Sacrificial polyvinyl alcohol substrates to transfer atomic layer deposition grown dielectric thin films Authors: Gillian P Boyce, Alex T Hall, Nicholas A Blumenshein, John Cumings, Raymond J Phaneuf, Aubrey T Hanbicki, Adam L Friedman Published: 2026 ABSTRACT AbstractIn this work, we demonstrate the exfoliation and transfer of metal oxide thin films from sacrificial polyvinyl alcohol (PVA) substrates to target surfaces. The films were deposited directly onto PVA using a standard amorphous AlO (alumina) atomic layer deposition (ALD) process and subsequently transferred to SiO or Au. The resulting transferred films are 32 nm-thick and continuous over millimeter-scale lateral dimensions. They exhibit angstrom-scale surface roughness and are structurally robust, remaining physically intact after multiple exfoliations. Our results address not only the transferred films but also the sacrificial PVA substrates–specifically their interaction with chemical vapor deposition processing at elevated temperatures. Most practically relevant, thermal analysis of the PVA reveals a well-defined melting temperature and no observable glass transition, motivating a recommended maximum processing temperature of 226 C. Additional thermal factors affecting ALD growth and final film quality, such as the mismatch in thermal coefficient of expansion between PVA and the metal oxide layer, are also discussed. Finally, at the interface between the transferred films and target substrates, we identify residual PVA-derived carbon contamination despite extensive O plasma processing. This finding is consistent with the observed degradation of the transferred films’ dielectric properties, likely due to conductive pathways formed by residual carbon. Addressing this carbon contamination is therefore a critical direction for future work.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-55074-y.Subject terms: Materials science, Nanoscience and technology, Physics Introduction As devices move beyond Si with the goal of creating nanoscale gate lengths and channel thicknesses, integrating atomically thin 2D materials for device channels has emerged as a promising approach1–3. For conventional 3D field effect transistors (FETs), atomic layer deposition (ALD) is widely chosen for depositing metal oxide gate dielectrics. ALD relies upon reactive chemical terminations, or “dangling bonds” at the substrate surface to act as nucleation sites. These reactive sites are critical for initiating the self-limiting chemical reactions between the substrate and the sequentially introduced precursor gases in ALD. It is the self-limiting nature of these reactions that makes ALD highly favorable for growing precise and conformal films, resulting in high quality devices with sharp interfaces4. However, ALD-produced gate dielectrics on 2D semiconductor materials such as MoS, MoSe, and WSe remains a major challenge5. The difficulty in ALD processing stems from the inert nature of many 2D material surfaces, which lack dangling bonds6. The chemical nature of layered materials that enables 2D processing–strong covalent in-plane bonds and weak Van der Waals out-of-plane bonds–presents a major challenge for ALD due to the lack of reactive surface sites to initiate growth. As a result, ALD cannot proceed on most canonical 2D material surfaces and can only nucleate on select defect sites or grain boundaries where islanding can occur and potentially coalesce with thicker films7. Recent research has addressed the challenge of integrating dielectrics onto inert surfaces by creating reactive sites to encourage ALD nucleation and growth8. For example, plasma pretreatments and plasma enhanced ALD are effective at creating higher densities of nucleation sites9,10. However, using plasma modification degrades the interface between the 2D material and the dielectric layer because of the damage induced by the plasma to create the available dangling bonds for ALD11. Other surface modification strategies such as UV-ozone pretreatments12,13 and functionalized interlayer insertion14 are also being explored to increase nucleation sites for ALD with varying levels of success. Given the incompatibility of ALD-grown dielectric layers and 2D materials, it is necessary to explore novel integration methods that maintain favorable properties such as bandgap and dielectric constant while minimizing interfacial disorders and defects. From a fabrication and processing perspective, the interfaces in a device can pose the greatest challenge for minimizing impurities, strain, and other defects, but are crucial for achieving good device performance. Interfacial defects can act as scattering centers, directly impacting charge carrier transport in 2D semiconductors. Interfacial disorders, originating from charge impurities, adsorbates, and local strain, can further induce mid-gap states and trap states15. In addition, poor dielectric integration, such as the presence of pinholes, cracks, or other defects, can lead to gate leakage and other performance issues16. One film integration method that is promising for circumventing interfacial issues involves growing an ALD dielectric onto a sacrificial layer that is then used to transfer the dielectric onto an inert surface via Van der Waals stacking. The sacrificial layer acts as a favorable growth substrate and is removed from the dielectric film after growth and before the dielectric