A robotic production line revisits 2D materials to produce multi-layered structures | To research


A new technique for scalable production of twisted van der Waals heterostructures has been used to produce twisted stacks with many more layers than before. The applications are not yet clear, but it could allow engineers to exploit the many fascinating phenomena discovered in twisted multilayer heterostructures such as superconductivity.

Following theoretical predictions by Allan MacDonald of the University of Texas at Austin and colleagues, Pablo Jarillo-Herrero of the Massachusetts Institute of Technology and colleagues showed in 2018 that in bilayer graphene, if the sheets were twisted together compared to others by only 1.1°, the material could become either a superconductor or an insulator. Subsequently, Jarillo-Herrero’s group and others discovered exotic behavior in graphene comprising three, four, and five layers, as well as in twisted multilayers of materials like transition metal dichalcogenides and heterostructures.

Achieving these structures, however, involves carefully searching for single-crystal regions on the surface of the bulk material, mechanically exfoliating them, and finally stacking them at the required offset angle. Andrew Mannix, Jiwoong Park and their colleagues at the University of Chicago in Illinois and elsewhere wanted a more scalable method. Wafer-scale versions of 2D materials “good enough for LEDs and transistors” can now be produced by chemical vapor deposition, Park says. “We wondered how to integrate these large-scale materials into devices in which these complicated structures can be produced in large quantities and with high throughput,” he explains.

The technique involves automatically cutting a slice of each material into a large number of identical, pre-programmed and customizable shapes. A robotic arm is then used to pick up the monolayer, move it to the next wafer, twist it to the desired angle, and stack it on top. In this way, an entire heterostructure can be assembled in a process similar to silicon chip fabrication today. The wafers are of uniform quality, so each wafer can be cut in the same position, allowing parallel processing of the manufacturing process without the need to search for perfect single crystal regions and carefully position them on top of each other. “All the basic pieces were there,” Park explains. “Like everything else, what’s really important is the willingness to integrate everything.” When the researchers tried the technique with exfoliated samples, the quality of the material was preserved – the quality of the finished heterostructure is therefore only limited by the quality of the wafers.

The researchers produced and characterized several heterostructures that could not be fabricated with current methods, such as one comprising 25 layers of molybdenum disulfide diagonally interspersed with tungsten disulfide. “We performed clear optical spectroscopy measurements on all of these materials continuously as we added one layer, then another layer, and what we found – somewhat surprisingly – was that the individual colors add up like paint, and it becomes denser and more opaque. This is something one would expect if it were a totally non-interacting material.

Heterostructures avoid graphene and hexagonal boron nitride because reliably separating them from their growth substrates and depositing them into heterostructures without damage is not, at present, practical, Park says. Researchers can produce heterostructures at a rate of around 30 layers per hour, and Park thinks this could be significantly improved with commercial investment. “The question is always what will it replace and why? Park said. “We are showing that it is possible and we hope it will allow people to think more freely.”

David Goldhaber-Gordon of Stanford University in California, who was not involved in the research, is impressed. “When I imagine trying to create a four- or five-layer twisted graphene stack with mostly manual techniques, I find it terrifying,” he says. “When I talk to people in the field, they often tell me: “We take 20 samples and we take the one that works well”. You can do a lot of good this way. but I think this method that Andy and Jiwoong demonstrated really provides the confidence of being able to accurately create multiple samples with the same structure and also opens up a realm of dozens of layers that just wouldn’t happen with manual techniques.


Comments are closed.