Wafer-scale synthesis of air-stable two-dimensional magnets by pulsed chemical vapour deposition

Abstract

Scalable synthesis remains a bottleneck for two-dimensional magnets. Exfoliated chromium halides and tellurides established the physics of atomically thin magnetism, but many candidate materials are air sensitive, compositionally narrow, or available only as small flakes. We report a pulsed chemical vapour deposition route for wafer-scale, air-stable 1T-CrTe2 films with thicknesses from 2.4 to 14.8 nm on 100 mm sapphire, thermally oxidised silicon, and transferred hBN. Alternating chromium-precursor and tellurium-overpressure pulses separated by hydrogen/argon purge steps suppress secondary Cr2Te3 and Te-rich phases while allowing low-temperature lateral growth at 385-425 deg C. On sapphire, 7.1 nm films showed 96.4% surface coverage, root-mean-square roughness of 0.48 nm over 25 um2, and a wafer-scale Te:Cr ratio of 1.98 +/- 0.05 by X-ray photoelectron spectroscopy. Cross-sectional transmission electron microscopy confirmed layered 1T stacking over micrometre domains separated by low-angle rotational boundaries. Magnetic measurements gave perpendicular ferromagnetism with Curie temperature 286 +/- 8 K for 7 nm films and 318 +/- 7 K for 14 nm films, coercive field 78-132 mT at 10 K, and anomalous Hall hysteresis in patterned Hall bars. After 90 days in laboratory air at 35-55% relative humidity, 7 nm films retained 91% of their saturation Kerr contrast and showed only a self-limited surface oxide below 0.8 nm. The process does not yet produce monolayer magnets with uniform long-range order, but it demonstrates a manufacturable route to wafer-scale air-stable van der Waals ferromagnetic films compatible with back-end thermal budgets.

Introduction

Magnetism in strictly two-dimensional crystals is constrained by thermal fluctuations unless anisotropy or long-range interactions open a spin-wave gap [1]. The modern field of van der Waals magnetism grew rapidly after intrinsic ferromagnetism was observed in few-layer Cr2Ge2Te6 and monolayer CrI3 [2,3,4]. Since then, metallic Fe3GeTe2, chromium tellurides, chalcogenide-halides, and related heterostructures have widened the materials palette and raised Curie temperatures toward device-relevant ranges [5,6,7,8].

The synthesis problem has not kept pace with the physics. Many benchmark 2D magnets are exfoliated from bulk crystals, which gives high crystalline quality but poor wafer coverage. CrI3 is chemically fragile in air. Fe3GeTe2 and Fe5GeTe2 can show high Curie temperatures and have been grown as large-area films, but stoichiometry and interfacial coupling remain sensitive to growth conditions [5,6,7,8]. Air-stable magnetic semiconductors such as CrSBr are promising, yet their scalable thin-film synthesis is less mature than their exfoliated-crystal physics [15,16,17].

Chromium tellurides occupy an attractive middle ground. CVD-grown 1T-CrTe2 has been reported as an air-stable two-dimensional ferromagnet with unusual thickness-dependent Curie temperature [9]. Epitaxial CrTe2 ultrathin films can be intrinsically ferromagnetic near room temperature [10], and recent work has pushed large-scale CrTe2 integration onto amorphous silicon wafers and Si-CMOS-compatible processes [12,13]. These advances suggest that CrTe2 is a realistic platform for process engineering, not only for small-flake studies.

This article reports a fictional pulsed-CVD process designed to make that platform more uniform. The central idea is simple: separate chromium delivery from tellurium overpressure in time. Continuous co-flow growth produces narrow process windows because Te deficiency creates Cr-rich phases while excess Te produces droplets and weakly bonded surface layers. Pulsed delivery lets surface reactions proceed under alternating metal-limited and chalcogen-rich conditions, improving stoichiometry control without raising the wafer temperature beyond a low thermal budget.

The goal is not to claim a record Curie temperature. It is to establish a reproducible synthesis window, quantify wafer-scale variation, and test air stability with the same seriousness as magnetic hysteresis. For a practical 2D magnet, a sharp Kerr loop on one flake is not enough; the film must survive lithography, storage, and spatial mapping.

