Remote Epitaxy of BaTiO₃/Graphene

• The paper demonstrates that remote epitaxy of BaTiO₃/graphene harnesses substrate potentials transmitted through graphene to achieve epitaxial registry while enabling membrane exfoliation.
• Methodological advances include a two-step PLD process using inert and oxidizing atmospheres to mitigate both ballistic and oxidative damage to graphene.
• The study shows that optimizing graphene microstructure and transfer protocols yields high-quality, freestanding BaTiO₃ membranes with robust ferroelectric properties for device integration.

Remote epitaxy via two-dimensional (2D) interlayers such as graphene enables the synthesis of single-crystal oxide membranes that inherit the crystallographic registry of their substrates, yet can be mechanically exfoliated and transferred to diverse platforms. For the BaTiO$_3$/graphene system, fundamental advances center on the interplay between oxide nucleation, graphene integrity under pulsed laser deposition (PLD), and scalable transfer processes. This synthesis provides an avenue to combine functional perovskite oxides with silicon technologies or flexible electronics, contingent on controlling atomic-scale phenomena during thin-film growth (Haque et al., 15 Aug 2024, Haque et al., 14 Nov 2025).

1. Principles of Remote Epitaxy Through Graphene

Remote epitaxy exploits the partial transmission of long-range electrostatic or ionic substrate potentials through an atomically thin interlayer—such as monolayer graphene (thickness $t \approx 3.4$ Å). The substrate’s potential, typically characterized by $\Phi(z)\propto \Phi_0 e^{-kz}$ where $z$ is vertical distance and $k \approx 2\pi/a_\mathrm{sub}$ is determined by substrate lattice constant $a_\mathrm{sub}$, tailors adatom adsorption and diffusion atop the 2D sheet. In BaTiO$_3$/graphene/SrTiO$_3$ heterostructures, this potential tail exceeds typical thermal energies during film growth ($T \sim 700\,^\circ$C), establishing epitaxial registry through the graphene, even while van der Waals binding is weak. The net adatom–substrate interaction energy is modeled as $$E_\mathrm{int} = \int\rho_\mathrm{ad}(r)\Phi(r)\,d^3r$$, showing that the templating effect persists despite the nominal chemical inertness of graphene (Haque et al., 15 Aug 2024).

Misfit strain, parametrized by $$\epsilon = (a_\mathrm{film} - a_\mathrm{sub})/a_\mathrm{sub}$$ (with $\epsilon \approx +1.4\%$ for BaTiO$_3$/SrTiO$_3$), is only partially relieved in the presence of graphene due to reduced film–substrate binding energy. This translates into unique strain states both before and after exfoliation, directly impacting functional properties.

2. Pulsed Laser Deposition Schemes and Damage Mitigation

PLD, the primary method for BaTiO$_3$ remote epitaxy on graphene, involves ablation with a 248 nm KrF excimer laser. Key parameters include laser fluence ($F = 1.04\rm\,J\,cm^{-2}$ to $2.2\rm\,J\,cm^{-2}$), repetition rate ($1$–$5\rm\,Hz$), and chamber background (O$_2$ or Ar). The kinetic energy distribution of ionic plume species is bimodal: “fast” Ba-ion components ($E_\mathrm{fast} \approx 110\,\rm eV/Ba$) can directly induce ballistic defects, while “slow” scattered components ($E_\mathrm{slow} \approx 6.6\,\rm eV/Ba$ at $60\,\rm mTorr$) are chemically reactive but less damaging (Haque et al., 14 Nov 2025).

Preserving graphene requires minimizing both thermal oxidation and ballistic impact. Autonomy-driven parameter mapping demonstrates that monolayer graphene survives at $T_\mathrm{sub} \lesssim 400^\circ$C and $p_\mathrm{O_2} \lesssim 100\,\rm mTorr$, but these conditions yield poorly crystallized BaTiO$_3$. Consequently, optimized workflows employ two-step growth:

• Nucleation in inert Ar ($p_\mathrm{Ar} = 60\,\rm mTorr$, $T_\mathrm{sub} = 700^\circ$C, $F \approx 2\,\rm J\,cm^{-2}$), generating $\sim$5–10 nm of c-axis BTO nuclei.
• Stoichiometric BTO growth in O$_2$ ($p_\mathrm{O_2} = 60\,\rm mTorr$), completing the remaining 40–50 nm for ferroelectricity, after much of the graphene is shielded.

Damage is further reduced by laser aperture control, restricting the ablation spot during nucleation, which decreases defect metrics quantified by Raman D:G intensity ratio ($I_\mathrm{D}/I_\mathrm{G}$) and D-peak full width at half maximum ($\Gamma_\mathrm{D}$). Typical metrics with damage-mitigating slit: $I_\mathrm{D}/I_\mathrm{G} \sim 1.3 \pm 0.1$, $\Gamma_\mathrm{D} \approx 24\,\rm cm^{-1}$; without slit, values are $1.6 \pm 0.07$, $68\,\rm cm^{-1}$ (Haque et al., 15 Aug 2024).

