Scientists at the National Institute of Standards and Technology (NIST) have now unveiled the long-mysterious inner workings of these semiconductor elements, which can act like the short-term memory of nerve cells.
Just as the ability of one nerve cell to signal another depends on how often the cells have communicated in the recent past, the resistance of a memristor depends on the amount of current that recently flowed through it. Moreover, a memristor retains that memory even when electrical power is switched off.
Above – Illustration shows an electron beam impinging on a section of a memristor, a device whose resistance depends on the memory of past current flow. As the beam strikes different parts of the memristor, it induces different currents, yielding a complete image of variations in the current throughout the device. Some of these variations in current indicate places where defects may occur, indicated by overlapping circles in the filament (titanium dioxide), where memory is stored. Credit: NIST
But despite the keen interest in memristors, scientists have lacked a detailed understanding of how these devices work and have yet to develop a standard toolset to study them.
Now, NIST scientists have identified such a toolset and used it to more deeply probe how memristors operate. Their findings could lead to more efficient operation of the devices and suggest ways to minimize the leakage of current.
o explore the electrical function of memristors, the team aimed a tightly focused beam of electrons at different locations on a titanium dioxide memristor. The beam knocked free some of the device’s electrons, which formed ultrasharp images of those locations. The beam also induced four distinct currents to flow within the device. The team determined that the currents are associated with the multiple interfaces between materials in the memristor, which consists of two metal (conducting) layers separated by an insulator.
“We know exactly where each of the currents are coming from because we are controlling the location of the beam that is inducing those currents,” said Hoskins.
In imaging the device, the team found several dark spots—regions of enhanced conductivity—which indicated places where current might leak out of the memristor during its normal operation. These leakage pathways resided outside the memristor’s core—where it switches between the low and high resistance levels that are useful in an electronic device. The finding suggests that reducing the size of a memristor could minimize or even eliminate some of the unwanted current pathways. Although researchers had suspected that might be the case, they had lacked experimental guidance about just how much to reduce the size of the device.
Because the leakage pathways are tiny, involving distances of only 100 to 300 nanometers, “you’re probably not going to start seeing some really big improvements until you reduce dimensions of the memristor on that scale,” Hoskins said.
To their surprise, the team also found that the current that correlated with the memristor’s switch in resistance didn’t come from the active switching material at all, but the metal layer above it. The most important lesson of the memristor study, Hoskins noted, “is that you can’t just worry about the resistive switch, the switching spot itself, you have to worry about everything around it.” The team’s study, he added, “is a way of generating much stronger intuition about what might be a good way to engineer memristors.”
Metal oxide resistive switches are increasingly important as possible artificial synapses in next-generation neuromorphic networks. Nevertheless, there is still no codified set of tools for studying properties of the devices. To this end, we demonstrate electron beam-induced current measurements as a powerful method to monitor the development of local resistive switching in TiO2-based devices. By comparing beam energy-dependent electron beam-induced currents with Monte Carlo simulations of the energy absorption in different device layers, it is possible to deconstruct the origins of filament image formation and relate this to both morphological changes and the state of the switch. By clarifying the contrast mechanisms in electron beam-induced current microscopy, it is possible to gain new insights into the scaling of the resistive switching phenomenon and observe the formation of a current leakage region around the switching filament. Additionally, analysis of symmetric device structures reveals propagating polarization domains.
Understanding the physics of EBIC imaging in resistive switching devices has broader implications for the metrology of resistive switching. The direct observation of hot electron currents opens the door to other hot electron techniques such as ballistic electron emission microscopy (BEEM) and internal photoemission (IPE). IPE, being the optical analog to EBIC, sacrifices spatial resolution for precise spectroscopic information. While IPE has long been used to study MIM diodes and other electronic devices its use has not been demonstrated on filaments, probably due to the small active area. An IPE system, most likely combined with high brightness sources, focusing optics and phase sensitive detection, could make it possible to deconstruct the underlying electronic structure of the filament–electrode interface as a function of state.
Though EBIC is clearly applicable for conventional device geometries, our results also show its applicability for other geometries, like lateral devices. If the devices studied here were rotated on their side, with EBIC it would be possible to probe the device depletion region at its interface, as in conventional EBIC, and also determine the onset of filament formation by observing the emergence of ISEE at the electrodes. With a lateral device, it would be easier to see structural changes (such as with electron back scattering diffraction or X-ray absorption) as the device switches, but this could also lead to false positives as regions unrelated to the switching are changed by Joule heating. This problem is particularly acute at large biases where leakage currents could dominate the electrical properties. EBIC then can be an effective, rapid means of disentangling resistive switching from artifacts.
They demonstrate energy-dependent and stateful EBIC measurements on conventional resistive switching devices. Comparing these measurements to Monte Carlo simulations reveals two competing forms of contrast that have not previously been distinguished: classic electron–hole pair separation, and ISEE. Differentiating between these two forms of current generation makes it possible to distinguish the filament from its surrounding recrystallized region. Stateful measurements of the ISEE current show different scaling relationships for the turn-on and turn-off branches, which suggests the existence of different, hysteretic mechanisms for filament formation and dissolution. Symmetric device structures show propagating fronts of different polarizations, depending on the direction of the applied bias prior to the image acquisition. This large area switching suggests that the details of device manufacture and geometry can have a significant effect on the underlying scaling of the resistive switching. These effects are difficult to observe spectroscopically, but become clear with EBIC.