Researchers have built a primitive microprocessor out of a two-dimensional material similar to graphene, the flexible conductive wonder material that some believe will revolutionize the design and manufacture of batteries, sensors and chips.
With only 115 transistors, their processor isn’t going to top any benchmark rankings, but it’s “a first step towards the development of microprocessors based on 2D semiconductors,” the researchers at Vienna University of Technology said in a paper published in the journal Nature this month.
Two-dimensional materials have the benefit of flexibility, meaning that they can be incorporated more easily into wearable devices or connected sensors, and potentially making them less breakable: Picture a smartphone that bends rather than breaks if you drop it.
Today’s semiconductors and screens are already pretty thin, but they still rely on the three-dimensional physical properties of the materials they’re made from in order to function. Bend a silicon wafer and it will crack. But 2D materials like graphene or the transition-metal dichalcogenide (TMD) used by the Vienna researchers, are truly two-dimensional, made with crystals just one layer of atoms or molecules thick, allowing them to flex.
TMDs are compounds composed of a transition metal such as molybdenum or tungsten and a chalcogen (typically sulfur, selenium or tellurium, although oxygen is also a chalcogen). Like graphene, they form into layers, but unlike graphene which conducts electricity like a metal, they are semiconductors, which is great news for flexible chip designers.
Stefan Wachter, Dmitry Polyushkin and Thomas Mueller of the Institute of Photonics, working with Ole Bethge of the Institute of Solid State Electronics in Vienna, decided to use molybdenum disulfide to build their microprocessor.
They deposited two molecule-thick layers of it on a silicon substrate, etched with their circuit design and separated by a layer of aluminium oxide.
“The substrate fulfills no other function than acting as a carrier medium and could thus be replaced by glass or any other material, including flexible substrates,” they wrote.
Recent Intel microprocessors act on data in 64-bit “words,” can understand hundreds or even thousands of different instructions, depending on how you count them, and contain hundreds of millions of transistors.
In contrast, the microprocessor built by the researchers is only capable of acting on data one bit at a time, using a set of just four instructions (NOP, LDA, AND and OR), and the circuit features used to build it are of the order of 2 micrometers across, 100 times larger than those found in the latest Intel and ARM processors.
With more work, though, the microprocessor’s complexity could be increased and its size reduced, the researchers said. They deliberately chose an overly large feature size for their manufacturing process to reduce the effects of holes, cracks and contamination in the molybdenum disulfide film and to make it easier to inspect the results with an optical microscope.
“We do not see any roadblocks that could prevent the scaling of our 1-bit design to multi-bit data,” they said, and only the challenge of lowering contact resistance stands in the way of sub-micrometer manufacturing.
That’s not to say it will be easy: Although the manufacturing yield for subunits was high, with around 80 percent of the arithmetic-logic units fully functional, their non-fault tolerant design meant only a few percent of finished devices worked properly.
Commercial microprocessor manufacturers deal with yield problems by making their chip designs modular, and testing them at a variety of speeds. Chips that work at higher speed fetch higher prices, while faulty subcomponents can be permanently disabled and the resulting chips, otherwise fully functional, sold as lower-spec models.
It’s taken 46 years for Intel to get from the 4004, a four-bit central processor with 46 instructions, to the latest incarnation of the x86 architecture, Kaby Lake: With all that the industry has learned about micromanufacturing since then, progress with flexible semiconductors may be a little faster.