoristor, with unique properties, if developed on a nanometric scale. Then a theoretical and mathematical concept, this element would be able to oscillate, almost instantaneously, from the behavior of an insulator to that of a semiconductor and “remember” its last level of electrical resistance when it stopped receiving current. Only in 2008, 37 years later, did a team at HP Labs, in the United States, produce the first circuit based on the elusive component. The researchers created a titanium oxide nanofilm with memoristors measuring 15 nanometers. Based on this work, the memoristor began to be promoted as a possible wildcard from nascent nanoelectronics. It could potentially carry out the two most basic tasks of a computer more quickly, with lower energy consumption and in a smaller space — namely process information (like a chip with silicon transistors) and store information (like PC hard drives and the flash memory of pen drives).

Even now, no one knows exactly why memoristors act in such a unique manner, although some companies—like giant Panasonic and tiny Knowm, in New Mexico—are already beginning to sell a few modest versions of chips based on this component. The movement of some oxygen atoms inside nanofilms made of metallic oxides, when submitted to different electric currents, is the most accepted hypothesis for the unique properties of the memoristor. In July 2016, in an article published in the journal Scientific Reports, a team of theoretical physicists from the Federal University of the ABC (UFABC), São Paulo State University (Unesp), and the National University of Yokohama (Japan) proposed an alternative explanation for the phenomena: the circulation of elections could be responsible for the characteristics of this component, given that the movement of atoms is not believed to be fast enough to produce the effects attributed to memoristors.

These components can change their resistance when an electric current is passed through them for a few picoseconds (one trillionth of a second is a picosecond). “We are not saying that this effect is just an electronic phenomenon,” explains Gustavo Dalpian, a physicist at UFABC and coordinator of the team that produced the theoretical study as part of a FAPESP thematic project. “But we believe that just the oscillation of the atoms inside the material is not sufficient to explain the attributes of memoristors.” According to the article, in certain internal configurations of their atoms, such as in what ore known as oxygen-deficient phases of titanium oxide, the memoristors are able to store charge. “This alters their electronic properties and, thus, their ability to conduct electricity,” says physicist Antonio Claudio Padilha, another co-author of the study who did his doctorate on the topic at UFABC and is currently undertaking postdoctoral research at the University of York, England.

Memoristor_247The theory’s new attempt to clarify the nature of how the memoristor works needs to be supported by experimental data. Some researchers who have been working longer in the field are skeptical about seeking an explanation by focusing on electrons rather than atoms. This is the case of physicist Gilberto Medeiros-Ribeiro, of the Federal University of Minas Gerais (UFMG). In April 2016, three months before the publication of the article by Dalpian and his colleagues, Ribeiro and a team of researchers from HP reinforced the traditional hypothesis with respect to the mechanism behind this type of component with new evidence.

In an article published in Nature Communications, the scientists reported the measurement of internal noise originating from the movement of ions (atoms that have lost or gained electrons) in a system with memoristors made of tantalum oxide. “The noise level was 10,000 times higher at the points of contact between the atoms and the circuit’s electrodes,” says Ribeiro, who worked at HP Labs for four and a half years as the manager of memoristor research before being hired by the university in Minas Gerais. “On the scale of our devices, just one oxygen atom ‘moving’ one atomic position in the memoristor reduces its resistance by a factor of 10.” In the study, Ribeiro and his colleagues at the U.S. company developed memoristors in which the internal channel, the space in which the ions can move, was the thickness of one atom. Similar to how cosmic radiation is evidence of the existence of the Big Bang, this excessive internal noise in the memoristors, which occurs only when atomic contact takes place, is believed to be evidence of the movement of the ions inside the material.

Despite the memoristors not needing special conditions in order to operate, prior studies by Ribeiro and other researchers indicate that enormous variations in temperature at specific points in these components can occur. “The circuit as a whole is at room temperature, but the contact points between the metallic oxides and the electrodes can reach 800 degrees Celsius,” explains Ribeiro. This accumulation of heat in certain regions also explains the rapid movement of atoms inside these components, according the UFMG physicist.

Prototype of component using HP’s new technology...

