Analysis
Researchers at South Korea's KAIST and the startup TDS Innovation published a technique in Nature on October 7 that controls where 2D semiconductor crystals begin growing on a wafer, reported by The Register on October 11. Uncontrolled growth normally produces multiple crystals whose boundaries degrade a chip's electrical performance; the new method, which the team calls etching-flux-mediated single-centered nucleation, uses oxygen released from an oxide barrier to suppress crystal formation everywhere in a growth region except its geometric center, forcing one clean crystal per patterned site.
Tested across a 2-centimeter substrate of molybdenum disulfide, the technique produced single crystals at 397 of 400 patterned sites -- a 99.3% yield -- and the resulting field-effect transistors showed higher charge-carrier mobility than previously reported for selectively grown MoS2 devices. That's a specific, reproducible number in a peer-reviewed paper, a higher evidentiary bar than most "breakthrough chip material" announcements clear.
“That's a specific, reproducible number in a peer-reviewed paper, a higher evidentiary bar than most "breakthrough chip material" announcements clear.”
TDS Innovation, co-founded by paper co-author Kibum Kang, makes equipment and materials for 2D semiconductor devices, so the Nature result functions as validation for the startup's own roadmap as much as a general scientific advance. TDS representative Jony Jung said the company is "aiming to enable commercial use of 2D semiconductors around 2030," with timing dependent on further validation -- a multi-year horizon that puts this firmly in deep-tech territory rather than a near-term product story.
The strategic case for 2D semiconductors is about geometry: stacking logic and memory closer together, including through complementary FETs that place n-type and p-type transistors vertically, could shrink the distance data has to travel on a chip and cut the energy that distance costs -- a real constraint on how dense and efficient AI accelerators can get as silicon scaling slows. TDS points to AI demand, including physical AI and robotics, as a likely accelerant for adoption, though The Register notes the paper itself doesn't directly tie the result to AI compute, and a 99.3% lab yield on a 2-centimeter test wafer is still a long way from a production-scale 300mm fab line. For deep-tech investors, the open question isn't whether the chemistry works -- it's whether a university spinout can turn a Nature paper into a foundry relationship before a better-funded incumbent reproduces the result.
