Researchers make RAM from a phase change we don’t entirely understand

By John Timmer

Illustration of atoms.
Enlarge / Two layers of one of the materials used in this work.

We seem to be on the cusp of a revolution in storage. Various technologies have been demonstrated that have speed approaching that of current RAM chips but can hold on to the memory when the power shuts off—all without the long-term degradation that flash experiences. Some of these, like phase-change memory and Intel's Optane, have even made it to market. But, so far at least, issues with price and capacity have kept them from widespread adoption.

But that hasn't discouraged researchers from continuing to look for the next greatest thing. In this week's edition, a joint NIST-Purdue University team has used a material that can form atomically thin sheets to make a new form of resistance-based memory. This material can be written in nanoseconds and hold on to that memory without power. The memory appears to work via a fundamentally different mechanism from previous resistance-RAM technologies, but there's a small hitch: we're not actually sure how it works.

The persistence of memristors

There is a series of partly overlapping memory storage technologies that are based on changes in electrical resistance. These are sometimes termed ReRAM and can include memristors. The basic idea is that a material can hold a bit that is read based on whether the electrical resistance is high or whether electrons flow through like it was a metal. In some of these, the resistance can be set across a spectrum that can be divided up, potentially allowing a single piece of material to hold more than one bit.

So far, two mechanisms have been used to change the resistance. The first is phase-change memory, which relies on materials that can take on two forms when they solidify after being melted. If they're cooled quickly, they form a disorganized structure that conducts electricity poorly. A slow cool-down will allow them to form an organized, crystalline structure that's a better conductor. In an alternate approach, heating accompanied by high voltage will force some materials to allow a small thread of metal to form between electrodes, dramatically dropping the resistance between them.

The team behind this new work was looking to do a slightly different phase transition using a class of atomically thin materials. Collectively termed metal dichalcogenides, these materials form single-molecule sheets with a hexagonal arrangement of atoms similar to graphene. But, rather than being a completely flat, planar molecule, these materials have a somewhat more complex arrangement of atoms, with some slightly above or below a plane running through the center of the sheet.

In phase

The most commonly used example is molybdenum disulfide (MoS2), but it is a complex family of materials. It's possible, for example, to substitute out either of these elements with their equivalents in the same column of the periodic table. It's also possible to make mixed materials, with both molybdenum and tungsten in the same chunk of material. There's also some flexibility in how the atoms arrange within a sheet. And, as solids can be built up by layering multiple sheets, there's also flexibility in how neighboring sheets line up.

All of these complexities allow these materials to adopt multiple phases while remaining solid. (Think of these like the various forms of water ice that can form at different temperatures and pressures.) These phases have different conductive properties, and they include semiconductors, semi-metals, and metallic phases, which makes them great candidates for resistance-based RAM.

To construct a ReRAM device, the researchers kept things simple: a conducting base with some of the material—molybdenum ditelluride, in this case—on top of it and another electrode layered on top of that. Rather than a single sheet of MoTe2, they used chunks that were anywhere from five to 25 nanometers thick (a single layer is only about a third of a nanometer). Despite the simplicity, this worked. A high-voltage pulse of less than 10 nanoseconds was enough to shift them from a high-resistance state to one where current flowed over 50 times more readily when the voltage was dropped. A second pulse of higher voltage could shift it back, with the precise voltages needed determined by the thickness of the MoTe2 used.

It was only when the researchers started looking at why this happens that things got a bit strange. One possibility was the formation of metal filaments between the electrodes. The researchers eliminated this explanation by changing the electrodes to graphene and showing that the device still worked. The MoTe2 shouldn't undergo the sort of crystalline/amorphous change that underlies phase-change memories, either. So what exactly is going on?

To find out, the authors set the state of the device, removed the MoTe2 layer, and then subjected it to electron microscopy. This isn't sufficient to determine the structure down to the atomic detail. But it did give an idea where the molybdenum atoms were located and how the spacing between layers was arranged.

Away from the electrodes, the MoTe2 was in a previously known phase. But in the area directly under the electrodes, where the resistance had changed, this ordered structure had shifted to a different form. The new structure didn't look like anything that had been described for MoTe2 previously. The authors suspect that it's normally a temporary, transitional state between phases that somehow ends up stabilized in these circumstances. Somehow, the electric field seems to have triggered this phase change and its accompanying change in conductivity.

But wait, there’s more!

The folks at Purdue weren't the only ones thinking about the potential utility of getting these metal dichalcogenides to change state. As this report was being written last week, we became aware that a paper from researchers at the University of Michigan would be released this week describing something substantially similar. That team controlled the transition of molybdenum disulfide between insulating and metallic states by using an electric field. In this case, the electric field controlled the presence of lithium ions in the spaces in between layers of the MoS2.

Sufficiently high levels of lithium result in the transfer of an electron to molybdenum (which in turn alters the structure of the material) and a boost of its conductance. Reversing the control voltage could then force the lithium back out, returning it to a semiconducting state.

In this case, the researchers built several devices within a larger sheet of multi-layer MoS2. This allowed the researchers to have one device influence its neighbors through mechanisms like having them all share a limited pool of lithium. In this case, one device being set in the "on" state makes it more difficult for its neighbors to adopt a similar state. At the same time, clearing the lithium out of a device changes the conductance of all its neighbors.

This, the researchers argue, can mimic the behaviors seen in some neurons, where strengthening the connection on one collection between two neurons can influence the behavior of neighboring connections.

There's a big gap between these test devices and a technology that can scale and be easily integrated with electronic manufacturing. So these demonstrations have a long way to go before they end up in your phone. But the interesting thing about them is that they don't rely on the fact that the metal dichalcogenides can be used as an isolated, atomically thin sheet. Instead, it's the fact that these sheets can form layers that determines the behavior of these ReRAM devices.

Nature Materials, 2017. DOI" 10.1038/s41563-018-0234-y, 10.1038/s41563-018-0248-5  (About DOIs).