HP Labs Prototype of component using HP’s new technology…HP Labs

The structure of a memoristor is extremely simple. It is a nanofilm composed of strands of a metallic oxide with widths between 20 and 50 nanometers, connected to two metal electrodes, which are the points or poles of contact. If it were not for its tiny scale, essential to its characteristics, the memoristor could be mistaken for a conventional resistor, one of the three fundamental passive components (that do not generate energy) in electronic circuits, together with capacitors and inductors. “It is a relatively easy component to manufacture, although there are still many questions yet to be answered with respect to how it works,” comments Dalpian. Nanometric sheets of memoristors can be piled into hives.

In functional terms, a memoristor, whose processing and storage properties are usually compared to those of neurons, can do much more than a resistor can. The latter has constant electrical resistance. Its ability to oppose the passage of electric current in a circuit is constant, independent of the voltage across it. In other words, its electrical conductivity, large or small depending on the material used to manufacture it, never changes. This is why the resistor is a fundamental component that limits and stabilizes the current in a system.

...commercial circuit with eight memoristors from the company Knowm...

Knowm …commercial circuit with eight memoristors from the company Knowm…Knowm

The behavior of the memoristor is different. When subjected to a given voltage in one direction, it acts almost like an insulator: the electric current barely passes through the material. In other words, it is very resistant to current, with low conductivity. If the voltage and the direction of the current are altered, the component becomes a semiconductor or even a metal, with low electrical resistance. The current flows easily. The ability to alternate its conductivity and resistance means that the memoristor can encode information in binary form (0 and 1), like current computer chips. The insulating mode would be 0 and the semiconductor 1 or vice-versa.

The surprising thing is that, in addition to processing data, the memoristor can also store it. This is because the component “remembers” its last conductivity state, if it was the mode equivalent to 0 or 1. When the current feeding it is turned off, the memoristor “remembers” if it was operating in the almost-insulator or semiconductor regime. This is called non-volatility. In computational terms, it means that a circuit based on a type of non-volatile memory can be turned off and turned on again without erasing the information stored in it. Most current computer storage memory, such as hard drives and flash memory, are of this type. “The time it takes to record data in a memoristor is very low, on the order of nanoseconds, and the information is retained for years,” comments Padilha.

In addition to acting like a hard drive to store information long-term, the memoristor can also act like the other type of memory present in computers, RAM (random-access memory). This is a type of volatile memory. When the machine is turned off, everything in RAM is lost. This is the type of memory that allows us to load the programs that are installed on the computer. “By allowing the integration of volatile and non-volatile memory in a single device, a hypothetical computer based on memoristors could simply be unplugged without losing the programs and information stored,” comments Padilha. And, when it is turned on again, the machine would instantly begin to display the data from when it stopped working.

...and chip created by a German university: nanocircuits closer to being marketed

University of Bielefeld …and chip created by a German university: nanocircuits closer to being marketedUniversity of Bielefeld

The possibility of memoristors becoming the heart of a new generation of computers, with an architecture integrating processing chips and two types of memory in a single component, seems reasonable given the advances in nanotechnology. Researchers in academia and from companies like IBM are working on the idea that memoristors are the components most similar to human neural networks and could be capable of imitating synapses.  HP, the leading company in memoristor studies, had promised to launch a computer named “The Machine” in 2016 based on this new technology. But the plans were postponed, officially because of economy of scale concerns.

“Computers based on memoristors are much more feasible than the promises of quantum computing, which need extremely controlled conditions in order to work,” says Ribeiro. “But it is not easy to transfer memoristor technology to an assembly line and manufacture a commercial product. Regardless of their disagreement on the mechanisms that generate the characteristic properties of these components, his colleague Dalpian agrees. “Although it would be easy, in principle, to build memoristors, there are questions related to component quality control that have not yet been totally resolved,” he notes.

Electronic, magnetic and transport properties of nanostructures (nº 2010/16202-3); Grant Mechanism Thematic Project; Principal Investigator Adalberto Fazzio (IF-USP); Investment R$ 1,327,201.88.

Scientific articles
PADILHA, A. C. M. et al. Charge storage in oxygen deficient phases of TiO2: Defect Physics without defects. Scientific Reports. July 1, 2016
YI, W. et al. Quantized conductance coincides with state instability and excess noise in tantalum oxide memoristors. Nature Communications. April 4, 2016