Friday, 20 November 2026

Researchers guide 2D semiconductor growth to steadier chip yields

A KAIST-led team used oxygen leaking from a barrier to make one perfect crystal start in exactly the right spot, a step toward stacked AI chips.

Researchers in a lab setting examining materials

The short version

  • A KAIST-led team published a method in Nature that decides exactly where a 2D semiconductor crystal starts to grow.
  • Oxygen released from a hafnium oxide barrier blocks crystals near the edges, so a single crystal forms in the center.
  • The team grew one molybdenum disulfide crystal at the chosen spot in more than 99% of 400 test patterns.
  • The goal is stacked AI chips that put memory right on top of logic, but wafer-scale tests are still needed.
Quick read · 1 min

Researchers led by Professor Kibum Kang at KAIST in South Korea have found a way to make atom-thin semiconductor crystals start growing exactly where they want. Normally, several crystals form at random inside each patch, and the seams between them hurt performance.

The team surrounded each growth area with a hafnium oxide barrier. Oxygen from the barrier blocks crystals from forming near the edges, so only one starts in the middle and grows outward. It worked in more than 99% of 400 test patterns, and the study appeared in Nature in early October.

  • Why it matters: it helps stack memory on top of AI processors
  • Possible payoff: faster AI responses, lower data center power use
  • Timing: chipmaking like this is expected in the 2030s

Next, the team and startup TDS Innovation have to prove it works across whole wafers and in full circuits.

When atom-thin semiconductor crystals grow on a chip, they usually pop up wherever they like. A team in South Korea has now made them start in one chosen spot, almost every time. In more than 99% of 400 test patterns, a single crystal formed exactly at the center.

That sounds like lab trivia, but it goes to the heart of why AI is so power-hungry. A big share of the time and energy an AI chip uses goes into shuttling data between where it’s stored and where it’s processed. Stacking memory directly on top of computing circuits would shorten that trip, and 2D semiconductors are one of the best materials for building those extra layers. The catch has always been growing them cleanly, in the right place. This work tackles that 2D semiconductor growth problem head on.

01

Who did it and where it was published

The research was led by Professor Kibum Kang of the Department of Materials Science and Engineering at KAIST, South Korea’s leading science and technology university. He worked with teams at UNIST and Hanyang University and with TDS Innovation, a startup that makes equipment and materials for 2D semiconductor devices. Kang is also one of its two co-CEOs.

The paper appeared in the journal Nature in early October. The team calls the method etching-flux-mediated single-centered nucleation, or EF-SCN for short.

Rocket launch
02

Why random crystals ruin 2D chips

Random starting points are the problem because they create flaws where crystals collide. A 2D semiconductor is only a few atoms thick. Graphene is the famous 2D material, but it can’t switch off like a transistor needs to. Materials such as molybdenum disulfide (MoS₂) can, which is why chipmakers are interested.

When these films grow, crystals begin forming at several random points inside each patch at once. Each one has its atoms lined up in its own direction. When neighboring crystals grow into each other, they leave seams called grain boundaries. Electrons trip over those seams, so the material conducts worse and devices on the same wafer behave differently from one another.

Chipmakers could already limit growth to a marked-out region. What they couldn’t control was where inside that region crystals would start, or how many would. Kang described crystal formation as “a matter of probability” in conventional thin-film growth.

03

How oxygen steers 2D semiconductor growth

The trick is to use a chemical brake that is strongest at the edges of each patch and weakest in the middle. Each growth area is surrounded by a barrier made of an oxide, such as hafnium oxide. That barrier gives off oxygen, which gently eats away at any new crystal trying to form nearby. Scientists call that kind of chemical flow an etching flux.

Because the barrier rings the patch, the braking effect fades toward the center, the point farthest from every edge. So the only place a crystal can get started is right in the middle. Once that single seed exists, the same effect keeps new crystals from sprouting around it, and the one crystal grows outward into a larger piece with no seams.

The results were strong. The team grew one MoS₂ crystal at the planned central spot in more than 99% of 400 patterns. It also built working transistors, the on-off switches chips are made of, and reported that charge moved through them more easily than in earlier MoS₂ transistors grown in selected areas. In another test, they grew side-by-side patches of two different 2D materials and coated just one of them with a thin layer of ruthenium metal, showing how precise the placement can be.

Telescope
04

What this could mean for your phone, laptop and AI bills

For everyday people, the payoff is faster AI responses and less electricity burned in data centers, if the method makes it into factories. That’s the researchers’ stated aim, not a promise.

Today, a chip spends energy moving data back and forth between processors and memory. Stacking memory directly over the logic that uses it shortens that distance, which can cut delay and power use. To do that, makers need to grow good semiconductor layers right on top of circuits that are already built, rather than making the material elsewhere and peeling it over. Growing crystals in exactly the planned spots is what makes that kind of bottom-up building realistic.

Two-dimensional materials are also being studied for a stacked transistor design in which the two types of transistors that make up a logic circuit sit one above the other instead of side by side. IBM and Intel have both shown off this kind of 3D stacking.

None of this will change the phone in your pocket next year. KAIST expects chips built with deposited semiconductors, layers grown directly onto existing circuits, to be adopted in the 2030s. TDS Innovation has said it is aiming for commercial use of 2D semiconductors around the turn of the decade, and thinks demand from AI, robotics and physical AI could speed that up.

05

What still has to happen

Several big tests remain before any chipmaker uses this. The team still needs to prove the method works evenly across an entire wafer, not just on small test areas, and to check how reliably full devices and circuits perform.

TDS Innovation plans to combine the research with its own equipment, materials and processes to turn it into a manufacturing method for whole wafers. The company says production timing depends on that further validation. For now, the result shows that where a crystal begins can become a setting engineers design, rather than luck.

06

Quick answers

What are 2D semiconductors?

They’re materials only a few atoms thick that still behave like semiconductors, such as molybdenum disulfide. Because they’re so thin, they can be stacked in layers on top of existing chip circuits.

What did the KAIST team actually achieve?

They made a single crystal start growing at the center of each patterned area, in more than 99% of 400 tests, by using oxygen from a surrounding barrier to block growth near the edges.

When could chips made this way reach products?

Not soon. KAIST expects this kind of chipmaking to arrive in the 2030s, and wafer-scale testing still has to be done.

You're reading the quick version.