Etching flux pins single crystal growth of 2D semiconductors at exact pattern centres

For decades, semiconductor engineers have been able to decide roughly where a material should grow, but not precisely where within that region the very first crystal will appear. That missing degree of control has been one of the quiet frustrations of two-dimensional electronics, where atomically thin semiconductors such as molybdenum disulfide promise to carry transistor technology beyond the limits of silicon, yet resist the kind of deterministic manufacturing that the chip industry demands. Now a team of researchers in South Korea and the United States reports a strategy that closes this gap, using a lateral flux of etching species released from a patterned oxide barrier to force a single crystal nucleus to form at the geometric centre of each growth site, every time.

The work, published in Nature, describes a technique the authors call etching-flux-mediated single-centred nucleation, or EF-SCN. Conventional area-selective growth relies on a difference in surface binding energy between a growth region and a surrounding barrier: precursor molecules stick preferentially to the growth area, so deposition occurs only where intended. The problem is that the entire growth region behaves as a nucleation-active zone with a roughly uniform probability of crystal birth. Nuclei appear at random positions, often clustering at pattern corners and edges, and even when the region is shrunk enough to permit only one nucleus, its position remains a matter of chance. EF-SCN changes the physics of the problem by adding an in-plane chemical flux that reshapes the nucleation probability landscape itself.

The key ingredient is the barrier material. The researchers found that amorphous hafnium oxide, a common high-k dielectric, is thermally unstable at the temperatures used for molybdenum disulfide growth. X-ray diffraction showed monoclinic crystallization peaks appearing above roughly 550 degrees Celsius, and X-ray photoelectron spectroscopy revealed that at the typical growth temperature of 750 degrees Celsius the oxygen content of the film dropped rapidly from HfO1.95 to HfO1.83 within ten minutes before saturating. In other words, the barrier steadily exhales oxygen during the deposition process. Because oxygen is a known etchant for molybdenum disulfide, this released gas acts as a lateral etching flux that sweeps across the growth region from the pattern boundary inward.

The consequence is a spatially graded competition between growth and etching. Close to the barrier, the oxygen concentration is high enough that any nascent molybdenum disulfide nucleus is etched away as fast as it forms, creating a nucleation-suppressed region hugging the pattern edge. Farther toward the centre, the oxygen flux decays exponentially, following a diffusion model the team derived from Fick’s first law, and at some critical distance the net growth rate turns positive. Only the central core of the pattern remains a viable nucleation-active region, and within that core the probability of nucleation peaks at the geometric centre. The result, confirmed by scanning electron microscopy across large arrays of triangular and square patterns, is that a single molybdenum disulfide flake nucleates dead centre and then grows laterally until it fills the entire region as one seamless crystal.

What makes the demonstration compelling is not just the concept but the statistics. Across a two-centimetre substrate divided into sixteen regions of twenty-five patterns each, four hundred growth sites in total, the team achieved a single-crystal yield exceeding 99 percent. The resulting flakes reached approximately ten micrometres across, an order of magnitude relevant to practical device layouts, and atomic force microscopy confirmed monolayer thickness of 0.75 nanometres. Rigorous structural characterization, including transmission electron diffraction, atomic-resolution scanning transmission electron microscopy, atomic-force imaging and angle-resolved second-harmonic-generation mapping, consistently showed that the flakes were genuinely single-crystalline rather than stitched-together polycrystals, a distinction that matters enormously for carrier transport.

Perhaps most striking is the breadth of process compatibility. The technique worked on crystalline sapphire but also on amorphous silicon dioxide, where uniform single-crystal growth was achieved over a four-hundred-micrometre area even though crystal orientations varied randomly between sites, as expected on a non-epitaxial surface. Pattern densities spanning nearly an order of magnitude still yielded effective single-crystal growth. And by swapping the hafnium oxide barrier for zirconium dioxide, which releases its etching flux at lower temperature, the researchers pushed the entire process down to 430 degrees Celsius, producing well-defined centred single crystals and even fully filled ten-micrometre single-crystalline films. Low-temperature growth is a prerequisite for back-end-of-line integration, where atomically thin transistors would be fabricated atop finished silicon circuitry without damaging it.