Pulsed-CVD growth design

Growth was performed in a horizontal cold-wall CVD reactor modified with independent fast valves for chromium and tellurium precursor lines. Chromium hexacarbonyl was delivered from a temperature-stabilised bubbler, while elemental tellurium was evaporated from a separate upstream zone. The carrier gas was 95:5 Ar/H2 at 12-28 Torr. One growth cycle consisted of a chromium pulse, a purge, a tellurium pulse, and a second purge. The standard recipe used 8 s chromium delivery, 20 s purge, 12 s tellurium delivery, and 28 s purge, repeated 90-540 times depending on target thickness.

Substrates were 100 mm c-plane sapphire, 100 mm Si/SiO2 with 285 nm thermal oxide, and exfoliated hBN transferred onto SiO2 witness wafers. Sapphire was used for the primary wafer-scale mapping because it provided the lowest roughness and clearest X-ray diffraction peaks. SiO2 was used to test integration with ordinary lithography, while hBN was used to separate intrinsic film quality from substrate-induced disorder.

The substrate temperature was varied from 350 to 460 deg C. Below 370 deg C, nucleation density was high but lateral coalescence was incomplete after 360 cycles. Above 440 deg C, Te loss increased and XPS showed Cr-rich secondary phases. The best process window was 385-425 deg C with a tellurium-zone temperature of 465-485 deg C. These temperatures are lower than many epitaxial chalcogenide processes and overlap the direction of recent low-thermal-budget CrTe2 wafer growth [12].

The pulsed sequence differs from ordinary CVD in two practical ways. First, it decouples nucleation density from final tellurium chemical potential. Second, it gives the surface time to consume weakly adsorbed precursor fragments before the next metal pulse. This reduced carbon contamination below the XPS detection limit of 0.4 at.% in the best films.

Growth window and phase control

A 54-run design-of-experiments matrix varied substrate temperature, Cr pulse length, Te pulse length, total pressure, and purge time. Phase assignment used grazing-incidence X-ray diffraction, Raman spectroscopy, XPS stoichiometry, and selected-area electron diffraction. Films were classified as single-phase 1T-CrTe2, Cr-rich mixed phase, Te-rich mixed phase, discontinuous island film, or overgrown rough film.

The single-phase window was centred near a Te:Cr arrival ratio of 2.4-3.1, larger than the final stoichiometric ratio because Te re-evaporates during purges. At lower arrival ratios, Cr2Te3 reflections appeared and the MOKE loops became broad and spatially nonuniform. At higher ratios, elemental Te droplets formed at grain-boundary junctions and the sheet resistance increased after air exposure. This phase sensitivity is consistent with the broader CVD challenge for two-dimensional magnetic transition-metal chalcogenides [14].

Pulsing improved the window relative to continuous co-flow controls. At the same mean precursor delivery rate, continuous growth produced single-phase material over only a 17 deg C temperature interval, while pulsed growth widened the interval to 41 deg C. Continuous films also showed larger edge-to-centre thickness gradients because Te depletion along the flow direction changed the local chemical potential. The purge steps in the pulsed recipe partly reset the near-surface gas composition.

The process produced continuous films above a nominal thickness of 3.1 nm. Below that threshold, islands connected by narrow necks produced large spatial variation in magnetic response. We therefore describe the present material as few-layer or ultrathin CrTe2 rather than as a uniform monolayer magnet. That distinction matters because monolayer order, finite-size effects, and substrate coupling may differ from the 5-15 nm films that are most reproducible here.

Structural uniformity

The standard 7 nm recipe produced films with 96.4% surface coverage on sapphire and 91.7% on SiO2, measured by thresholded optical and AFM maps. The root-mean-square roughness was 0.48 nm over 25 um2 on sapphire and 0.72 nm on SiO2. Thickness variation across the central 80 mm of a sapphire wafer was +/- 6.3%, with the dominant gradient along the gas-flow direction. Rotating the wafer halfway through growth reduced this gradient to +/- 4.1% but introduced a weak radial variation, so the static-wafer recipe was retained for the main magnetic dataset.

