Remote Epitaxy: Mechanisms & Innovations

• Remote epitaxy is the growth of crystalline films on substrates through an atomically thin 2D interlayer that transmits the substrate’s lattice template, enabling exfoliation of free-standing membranes.
• It involves a complex interplay of direct bonding, pinhole-seeded growth, and van der Waals interactions where the substrate’s signal competes with graphene screening and defect-assisted nucleation.
• The process is highly influenced by adsorption kinetics, interfacial screening, and defect control, making it a versatile strategy for fabricating reusable semiconductor, oxide, and metallic films.

Searching arXiv for recent and foundational papers on remote epitaxy, including the papers we found on arXiv in the provided data block. Remote epitaxy is the growth of a crystalline film on a substrate covered by an atomically thin two-dimensional interlayer, most commonly graphene, in which the film is oriented to the underlying substrate rather than to the graphene itself. In the standard formulation, the interlayer weakens direct bonding sufficiently to enable exfoliation of single-crystalline membranes and substrate reuse, while the buried substrate still transmits a crystallographic template through the interlayer (Yoon et al., 2022). The subject has developed simultaneously as an overview method and as a mechanistic controversy: many reported outcomes that are macroscopically consistent with remote epitaxy are also consistent with pinhole-seeded lateral epitaxy, selective-area epitaxy, or van der Waals epitaxy, so the central questions concern what the film actually “feels” at nucleation, how graphene modifies that potential landscape, and which structural signatures are specific to a genuinely remote mechanism (LaDuca et al., 29 Jul 2025).

1. Definition and distinction from related growth modes

Remote epitaxy is conventionally distinguished from both direct epitaxy and van der Waals epitaxy. In direct epitaxy, the film bonds directly to the substrate. In van der Waals epitaxy, the film primarily interacts with the 2D surface itself. In remote epitaxy, by contrast, “the film is oriented to the underlying substrate” and is “oriented by the interatomic potential from the substrate penetrating through the van der Waals material” (Yoon et al., 2022). This distinction is central because membrane exfoliation and substrate reuse are not unique to a remote mechanism.

Three mechanistic classes recur throughout the literature.

Mechanism	Primary registry source	Typical consequence
Remote epitaxy	Buried substrate through an intact 2D interlayer	Substrate-aligned growth with weak adhesion
Pinhole-seeded lateral epitaxy	Exposed substrate at defects or openings	Direct nucleation plus lateral overgrowth
Van der Waals epitaxy	Graphene or other 2D surface	Graphene-controlled or weakly constrained registry

A recurring misconception is that exfoliable single-crystalline films, by themselves, establish remote epitaxy. GaSb on graphene-terminated GaSb(001) was shown to grow by pinhole-seeded lateral epitaxy, yet the resulting continuous film could still be exfoliated as a free-standing membrane (Manzo et al., 2021). Patterned graphene masks on Ge(001) likewise showed that GaAs nucleates primarily on exposed Ge windows, with $>99\%$ nucleation selectivity for $T \gtrsim 610^\circ\text{C}$, even when graphene stripe widths or spacings were as large as $10\ \mu\text{m}$; this result set an experimental constraint on claims of growth through continuous graphene (Lim et al., 2021). The distinction is therefore not merely semantic: the same macroscopic endpoint can arise from fundamentally different interfacial physics.

2. Potential transmission, screening, and “graphene transparency”

A central premise of remote epitaxy is that the substrate’s lattice or bonding potential survives transmission through graphene strongly enough to bias nucleation. Recent work has challenged any treatment of graphene as a passive, invisible spacer. A surface-science perspective decomposes the total potential into substrate, graphene, and reconstruction or supercell terms,

$$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$

and argues that the field has often over-interpreted “graphene transparency” without directly quantifying the relative amplitudes of these contributions (LaDuca et al., 29 Jul 2025). In that framework, Fourier and beating analysis was proposed as a bias-free way to separate periodicities associated with graphene, the substrate, and graphene-induced reconstructions.