Beyond demonstrating the phenomenon, the team distilled it into a quantitative design rule. By varying the growth-region size, the process time and the precursor flows during metal-organic chemical vapour deposition, they mapped out three distinct regimes: zero nucleation, when the nucleation-active region never forms; single nucleation, when only the central point survives; and multiple nucleation, when the active region broadens enough to host several nuclei. Growth regions of three, five and seven micrometres produced exactly these three outcomes under fixed conditions, while tuning the diethyl sulfide chalcogenine flow from 0.5 to 7 standard cubic centimetres per minute traced the same progression. The chalcogen flow primarily governed nucleation density, whereas the molybdenum hexacarbonyl flow controlled lateral growth rate, giving engineers two independent knobs. A kinetic Monte Carlo emulation with spatially separated precursor and etchant reservoirs reproduced all three regimes, reinforcing the physical picture.

The device implications follow directly. Because nucleation position and growth region can be decoupled, the researchers joined a large nucleation zone to a narrow growth-only channel, coaxing a single crystal to nucleate in one place and extend as a line-shaped stripe into another. These line-shaped channels, roughly 6.8 micrometres long, became the backbones of field-effect transistors exhibiting a peak field-effect mobility of 117 square centimetres per volt-second, the highest reported for selectively grown molybdenum disulfide, with an on-current density of 12 microamperes per micrometre. Even more intriguingly, four separate transistors were defined along a single ten-micrometre crystal, and their mobility variation collapsed to a relative standard deviation of 7.8 percent, a 65 percent improvement over devices carved from separate crystals. Sharing one flawless lattice apparently irones out much of the device-to-device scatter that has plagued two-dimensional electronics.

The same spatial programming enabled something previously out of reach: self-aligned contacts built from aligned lateral heterostructures. By alternating precursor supplies, the team grew stripe-shaped molybdenum disulfide and molybdenum diselenide superlattices with interfaces sharper than two nanometres, positioned and oriented identically across whole arrays. Because local strain at the heterojunction makes area-selective atomic layer deposition of ruthenium favour the diselenide segment, metal contacts deposited themselves precisely where the heterostructure dictated, yielding self-aligned field-effect transistors with on/off ratios above ten million and subthreshold swings of 170 millivolts per decade, achieved despite the harsh oxygen exposure of the ruthenium process. Taken together, the results suggest that the hardest part of two-dimensional manufacturing, deciding where a crystal begins, may finally be a matter of barrier design rather than luck, opening a route toward wafer-scale integration of atomically thin logic, optoelectronics and strain-engineered devices built crystal by crystal, site by site.

Subject of Research: Deterministic control of nucleation site in the selective growth of two-dimensional semiconductor single crystals using an oxygen etching flux released from oxide barriers

Article Title: Spatially deterministic nucleation of 2D semiconductors by etching flux

Article References: Park, J., Kim, J., Kang, S., Oh, J., Sunwoo, Y., Oh, S., Shin, S., Yang, S., Jo, S., Lee, J., Lee, D., Kim, M., Eom, C., Jeong, S., Lim, C., Moon, G., Kang, M., Kim, M.-G., Kim, J., … Kang, K. (2026). Spatially deterministic nucleation of 2D semiconductors by etching flux. Nature, 658(8135), 398-406. https://doi.org/10.1038/s41586-026-11095-1

Image Credits: AI Generated

DOI: 10.1038/s41586-026-11095-1

Keywords: two-dimensional materials, molybdenum disulfide, nucleation control, area-selective growth, chemical vapor deposition, single crystals, field-effect transistors, lateral heterostructures, hafnium oxide, semiconductor manufacturing, etching flux, 2D electronics