Engineers from US Department of Energy's Lawrence Berkeley National Laboratory have developed a method for fabricating transistors and circuits that are just a few atoms thick. The technique depends on two key materials, conducting graphene and a semiconductor called transition-metal dichalcogenide (or TMDC). Because both materials are atomically thin, this opens the door to electronics that are effectively two dimensional. Moreover, the method is already scalable to practical applications.
This is good news for Moore's law, of course, which predicts a doubling of transistor density every two years. Said law has been guiding and inspiring semiconductor research for close to 50 years, but it's now entering its most precarious era. At 5 and 10 nanometer scales, computer engineering is running up against some very fundamental laws of physics, particularly when it comes to building things with silicon.
Simply, silicon is a bulk material. Silicon's semiconducting properties occur thanks to additional dopants added to its crystal lattice structure, and taking advantage of this arrangement requires the usage of a bulk quantity of the material. Getting around this apparent lower limit of silicon needed to produce a semiconductor will require new materials, it seems.
"When you get to these scales, you won't be able to turn silicon on and off anymore," Mervin Zhao, lead author of a paper published Monday in Nature Nanotechnology describing the Berkeley research, told me. "It acts as a bad switch."
Enter graphene, a tantalizing conducting supermaterial with the very desirable property of being a two-dimensional crystal. While single-layer graphene lacks an electronic bandgap (and thus the ability to be turned "off") and won't work so well within a transistor itself (which is essentially an off/on switch), it may be practical to employ it in wiring and interconnections within next-generation devices.
Here, the idea is to use graphene as a conductive substrate through which channels can be lithographically etched and then seeded with a TMDC called molybdenum disulfide, which then continues to grow on its own (as is common in nanoscale construction). The etching has the effect of leaving tiny defects and barbs along the newly carved edges of graphene, which induce the TMDC to nucleate and grow preferentially along these edges. In this way, the TMDC grows only within the channels and not on the relatively smooth upper and lower surfaces of the graphene.
While 2D graphene can't easily be turned off, its electrical properties can be very carefully tuned. It's possible to tighten the electrical contact point between the graphene and the molybdenum disulfide into single Ohmic junctions, or points where current is passed from a semiconductor to a conductor and vise versa. This allows for the "heterostructure" needed to craft functional transistors where graphene is able to act as an electrode tasked with injecting current into the molybdenum disulfide.
In the transistors below, the graphene sections would be located underneath the off-white patches, while the small light-blue squares are the TMDC. The on/off state of the transistor would be given by the conductivity of the TMDC bridge connecting the graphene electrodes on either side. This state represents information.
As the paper notes, this basic idea has been realized before, but more so as a proof of concept and not with this degree of scalability. While previous attempts have utilized relatively "manual" methods of construction, here the researchers chemically grow their circuits. In a sense, they're self-assembling. This is made possible by the dual nature of graphene as both a growth template for the TMDC and as an electrical contact in the finished product.
To demonstrate the scalability of their technique, the Berkeley researchers built a functional millimeter-scale inverter, or NOT gate. "Having now demonstrated a logic application, it is clear that control over arbitrarily designed patterns will lay the foundation for chemically grown atomic computing," they conclude.
While Zhao and his team seem to have solved a very deep problem, it remains for now more fundamental rather than practical knowledge. While the transistors here may be a single nanometer thin, we're still at micrometer scales in terms of length and width. Scaling the technology down past the limits of silicon, where the transistor length and width are more comparable to the material's thickness, is not foretold.
"Whether when I go down to 5 nm it's going to be operational as a transistor, I don't know," Zhao said. "Maybe it will, but it's more addressing the problem of how to assemble a lot of materials together to make circuits. It's a first step kind of thing."