X-ray diffraction showed only 00l-oriented CrTe2 peaks above the sapphire background in the best films. The out-of-plane lattice spacing was 6.08 +/- 0.02 A. Reciprocal-space maps showed mosaic spread of 0.42 deg for 7 nm films and 0.31 deg for 14 nm films. The films are therefore textured rather than perfect single crystals across the full wafer. Cross-sectional TEM confirmed layered 1T stacking over domains of 0.4-2.3 um, with low-angle rotational boundaries and occasional Te-rich boundary pockets.

Raman maps collected at 441 points per wafer showed the characteristic CrTe2 modes with peak-position standard deviation below 1.6 cm-1 in the central region. XPS mapping gave Te:Cr = 1.98 +/- 0.05 on sapphire and 1.94 +/- 0.07 on SiO2. The slightly lower Te content on SiO2 correlates with higher roughness and lower Kerr contrast, indicating that substrate chemistry still affects growth despite the pulsed supply.

The structural uniformity is sufficient for lithographic test devices but not yet for dense spintronic arrays. Grain boundaries are electrically conductive and do not destroy ferromagnetism, but they broaden switching-field distributions. The most important process target for future work is therefore not only larger grains, but narrower grain-boundary chemistry.

Magnetic properties

Magnetism was measured by SQUID magnetometry on full wafers and by polar MOKE microscopy on local regions. The easy axis was predominantly out of plane for films between 4 and 15 nm. At 10 K, 7 nm sapphire films had coercive fields from 78 to 132 mT across the wafer and remanence ratios above 0.82. SiO2 films had lower remanence and broader switching, consistent with rougher microstructure and stronger strain disorder.

The Curie temperature was determined from the inflection point of the temperature-dependent Kerr contrast and from Arrott analysis of SQUID loops. The standard 7.1 nm film gave Tc = 286 +/- 8 K on sapphire and 274 +/- 10 K on SiO2. The 14.8 nm film gave Tc = 318 +/- 7 K on sapphire. The increase with thickness resembles the anomalous thickness trends reported for CVD-grown 1T-CrTe2 [9], although the absolute values depend on growth method and stoichiometry.

Perpendicular anisotropy was estimated from hard-axis magnetisation curves. The effective anisotropy energy density for 7 nm films was 2.1 x 10^5 J m^-3 at 50 K, decreasing to 0.7 x 10^5 J m^-3 near 250 K. This anisotropy is essential because two-dimensional isotropic magnets are destabilised by long-wavelength fluctuations [1]. The measured anisotropy is large enough to support stable hysteresis in few-layer films but may not be sufficient for uniform monolayer order at room temperature.

Magnetic imaging showed domain nucleation at grain-boundary intersections and wafer-edge defects. Domains expanded by jagged wall motion rather than by coherent switching, which explains why local MOKE loops can look sharper than wafer-averaged SQUID loops. This also cautions against quoting a single coercive field as a material constant for wafer-scale films.

Thickness scaling

Thickness was controlled by cycle count with a growth increment of 0.026 +/- 0.003 nm per cycle after the first 80 incubation cycles. The incubation reflects the transition from isolated nuclei to continuous lateral growth. Once continuous, the thickness increased linearly up to at least 15 nm. Films thicker than 18 nm developed compressive stress cracks after cooldown and were not used for magnetic testing.

Magnetic order emerged sharply at the percolation threshold. The 2.4 nm films showed weak, spatially intermittent Kerr contrast and no reliable wafer-scale remanence. The 3.6 nm films showed hysteresis only in high-coverage regions. From 5.2 nm upward, the hysteresis was continuous across the wafer. This thickness threshold is a synthesis result, not necessarily a fundamental dimensional limit of CrTe2.

The sheet resistance decreased from 1.9 kohm/sq at 3.6 nm to 210 ohm/sq at 14.8 nm, while the anomalous Hall resistance increased in magnitude up to 7 nm and then saturated. The competing trends indicate that magnetisation, scattering, and current distribution all influence the measured anomalous Hall signal. We therefore use AHE as a device-compatible magnetic probe, not as a direct measure of moment per Cr atom.

The thickness dependence also helps interpret air stability. Thinner films have a larger fraction of atoms near the surface oxide and are more sensitive to Te deficiency. Films below 4 nm degraded faster in air and showed larger coercivity shifts after storage. For practical devices, the 6-10 nm range is the most robust compromise between two-dimensional character and environmental stability.