is stacked onto a target 2D material. As demonstrated by Lu et al., this technique was reportedly successful with a sacrificial polyvinyl alcohol (PVA) layer used to grow an ALD AlO (alumina) gate dielectric for a MoS FET17. It was reported that this process allowed wafer-scale transfer of alumina and HfO (hafnia) films as thin as 4.5 nm and 2.6 nm, respectively, without the films cracking, and while maintaining good interface quality17. Furthermore, the electrical character of the transferred films was remarkably good, demonstrating no major holes, stresses, or residual carbon contamination that would degrade the films’ insulating or structural character. While the results from this method are promising, the interactions between the sacrificial polymer layer and the ALD alumina film during and after growth remain to be fully explored17. Specifically, a more detailed review of the underlying chemical and thermal processes is still needed. It is known that ALD alumina growth is easily accomplished on some polymers; the precursor trimethyl aluminum (TMA) is a strong Lewis acid, it reacts with the most Lewis basic sites in a polymer, which tend to be oxygen lone-pairs18. In addition, TMA is a strong Lewis acid catalyst, so when TMA is bonded in an adduct state it increases reactivity for more TMA reactions with nearby organics. In particular, for ALD alumina grown on PVA it was found that TMA reacts with available OH as well as carbonyl groups present on un-hydrolyzed acetate groups on the polymer chain19. However, polymer substrates are highly permeable to precursor gases, and the mechanisms driving non-ideal ALD growth on soft materials have been the focus of numerous studies20. During initial ALD cycles, TMA can diffuse into the subsurface of PVA due to its reaction with existing carbonyl groups in the polymer, causing subsurface nucleation and growth. However, it has been demonstrated that PVA has a relatively low depth of penetration for ALD alumina subsurface growth in comparison to other polymers (polyamide, polyester, polycarbonate, etc.)20. The low penetration is likely due to the relatively high density of hydroxyl groups at or near the PVA surface, which react readily with the TMA. Through reaction of the TMA with surface hydroxyl groups, O–Al–(CH) units form at or just below the PVA surface, and further TMA diffusion into the bulk is hindered. Given enough growth cycles to nucleate on all surface and near-surface sites, the alumina film can eventually grow and coalesce on the PVA surface. The chemical structure of partially hydrolyzed and fully hydrolyzed PVA and a comparison of surface versus subsurface polymer ALD alumina growth is illustrated in Fig. 1. Chemical structure of PVA that is partially hydrolyzed (few surface OH terminations) and fully hydrolyzed (abundant surface OH terminations). The schematic illustrates how in the first steps of the ALD AlO cycle, TMA reacts with OH groups. For polymer substrates with abundant surface hydroxyl terminations, as in fully hydrolyzed PVA, a contiguous layer of chemisobed TMA is quickly established, prohibiting TMA diffusion into the subsurface. In contrast, for polymer substrates without abundant surface hydroxyl terminations, as in partially hydrolyzed PVA, the TMA molecules can more readily diffuse into the subsurface. In our work, sacrificial PVA layers are investigated for having the potential to transfer ALD-grown dielectric films to target substrates. While millimeter-scale films that are smooth and structurally robust can be successfully transferred, further characterization suggests completely removing the sacrificial PVA layer from the ALD-grown films and achieving a clean interface between the film and the target substrate remains a challenge. The incomplete removal of the PVA layer can be a major problem for device performance, manifesting itself as excessive leakage current in fabricated devices. Results and discussion The processing of the film stack is shown schematically in Fig. 2 and full details are given in the Methods section. Briefly, a cleaned Si wafer is treated with sequential spin coating and baking of poly(methyl methacrylate) (PMMA) and a sacrificial PVA layer. On top of the PVA layer, an ALD alumina film is grown at 150 C. A thick PMMA support layer is added to the top, and the stack is exfoliated using thermal release tape. The exfoliated layers are exposed to an O plasma to remove PVA, dry-laminated onto a target substrate, and the tape is released by heating. The PMMA is removed with acetone and the final film is exposed to an O plasma to remove PMMA residue. Schematic of ALD AlO transfer process with a PVA sacrificial substrate. A cleaned Si wafer is sequentially spin-coated and baked with PMMA and a sacrificial PVA layer, followed by ALD growth of AlO at 150 C. A thick PMMA support layer is added, then the stack is exfoliated using thermal-release tape. O-plasma is used to remove the PVA, and the stack is dry-laminated onto a target substrate. The tape is released by heating, the PMMA is removed with acetone, and the transferred film is finished with an O plasma exposure to remove PMMA residue. Evidence of successful film growth and transfer is shown in the light-optical and scanning electron microscopy (SEM) images in Fig. 3a and b. Here, millimeter lateral-scale flakes of the exfoliated alumina film were dry laminated onto a target surface of