3. Microstructure of Graphene and Epitaxial Film Quality

Graphene microstructure, specifically grain size and density of grain boundaries ($\rho_\mathrm{GB}$), directly governs local damage accumulation and resultant BaTiO$_3$ crystallinity. Wet PMMA transfer yields initially defect-free graphene (Raman $I_{2D}/I_\mathrm{G}\sim2$; RMS roughness $<0.3\rm\,nm$ post-anneal). Two regimes are observed:

• Small-grained graphene ($\langle d \rangle \approx 6\,\rm \mu m$, $$\rho_\mathrm{GB} \sim 3.1 \times 10^{10}\rm\,cm^{-2}$$) accumulates more damage and leads to BaTiO$_3$ X-ray rocking curve (RC) FWHM $0.90^\circ$.
• Large-grained graphene ($\langle d \rangle \approx 322\,\rm \mu m$, $\rho_\mathrm{GB} \sim 1.2\times 10^{8}\rm\,cm^{-2}$) yields $I_\mathrm{D}/I_\mathrm{G} \sim 0.8 \pm 0.1$ and RC FWHM $0.61^\circ$, approaching substrate-direct epitaxy ($0.48^\circ$). Grain boundaries act as “soft spots,” visible as defect peaks in Raman line scans every $\sim$300 μm.

The correlation between grain-boundary density and film RC width can be modeled semi-empirically: $$\mathrm{FWHM} \propto \mathrm{FWHM}_0 + \alpha \rho_\mathrm{GB}$$; $\alpha$ represents a damage-sensitivity coefficient (Haque et al., 15 Aug 2024).

4. Exfoliation and Transfer of Freestanding BaTiO$_3$ Membranes

The exfoliation protocol involves sequential Ni stressor deposition (100 nm e-beam evaporation, 1 μm sputtering) atop BaTiO$_3$/graphene/SrTiO$_3$, followed by thermal release tape (TRT) application and heating ($\sim$90–110 °C). Exfoliation is driven by the difference in work of adhesion ($W_\mathrm{ad}$) between interfaces:

• $$W_\mathrm{ad} = \gamma_\mathrm{BTO} + \gamma_\mathrm{graphene-STO} - \gamma_\mathrm{interface}$$
• $W_\mathrm{Ni-BTO} \gg W_\mathrm{ad}$

Resultant BTO/Ni/TRT stacks (up to $4\rm\,mm \times 5\,mm$) can be transferred onto SiO$_2$/Si substrates, retaining single-crystalline BTO as validated by cross-sectional HRTEM, STEM-EDS, and preservation of lattice fringes for Ba, Ti, O, and Ni (Haque et al., 15 Aug 2024).

Strain state evolves from initial compressive in-plane ($\epsilon_\mathrm{IP}$) in BTO/STO ($\epsilon_\mathrm{IP} = -0.4\%$; $c/a = 1.011$), toward near-relaxation post-exfoliation ($\epsilon_\mathrm{IP} \approx 0$, $c/a = 1.004$). Bilayer graphene further weakens substrate bonding, fostering membrane release and expanded unit cell volume ($66.57\,\rm \AA^3$ after exfoliation).

5. Functional Characterization and Integration Potential

Ferroelectricity is retained in freestanding BTO, as shown by piezoresponse force microscopy (PFM) domain writing: clear 180$^\circ$ phase contrast between $+7$ V and $-7$ V poled regions, amplitude variation indicative of ferroelectric switching. SS-PFM measurements display hysteresis below Curie temperature ($T_C \sim 120^\circ$C) with coercive bias $V_c \approx \pm 1.5\,$V; loops collapse above $T_C$, confirming a ferroelectric transition (Haque et al., 14 Nov 2025).

Obtaining large-area, single-crystal, freestanding oxide membranes via remote epitaxy and PLD (at $T \leq 750^\circ$C) enables direct integration with CMOS or flexible platforms. Functional properties—high dielectric constant, piezo- and ferroelectric response—combine with mechanical flexibility and reusable substrates, supporting scalable manufacturing for oxide memories, sensors, and MEMS.

6. Mechanistic Insights, Process Optimization, and AI Workflow

Graphene’s damage during PLD arises from combined ballistic (plume-induced) and oxidative (chemical etching) mechanisms. The critical ballistic threshold is $E_d \approx 22\,$eV per C-atom, exceeded by “fast” plume species but not “slow” ones. Raman spectroscopy tracks damage evolution ($I_\mathrm{D}/I_\mathrm{G}$ increases exponentially with $T_\mathrm{sub}$ or $p_\mathrm{O_2}$), and MD modeling describes oxidative degradation as $$C_\mathrm{loss}(T) \propto \exp[-E_\mathrm{ox}/(k_BT)]$$ with $E_\mathrm{ox} \approx 0.8$–$1.2\rm\,eV$.

AI-human collaborative workflows (HAIC), integrating hypothesis generation, policy updates, and active-learning driven PLD experiments, accelerate mapping of growth conditions, mitigation strategies, and mechanistic understanding. For BaTiO$_3$/graphene, HAIC facilitated identification of growth windows, refined Raman-based defect metrics, and generalized the two-step Ar/O$_2$ process to other oxide/2D material systems (Haque et al., 14 Nov 2025). A plausible implication is that this paradigm applies broadly, provided in situ diagnostics and adaptive protocols are employed.

7. Generalization to Other Oxide/2D Material Systems

The interplay of high-energy plume impact and reactive oxidation applies to a wide spectrum of oxygen-rich oxide PLD systems (e.g., SrTiO$_3$, VO$_2$, HfO$_2$). The two-step inert/oxidizing-atmosphere PLD protocol is directly transferable to systems where stoichiometric oxide growth and 2D interlayer preservation are mutually antagonistic. The mechanistic insights from BaTiO$_3$/graphene remote epitaxy thus extend to a broad class of functional heterostructures, contingent on appropriately balancing process parameters and exploiting AI-driven optimization strategies (Haque et al., 14 Nov 2025).