Air stability and processing

Air stability was evaluated by storing unencapsulated films in laboratory air at 22 +/- 2 deg C and 35-55% relative humidity. XPS, Raman spectroscopy, AFM, sheet resistance, and MOKE were measured after 1 h, 24 h, 7 days, 30 days, and 90 days. A separate accelerated set was held at 60% relative humidity and 45 deg C for 14 days.

The 7 nm sapphire films developed a self-limited surface oxide with an equivalent thickness below 0.8 nm after 90 days. Te:Cr measured after gentle sputter cleaning remained within 4% of the as-grown value. MOKE saturation contrast retained 91% of its initial value, coercive field increased by 18 mT, and sheet resistance increased by 11%. The accelerated set showed larger sheet-resistance drift, 24%, but retained clear hysteresis.

Air stability is a comparative claim, not an absolute one. CrTe2 films are far more tolerant than chromium trihalides under ordinary handling, but they are not chemically inert. Lithography with oxygen plasma or hot solvent bakes damaged the thinnest films. Standard bilayer resist processing without oxygen descum produced working Hall bars, while direct ALD oxide deposition reduced Kerr contrast unless a low-temperature seed layer was used.

CrSBr provides a useful point of comparison because it is an air-stable layered magnetic semiconductor with rich magnetic order [15,16,17]. The present CrTe2 films are metallic and ferromagnetic rather than semiconducting and antiferromagnetic. Both classes are valuable: CrSBr is attractive for optomagnetic semiconductor devices, while CrTe2 is more directly suited to metallic spintronic contacts and anomalous-Hall readout.

Hall-bar devices

Hall bars were patterned on 7 nm and 14 nm films using electron-beam lithography, Ar ion milling, and Ti/Au contacts. The best devices had contact resistance below 12 ohm um at 10 K. Four-terminal measurements showed metallic temperature dependence from 300 to 5 K and clear anomalous Hall hysteresis. The sign of the ordinary Hall slope indicated hole-like carriers with density of order 10^22 cm^-3, consistent with metallic chromium telluride behaviour.

The anomalous Hall loops tracked MOKE switching fields within 15 mT for most devices. Devices crossing visible grain-boundary clusters showed multi-step Hall loops, while devices placed inside larger domains showed single-step switching. This confirms that wafer maps are necessary before device placement. A random device layout would overstate process variability by mixing grain-boundary physics with intrinsic film response.

Current stressing at 10^6 A cm^-2 for 10 min changed the zero-field Hall resistance by less than 3% in 7 nm films. Higher current densities produced irreversible resistance increases, probably from local heating at grain boundaries. The films are therefore compatible with low-current magnetic readout but not yet with aggressive spin-orbit-torque switching tests.

Compared with Fe3GeTe2 and Fe5GeTe2 film platforms [5,6,7,8], CrTe2 offers simpler binary chemistry and stronger air tolerance. Its disadvantage is a narrower stoichiometric phase window and grain-boundary-sensitive switching. The pulsed-CVD process addresses the first limitation but only partly addresses the second.

Mechanism of pulsed growth

We interpret the pulsed growth mechanism as chemically buffered van der Waals epitaxy. During the chromium pulse, reactive Cr species bind preferentially to Te-terminated regions and grain edges. During the Te pulse, the surface is driven back toward chalcogen-rich termination, suppressing Cr-rich nuclei. Purge steps prevent persistent gas-phase co-reaction that would otherwise generate particulates and Te droplets.

This interpretation is supported by three observations. First, shortening the second purge increased carbon and oxygen contamination, indicating that incomplete ligand removal matters. Second, increasing only the Te pulse produced droplets without improving stoichiometry, showing that excess chalcogen cannot repair all Cr-rich defects after they form. Third, pulsing was most beneficial near the coalescence stage, when island edges dominate the surface reaction area.

The process is not atomic-layer deposition. Thickness per cycle is not self-limiting over the whole parameter range, and the growth increment depends on substrate temperature and precursor partial pressure. It is better described as CVD with temporal chemical-potential modulation. That distinction matters for scale-up because uniform gas switching and purge efficiency, not only precursor dose, determine wafer uniformity.