The quantitative severity of screening is emphasized by both analytical and phenomenological models. A Thomas–Fermi-like estimate with $\lambda_{TF}=1.7\ \text{\AA}$ and effective graphene thickness $\Delta z = 3\ \text{\AA}$ gives a transmitted field of $17\%$ through monolayer graphene and $3\%$ through bilayer graphene (LaDuca et al., 29 Jul 2025). A separate analytical model starts from a Morse bonding potential,

$$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$

adds spacer-induced separation via a Lennard-Jones term and graphene screening via $T_s$, and writes

$T \gtrsim 610^\circ\text{C}$0

That model concludes that $T \gtrsim 610^\circ\text{C}$1 for typical semiconductor and oxide substrates is only a few meV, comparable to graphene’s own van der Waals potential, with representative estimates of $T \gtrsim 610^\circ\text{C}$2 meV for ZnO, $T \gtrsim 610^\circ\text{C}$3 meV for GaN, $T \gtrsim 610^\circ\text{C}$4 meV for GaAs, $T \gtrsim 610^\circ\text{C}$5 meV for SiC, $T \gtrsim 610^\circ\text{C}$6 meV for Si, and $T \gtrsim 610^\circ\text{C}$7 meV for Cu through monolayer graphene (Kawasaki et al., 14 Jul 2025). The immediate implication is not that remote epitaxy is absent, but that the remote term is often energetically comparable to graphene’s own contribution and therefore susceptible to interference, reconstruction, and defect-mediated alternatives.

This competition is formalized most explicitly in GdAuGe on graphene/6H-SiC(0001), where the total interfacial potential was written as

$T \gtrsim 610^\circ\text{C}$8

with

$T \gtrsim 610^\circ\text{C}$9

The key regime was identified as

$10\ \mu\text{m}$0

so that no single registry dominates and long-range order becomes frustrated (Jung et al., 7 Dec 2025). This formulation shifts the subject away from a binary transparent-versus-opaque picture and toward a multicomponent interfacial potential landscape.

3. Kinetics, sticking coefficients, and island mobility

Remote epitaxy is not only a question of static templating. Several studies argue that its observability depends critically on adsorption kinetics, wetting, diffusion, and the ability of islands to move across graphene while retaining some substrate sensitivity.

The most direct practical manifestation is the sticking coefficient $10\ \mu\text{m}$1, defined as the probability that an incoming adatom sticks to the surface. For Ni$10\ \mu\text{m}$2MnGa grown by MBE on monolayer graphene-covered MgO(001), $10\ \mu\text{m}$3 was found to be element- and temperature-dependent rather than close to unity. On graphene/MgO at $10\ \mu\text{m}$4C by IBS and $10\ \mu\text{m}$5C by EDS, Ga accumulated almost as well as on bare MgO, whereas Ni and Mn showed only about $10\ \mu\text{m}$6–$10\ \mu\text{m}$7 of the areal density found on the MgO side. By initiating growth below $10\ \mu\text{m}$8C, where sticking coefficients were closer to unity and wetting improved, the authors obtained epitaxial Ni$10\ \mu\text{m}$9MnGa films with controlled stoichiometry and no impurity phases (LaDuca et al., 2023). This result directly linked graphene-modified adsorption kinetics to stoichiometric control.

Selective-area experiments on GaAs further underscored the kinetic role of graphene. Patterned graphene on Ge(001) behaved as a low-sticking, high-diffusion mask. As growth temperature increased from $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$0 to $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$1C, GaAs coverage on graphene dropped by about $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$2, while coverage on Ge dropped by less than an order of magnitude. Periodic supply epitaxy, in which Ga flux was pulsed and As remained on during annealing intervals, further enhanced selectivity by allowing desorption from graphene and diffusion toward exposed Ge (Lim et al., 2021). In that setting, diffusion length rather than remote substrate transmission determined where nucleation occurred.

A first-principles study generalized this kinetic perspective by arguing that electrostatic potential and single-atom adsorption are insufficient criteria for remote-epitaxy viability. Instead it proposed the sliding barrier of a relaxed film island as the most reliable metric. With island bonding energy

$$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$3

the sliding barrier is

$$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$4

Normalized by exposed island area, a value near $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$5 emerged as an empirical separator between systems in which remote epitaxy had and had not been observed (Campbell et al., 11 Mar 2026). This supports a kinetic picture in which remote epitaxy occupies an intermediate regime: islands must remain mobile enough to relax and align, but not so mobile that substrate-guided registry is lost.