gold-coated silicon. While millimeter-scale flakes are still advantageous for device processing throughput, it should be noted that wafer scale transfer (cm-scale) of the film as reported by Lu et al. was not able to be replicated following the reported processing information17. The energy dispersive spectroscopy (EDS) maps of aluminum and oxygen distribution verify that the transferred flakes are aluminum oxide. Further, the uniform distribution of the carbon signal across the chip indicates that the PVA layer was removed to a level at least below the detection limit (∼0.1 wt%21) for EDS. Materials characterization of the transferred AlO films. (a) Light-optical micrograph of AlO film transferred to gold-coated silicon. (b) SEM micrograph and corresponding EDS elemental maps of aluminum, oxygen, and carbon of the AlO film transferred to gold-coated silicon. (c) Raman spectra of AlO film transferred to gold-coated silicon showing distinct graphitic carbon peaks are present. (d) AFM area scans and height profiles of AlO transferred to silicon with a native oxide. The films show a ∼32 nm thickness and 0.326 ± 0.020 nm. (e) TEM micrograph of a standalone AlO film (exfoliated from gold-coated silicon) in a plan view. Evaluation of the topography and roughness of the transferred films is also crucial. For a transfer to be considered successful, the films should be smooth with minimal tears. To examine these aspects, a section of an exfoliated alumina film was transferred to a target Si substrate with a native oxide. The use of bare SiO, unlike the gold-coated silicon substrate used in the optical and scanning electron imaging, is important for these measurements to not introduce additional background roughness. Atomic Force Microscopy (AFM) scans of 10 μm × 10 μm regions of films transferred to the SiO show the surface is extremely smooth, with an RMS roughness of 0.326 ± 0.020 nm for 5 scans in various regions. In addition, a 20 μm × 20 μm region with a narrow section of the film shows the transferred alumina has a thickness of 32 nm, as indicated by the step height at the edge of the film, as shown in Fig. 3d. To characterize further the transferred alumina films, transmission electron microscopy (TEM) was performed on one of the films which had been transferred to a gold-coated SiO substrate. Initially, a cross-section of the alumina/gold structure was prepared to investigate the nature of the alumina/substrate interface. Specifically, any apparent roughness that could be a result of diffuse polymer-alumina mixing at the interface is of interest. However, during the final thinning in sample preparation, the stack separated, and the alumina film was exfoliated from the gold surface, perhaps due to the film’s inadequate surface adhesion. While the sample was no longer suitable for examining the nature of the interface, the resulting TEM images are useful in that they show a section of the alumina film was able to be exfoliated from its surface a second time while remaining intact. This indicates the film is structurally robust and can potentially be exfoliated and transferred numerous times. The TEM image of this stand-alone alumina region shows the film’s amorphous structure in a plan view in Fig. 3e. As shown in Fig. 3b, the carbon signal in SEM-EDS mapping appears uniform across the entire substrate, regardless of where the transferred alumina film is located. However, to examine the presence of carbon more carefully with a lower detection limit, Raman spectroscopy was performed on the alumina films transferred to the gold coated silicon with the resulting spectra shown in Fig. 3c. There are two sharp peaks that can be associated with the graphitic G peak of carbon at ∼1560 cm and ∼1585 cm. Further, there is no peak present at ∼1355 cm, which would be typical of the amorphous D peak of carbon. The G peak originates from the Eg phonon mode, corresponding to the in-plane bond stretching of all pairs of sp-hybridized carbon atoms in both rings and chains22. In contrast, the D peak arises from the Ag breathing mode of sp carbon atoms only in rings and becomes Raman-active only in the presence of structural defects22. In general, the presence of these peaks alone suggests that the PVA sacrificial layer was not entirely removed from the alumina film, as we now discuss. Interestingly, the two separate G peaks at ∼1560 cm and ∼1585 cm indicate that there are two distinct populations of sp-bonded carbon in the material. A likely explanation is the PVA polymer network, which is highly crystalline, has ordering that encourages formation of small, coherent aromatic clusters that produce the well-defined G peak at 1585 cm. Additionally, strained or slightly disordered graphitic patches could create the separate G peak at 1560 cm23. This seems a likely explanation as the PVA layer is likely to be strained during the ALD processing due to the mismatch in thermal coefficients of expansion of the PVA and alumina impacting film growth as well as stress induced by mechanical transfer. The lack of a D peak at ∼1355 cm in the Raman spectra indicate that the graphitic patches in the PVA are small but relatively defect-free, suppressing D-peak intensity24. Overall, these results indicate that the PVA sacrificial layer is not fully removed, despite the plasma processing intended to clean the surface, as described in Methods. In addition, similar results were found