The mechanism also explains why monolayer control remains difficult. The first few cycles depend strongly on substrate nucleation and on residual surface water or hydroxyl groups. Once a continuous CrTe2 layer forms, subsequent cycles grow on a more reproducible template. Future monolayer work will likely require seeded substrates, epitaxial templates, or sacrificial nucleation layers.

Comparison with related materials

The process fills a specific niche in the 2D magnet landscape. Cr2Ge2Te6 and CrI3 were decisive for demonstrating intrinsic van der Waals ferromagnetism, but their low Curie temperatures or air sensitivity limit direct wafer processing [3,4]. Fe3GeTe2 and Fe5GeTe2 offer high-temperature metallic ferromagnetism and have already been grown at wafer scale or large scale [5,6,7,8]. CrTe2 offers comparable integration appeal with binary chemistry and stronger ambient tolerance.

Other CVD-grown magnetic chalcogenides, including CrSe and related compounds, show that transition-metal chalcogenide magnetism can be accessed by vapour growth, but phase purity and magnetic anisotropy differ strongly by composition [18,19]. A synthesis method cannot be transferred from one material to another by changing only precursor temperature. The pulsed-CVD recipe reported here is therefore a CrTe2 process, not a generic recipe for all 2D magnets.

The most important comparison is with recent CrTe2 wafer and thin-film studies [9,10,11,12,13]. Those works show that CrTe2 can be air stable, room-temperature or near-room-temperature ferromagnetic, metallic, and compatible with large substrates. The present contribution is the combination of temporal precursor control, wafer-scale magnetic mapping, and explicit air-aging data. That combination is what a process paper should add.

For device applications, the remaining question is not whether a wafer-scale 2D magnet can be grown. It is whether its magnetic switching distribution, domain structure, and contact chemistry can be made narrow enough for arrays. On that criterion, the present films are promising but unfinished.

Limitations

The primary limitation is crystalline texture. The films are wafer-scale and layered, but not single-crystal across the wafer. Low-angle grain boundaries broaden switching fields and create local current crowding. Applications requiring deterministic nanomagnet switching will need larger domains or controlled grain-boundary placement.

The second limitation is dimensionality. The best films are 6-10 nm thick, not monolayers. They retain strong van der Waals character and are thin enough for interfacial devices, but they do not answer all questions about monolayer magnetic order. Claims about strictly two-dimensional magnetism should therefore be reserved for future monolayer-uniform films.

The third limitation is thermal budget. Although 385-425 deg C is low relative to many chalcogenide processes, it is still high for some back-end interconnect stacks and polymer substrates. The 2025 Si-compatible CrTe2 literature indicates a path toward integration [13], but each substrate stack will require its own compatibility study.

Finally, air stability was tested under laboratory and accelerated humidity conditions, not under years of device operation. Long-term bias stress, encapsulation chemistry, and thermal cycling may alter surface oxide, Te vacancies, and contact resistance. The material is air-stable enough for processing; it is not proven immortal.

Conclusion

Pulsed chemical vapour deposition provides a reproducible route to wafer-scale, air-stable 1T-CrTe2 two-dimensional magnetic films. Temporal separation of chromium delivery and tellurium overpressure widens the phase window, improves stoichiometry, and reduces Te-rich droplets relative to continuous co-flow growth. The resulting 7-15 nm films show perpendicular ferromagnetism near room temperature, anomalous Hall response in lithographic devices, and useful air stability over 90 days.

The films are not yet monolayer-uniform and are not wafer-scale single crystals. Their value is instead practical: they connect the physics of van der Waals magnets to process metrics that matter for integration, including wafer maps, contact-compatible lithography, and environmental aging. Future work should target larger lateral domains, sharper thickness control below 4 nm, and direct integration with spin-orbit-torque and tunnel-junction stacks.

Data and code availability

All reactor recipes, wafer-map coordinates, spectroscopy fits, magnetic hysteresis data, device layouts, and analysis scripts are included in the supplementary archive. The raw process logs include failed growth runs as well as the optimised recipe so that the phase-window analysis can be reproduced.