Research Study Identifies Tiny Material Differences and Their Importance for Memristive Device Behavior

Memristive devices have been heavily researched in both academia and industry for the last two decades. While they were originally proposed as binary non-volatile random access memories (RAM), and indeed early research was carried out in the search for higher density memories, their characteristics make them useful for more novel computing applications such as in-memory and analog computing. 

Capable of extremely low power operation and behaving similarly to the human brain’s neurons, memristive devices are thought to be a highly promising alternative to nanoelectronic elements in modern computer chips.

Now, researchers from the Jülich Aachen Research Alliance (JARA) and the German technology group Heraeus claim to have discovered how to control their functional behavior.

 

Tiny Material Differences

Memristive devices are being so heavily researched because of their advantageous functionalities and the promise they hold for electronics. Today, major companies like IBM, HP, Intel, and Samsung are all carrying out research into memristive elements and devices with a view to bringing new types of computer and storage devices to market. 

A fundamental feature—if not the foremost one—of memristors is their ability to switch from high resistance to low resistance and back again. In theory, this means the devices are adaptive, similar to the synapses in our own nervous system. "Memristive elements are considered ideal candidates for neuro-inspired computers modelled on the brain, which are attracting a great deal of interest in connection with deep learning and artificial intelligence," says Dr. Ilia Valov.

 

Small Material-Level Differences 

Now, memristive devices may find potential in a wider range of applications thanks to the team’s research. It highlights “the smallest differences” in material composition that make big differences in how a memristive device behaves in terms of efficiency and reliability. These differences are so small in fact that until now, researchers have failed to notice them. 

 

Dr. Ilia Valov (left) in the Oxide Cluster at Forschungszentrum Jülich, where the team’s experiments were carried. In the background are Michael Lübben (center) and Prof. Rainer Waser (right). Image credited to RWTH Aachen / Peter Windany

 

Selectively Controlling Switching and Neuromorphic Behavior

This research describes how these small differences make it possible to selectively control the switching and neuromorphic behavior of these memristive elements, in a similar way to semiconductor doping, and design very specific memristive systems. This is achieved by introducing foreign atoms which allows the control of the solubility and transport properties of the thin oxide layer found in memristive elements. 

According to the team’s findings, it is the switching oxide layer’s purity that is the crucial factor here.  "Depending on whether you use a material that is 99.999999 % pure, and whether you introduce one foreign atom into ten million atoms of pure material or into one hundred atoms, the properties of the memristive elements vary substantially," says Valov.

According to Valov, his team’s insights could be used by manufacturers to methodically develop memristive elements and only select the functions that they need. A higher doping concentration, for example, means the slower the resistance of the memristive elements changes as the incoming voltage pulses fluctuate, and the more stable this resistance remains. "This means that we have found a way for designing types of artificial synapses with differing excitability," explains Valov.