4. Mechanistic controversy: pinholes, thru-holes, frustration, and rotated domains

The strongest criticisms of remote epitaxy come from experiments showing that direct nucleation at sparse openings can reproduce its canonical outputs. In GaSb/graphene/GaSb(001), in-situ XPS showed native oxide removal during pre-growth annealing, while AFM and Raman revealed defect generation in graphene: $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$6 nm pinholes appeared at about $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$7C, and larger $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$8 nm holes at $$\phi_{total} = \phi_{sub} + \phi_{gr} + \phi_{supercell},$$9C. During subsequent MBE growth, GaSb nucleated selectively in those holes, grew laterally over graphene, and coalesced into a continuous film by about $\lambda_{TF}=1.7\ \text{\AA}$0 ML (Manzo et al., 2021). The same work estimated a pinhole spacing of about $\lambda_{TF}=1.7\ \text{\AA}$1 nm and a diffusion length on the order of $\lambda_{TF}=1.7\ \text{\AA}$2, so that $\lambda_{TF}=1.7\ \text{\AA}$3, favoring defect-seeded growth. It also compared a calculated remote potential modulation of $\lambda_{TF}=1.7\ \text{\AA}$4 meV with $\lambda_{TF}=1.7\ \text{\AA}$5 meV at growth temperature, arguing that direct bonding at holes, estimated at $\lambda_{TF}=1.7\ \text{\AA}$6 eV, is the more plausible driver.

The critique was extended by “thru-hole epitaxy,” which showed aligned GaN growth not only across transferred h-BN, but also across thick, polycrystalline, symmetrically incompatible h-BN and even across a $\lambda_{TF}=1.7\ \text{\AA}$7 nm amorphous SiO$\lambda_{TF}=1.7\ \text{\AA}$8 interlayer, provided nanoscale openings connected the film to sapphire (Jang et al., 2021). In that framework, alignment plus easy detachment does not imply remoteness; it only implies sparse direct connectedness plus epitaxial lateral overgrowth.

Against these critiques, several studies have proposed structural signatures that are difficult to reconcile with pinhole-seeded or simple serial mechanisms. On clean graphene/sapphire, GdPtSb showed an in-plane $\lambda_{TF}=1.7\ \text{\AA}$9 rotated epitaxial superstructure, denoted R30, in addition to the direct-epitaxy R0 state. Because pinhole-seeded growth should reproduce the direct sapphire orientation and van der Waals epitaxy on polycrystalline graphene should not generate a substrate-registered $\Delta z = 3\ \text{\AA}$0 superstructure, R30 was proposed as a possible experimental fingerprint of remote coupling (Du et al., 2022). The R30 volume fraction increased as growth temperature decreased from $\Delta z = 3\ \text{\AA}$1C to $\Delta z = 3\ \text{\AA}$2C, a trend interpreted as reduced diffusion favoring remote rather than pinhole epitaxy.

The most explicit structural argument for a nontrivial remote mechanism appears in GdAuGe on graphene/SiC. Two signatures were reported that were stated to be incompatible with the leading alternatives of pinhole-seeded epitaxy and simple serial or direct epitaxy: a few-atomic-layer-thick disordered interlayer at the GdAuGe/graphene interface, and a $\Delta z = 3\ \text{\AA}$3 rotated epitaxial relationship between GdAuGe and SiC on buffer graphene (Jung et al., 7 Dec 2025). STEM showed that the first $\Delta z = 3\ \text{\AA}$4–$\Delta z = 3\ \text{\AA}$5 atomic layers of GdAuGe were highly disordered on buffer graphene and epitaxial graphene, whereas this disorder was absent on H-intercalated graphene. DFT indicated that buffer and epitaxial graphene/SiC generate multiple comparable Fourier components, including $\Delta z = 3\ \text{\AA}$6, $\Delta z = 3\ \text{\AA}$7, $\Delta z = 3\ \text{\AA}$8, and a $\Delta z = 3\ \text{\AA}$9 pseudo-periodicity $17\%$0. For buffer graphene, the distorted freestanding graphene contribution at $17\%$1 was only about $17\%$2 meV, whereas the full buffer graphene/SiC slab had a reconstruction-related amplitude of about $17\%$3 meV, leading to the conclusion that the remotely screened substrate distortion dominates the reconstruction term. In that interpretation, the R30 state and disordered interlayer are manifestations of remote epitaxial frustration rather than failures of epitaxy.

5. Growth methodologies and materials platforms

A major development has been the adaptation of growth techniques to preserve the 2D interlayer while maintaining epitaxial quality. For oxides, hybrid MBE was used to grow SrTiO$17\%$4 on graphene without an independent oxygen source. Instead of RF plasma, ECR plasma, ozone, or molecular oxygen, the process used Sr from thermal sublimation and Ti plus oxygen from titanium tetraisopropoxide, with growth at $17\%$5C on $17\%$6 SrTiO$17\%$7(001) or LSAT(001). The four oxygen atoms in each TTIP molecule provided sufficient oxygen to obtain phase-pure SrTiO$17\%$8, thereby avoiding graphene damage. The films showed RHEED oscillations, atomically smooth surfaces, an epitaxial relationship of $17\%$9, and could be exfoliated and transferred while leaving the graphene on the original substrate (Yoon et al., 2022).

