Share This article

Memristors could be big — like, the future of computer memory, and even computing itself, big. But then again, the idea was first floated about 45 years ago, and we still can’t buy a product that uses it: a way of storing a bit of information in a stable, reversible change in electrical conductance. It’s a tough challenge, but engineers from companies like HP believe it could give us terabytes (and allegedly up to hundreds of terabytes!) of high-speed RAM, starting as soon as 2020.

Now, new research from Germany shows that two types of experimental memristor aren’t as different as previously believed, and that researchers still have quite a ways to go before they fully understand the behavior of their own creations.

The advantage of memristors is right in the name: memory resistors, or memristors. Turn the power off to the machine, and the memristors can each remember (maintain) their physical state. As a result, you don’t have to keep a power flowing through them just to keep them from forgetting their state, and so long term they consume considerably less energy than the transistors in classical forms of memory. Much like the liquid crystals in e-ink paper, you only need to use electricity to change the state of the system — it will then keep that state until you expend more energy to change it again.

There has even been very preliminary work trying to use more elaborate versions of memristors as the computational unit in physical neural networks, computers built to run the machine learning algorithms that are quickly dominating the world. Memristor-based neuromorphic chips could be extremely power efficient, but scientists will need to advance the technology quite a bit before they can actually build one. Right now, neuromorphic chips like IBM’s TrueNorth use a very different, “spike based” system.

An experimental HP memristor memory wafer.

Memristors are falling behind because the requirements of a real memristor are very difficult to satisfy, and manufacture. Two major technologies that can get the job done, at least in principle, are electrochemical metallization and valence change.

In electrochemical metallization, a “write” voltage is applied to a copper electrode, oxidizing it and causing copper ions to diffuse across an electrolyte — a diffusion mechanism not unlike that in a lithium ion battery. The copper ions bind to the far electrode and build up, eventually forming a conductive bridge between the two electrodes and changing the resistance of the system. A much smaller “read” voltage can check the resistance without changing it, and an inverted signal will return the system to its original state — functionally erasing the memory.

HP memristor die/wafer shot

Valence change memristors work by a similar mechanism, releasing positively charged metal ions and stimulating the release of both negatively charged oxygen ions from the electrolyte. This time, the bridge formation is mediated by the release of oxygen — at least, that’s what was believed up until now. This team’s results seem to call this into question, as they designed an experiment to tease out the relative importance of metal and oxygen ion mobility.

In a no-oxygen, and even an oxygen-blocking environment, the memristor still performed as expected. The team hopes that with a better understanding of this bridge formation will come a better ability to actually build them, and control them, to fully realize their potential.