A different route replaced transferred graphene with directly formed 2D interlayers inside the epitaxy system. In “advanced remote epitaxy,” h-BN was grown directly on GaN(0001) by MBE at $3\%$0C, while graphene was grown directly on GaAs(001) in MOCVD at around $3\%$1C using toluene, with an AlGaAs buffer for thermal stabilization. The directly grown 2D layers were often amorphous or nanocrystalline but remained predominantly $3\%$2-bonded. This enabled alternating semiconductor/2D/semiconductor/2D/semiconductor stacks in a single growth campaign, followed by layer-by-layer peeling. The work demonstrated three-stack GaN/h-BN and three-stack (Al)GaAs/graphene structures, exfoliation of full 2-inch GaN and GaAs membranes, and reuse of the same GaAs wafer three times, with post-exfoliation roughness values of $3\%$3, $3\%$4, and $3\%$5 nm (Kim et al., 2022). This established remote epitaxy as a manufacturing scheme rather than only a one-film release method.

The interlayer itself has also been generalized beyond graphene and h-BN. Low-temperature PECVD produced ultrathin amorphous carbon layers on III–V semiconductors at room temperature followed by a mild $3\%$6C UHV anneal. These films had RMS roughness $3\%$7 nm, predominantly $3\%$8-hybridized bonding, and monolayer-like thickness near $3\%$9 nm. Carbon thickness tuned the substrate–film interaction continuously: monolayer-like a-C yielded the best results, while thicker a-C screened the substrate more strongly and degraded morphology and dislocation density. Under optimized conditions, the method produced single-crystalline, (001)-oriented GaAs, In$$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$0Ga$$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$1As, cubic-AlN, and cubic-GaN on carbon-coated GaAs, InP, and 3C-SiC, with dislocation densities below $$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$2 and successful lift-off using a Ti/Ni stressor (Henksmeier et al., 2024). This broadened remote epitaxy from a transferred-graphene methodology into a more general thin-$$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$3-interlayer strategy.

6. Oxides, freestanding membranes, and functional responses

Oxide remote epitaxy has been especially demanding because conventional oxide growth conditions damage graphene. A PLD study of BaTiO$$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$4 on graphene/SrTiO$$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$5 identified a direct correlation between the evolving graphene microstructure, plume-induced defect formation, and BTO crystalline quality. A controlled aperture method reduced the kinetic and ionic severity of the plume during the first growth stage. Large-grain graphene with grain size greater than $$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$6 suffered less damage than $$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$7 graphene, and BTO on the large-grain case reached a rocking-curve half width of $$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$8, compared with $$\phi_{m}(z) = D \left[ e^{-2a(z-z_0)} - 2e^{-a(z-z_0)} \right],$$9 for BTO directly on STO. Bilayer graphene provided the most effective compromise between epitaxy and release: monolayer graphene preserved epitaxy but exfoliation was incomplete, while trilayer graphene yielded polycrystalline BTO. Using large-grain bilayer graphene, $T_s$0 oxide layers were exfoliated and transferred onto SiO$T_s$1-Si, and PFM showed a $T_s$2 phase contrast after poling (Haque et al., 2024).

A human-AI collaborative autonomous PLD workflow addressed the same BTO/graphene problem from the standpoint of synthesis optimization. In a six-phase campaign comprising $T_s$3 films and no repeated growth conditions, in-situ Raman spectroscopy, laser reflectivity, and an ion probe were coupled to Bayesian optimization. The study found that graphene preservation requires low oxygen pressure and low substrate temperature, whereas BTO crystallization on STO improves only above about $T_s$4–$T_s$5C. Ion-probe data showed a fast plume component of about $T_s$6 eV/Ba atom in vacuum, a slow component around $T_s$7 eV/Ba atom at $T_s$8 mTorr in Ar or O$T_s$9, and an extreme leading edge near $T \gtrsim 610^\circ\text{C}$00 eV/Ba atom; with a graphene displacement threshold of roughly $T \gtrsim 610^\circ\text{C}$01 eV/C-atom, ballistic damage seeded defects that oxygen chemistry then amplified at elevated temperature. The study concluded that a two-step Ar/O$T \gtrsim 610^\circ\text{C}$02 deposition is required to exfoliate ferroelectric BaTiO$T \gtrsim 610^\circ\text{C}$03 while maintaining a monolayer graphene interlayer (Haque et al., 14 Nov 2025).

Remote epitaxy has also been important as a route to functional freestanding metallic and magnetic membranes. Ni$T \gtrsim 610^\circ\text{C}$04MnGa grown on graphene/MgO(001) could be exfoliated as a freestanding membrane and then rippled by transfer to pre-strained polyurethane. In SQUID measurements at $T \gtrsim 610^\circ\text{C}$05 K with in-plane field, the coercive field increased from about $T \gtrsim 610^\circ\text{C}$06 Oe in the relaxed film to about $T \gtrsim 610^\circ\text{C}$07 Oe in the strained membrane, with ripple period about $T \gtrsim 610^\circ\text{C}$08, peak-to-peak height about $T \gtrsim 610^\circ\text{C}$09, and estimated peak strain magnitudes $T \gtrsim 610^\circ\text{C}$10 (LaDuca et al., 2023). Such results link remote epitaxy directly to strain engineering rather than only to membrane release.

7. Conceptual boundaries and adjacent membrane-based strategies

The modern literature increasingly treats remote epitaxy as one member of a broader family of membrane-enabled and 2D-interlayer growth strategies, not as a universal description of all epitaxy on graphene. Cold seeded epitaxy of GdAuGe on graphene/Ge(111) is a clear example. There, a $T \gtrsim 610^\circ\text{C}$11 nm seed was deposited at room temperature, annealed at $T \gtrsim 610^\circ\text{C}$12C, and then used as the template for continued growth to about $T \gtrsim 610^\circ\text{C}$13–$T \gtrsim 610^\circ\text{C}$14 nm. The resulting films had rocking-curve widths of $T \gtrsim 610^\circ\text{C}$15 arc sec on graphene/Ge and $T \gtrsim 610^\circ\text{C}$16 arc sec on graphene/SiC, with atomically sharp interfaces and continuous exfoliated membranes. However, the authors interpreted the two $T \gtrsim 610^\circ\text{C}$17-rotated in-plane orientations as matching the two graphene domain orientations and therefore as evidence that the film was epitaxial to graphene rather than to the underlying Ge. They explicitly placed the work closer to seed-assisted van der Waals or graphene-mediated epitaxy than to strict remote epitaxy (LaDuca et al., 2024). This provides an important boundary condition: graphene can enable membrane growth without the buried substrate being the dominant template.

Other adjacent concepts are even more clearly distinct. Double-sided van der Waals epitaxy of Bi$T \gtrsim 610^\circ\text{C}$18Se$T \gtrsim 610^\circ\text{C}$19 and Sb$T \gtrsim 610^\circ\text{C}$20Te$T \gtrsim 610^\circ\text{C}$21 on both sides of suspended graphene or hBN does not rely on a remote 3D substrate field at all; the atomically thin membrane itself is the crystalline scaffold, and in the hBN case also serves as a crystal-momentum-conserving tunnel barrier (Park et al., 2024). Membrane-based oxide interface engineering beyond epitaxy, such as the transfer of a 30-nm SrTiO$T \gtrsim 610^\circ\text{C}$22 membrane onto Al-terminated sapphire followed by annealing at $T \gtrsim 610^\circ\text{C}$23C in $T \gtrsim 610^\circ\text{C}$24 mbar O$T \gtrsim 610^\circ\text{C}$25, creates atomically clean but symmetry-forbidden interfaces with moiré-type reconstruction; this strategy is explicitly not remote epitaxy because the interface is formed by transfer and bonding rather than by substrate-guided growth through a 2D layer (Wang et al., 2024).

These neighboring approaches clarify the scope of remote epitaxy. In its strict sense, remote epitaxy requires a film whose crystallographic registry is set by the buried substrate through an intact, weakly bonding interlayer. The current literature suggests that this condition is real in some systems, but often only under narrow combinations of interfacial cleanliness, interlayer thickness, screening, lattice mismatch, reconstruction amplitude, sticking kinetics, and defect density (Jung et al., 7 Dec 2025). A plausible implication is that the field has moved from asking whether remote epitaxy exists in principle to asking under what precisely defined interfacial and kinetic conditions it can be separated from pinhole epitaxy, van der Waals epitaxy, and reconstruction-mediated templating.