for samples that were plasma processed at higher powers (up to 200 W) and for longer durations of time (up to 10 minutes) than is described in the Methods section. This strongly indicates that carbon originating in the PVA layer is embedded in the ALD grown films. An important temperature-related consideration for ALD on PVA is the stress and strain introduced to the metal oxide film because of the mismatch in thermal coefficients of the polymer substrate and the deposited metal oxide. The coefficient of thermal expansion for PVA is ∼125 × 10 1/C25 while ALD alumina has a much smaller coefficient of thermal expansion ∼4.2 × 10 1/C26. If ALD alumina is grown at elevated temperatures on PVA, the result is compressive stress on the alumina film as both materials cool to room temperature after deposition, with the PVA contracting 30× more than the alumina film26. This can result in a buckling of the ALD film as was observed for a thick film of alumina (500 cycles, ∼70 nm) grown on PVA, as shown in Fig. 4. While the buckling of the alumina film can perhaps be mitigated by slowly cooling the whole stack after deposition the result is indicative that likely intrinsic stresses will remain in the transferred film. (a) SEM micrograph in plan view of 500 cycles ALD AlO grown on PVA. The AlO has buckled which is consistent with the mismatch in thermal coefficients of expansion between the PVA substrate and the AlO film (b) Schematic illustrating how the PVA substrate expands when heated to the ALD growth temperature, and the AlO film is deposited with the substrate in the expanded state. When the PVA substrate cools to room temperature, it contracts 30× more than the AlO, leading to buckling of the film. To investigate the dielectric quality of the transferred alumina, the film was stacked onto a target Si substrate that was first coated with Ti/Au (5 nm/50 nm) using electron beam evaporation. Metal insulator metal (MIM) devices were fabricated by depositing top contacts which consisted of circular pads of Ti/Au (5 nm/50 nm) that varied in diameter from 50-400 μm onto the ALD alumina using electron beam evaporation and a bilayer etch process. Figure 5 shows a light-optical image of an example chip with MIM devices using the transferred alumina film. Example of a transferred AlO film with circular MIM devices. The I-V characteristic of these devices was evaluated, but none of the tested devices were functional capacitors. Excessive leakage current (greater than the 10 nA limit imposed by the instrument) was observed in every device, effectively functioning as short circuits. The deterioration of the dielectric properties of the transferred alumina film are not unexpected given the findings discussed above. Poor adhesion, pinholes, or contamination can lead to gate leakage and performance issues. Mainly, the graphitic carbon evidenced by Raman spectroscopy is likely creating conductive pathways, and the strain induced during the heated ALD growth can be similarly detrimental to the final film. Conclusions In this work, 32 nm thick, millimeter lateral-scale amorphous alumina films grown by ALD were successfully transferred to a target substrate with the aid of a sacrificial PVA layer. Similar to previous studies, the films had angstrom-scale smoothness and were structurally robust, with results showing the films could be exfoliated multiple times while remaining intact. However, contrary to previous reports, the dielectric properties of the films were not maintained during transfer, with strong evidence pointing to subsurface graphitic carbon contamination as the cause. Because the surface of the transferred films was found to be extremely smooth and did not show apparent cracks, characterizing the films’ bulk and buried interface between the alumina and the target substrate is critical. Specifically, creating cross sections of the sample stacks for analyzing the alumina/target substrate interface with TEM could provide insight into the quality of the dielectric integration and is a focus of future work. This future work may be guided by the results reported here, specifically, an appropriate maximum processing temperature of 226C is suggested based on the well-defined melting temperature and lack of glass transition for 99% hydrolyzed PVA. In addition, other thermal considerations that can influence growth and final quality of the ALD films such as the mismatch in thermal coefficient of expansion between PVA and the metal oxide should be considered to reduce strain (e.g. slowly ramping the cooling of the full stack after deposition of alumina). Finally, a key challenge for future work is addressing the carbon contamination from residual unremoved PVA at the buried interface. Methods Fabrication The processing of the full film stack is shown schematically in Fig. 2. A 1 × 1 inch Si wafer with a native oxide was cleaned by sonicating in acetone for 5 minutes, then isopropyl alcohol for 5 minutes, then immediately dried with N. The wafer was then processed with O plasma operated at 100 W for 2 minutes with 20 sccm of O flow. The wafer was then dehydrated by heating at 120C for 5 minutes. PMMA (Kayaku 495 A4) was pipetted onto the cleaned wafer and spin coated at 3000 rpm for 60 seconds to achieve a ∼250 nm thick layer. The wafer was then baked on a hot plate at 180C for 90 seconds, driving off the anisole solvent and hardening the PMMA layer. The PVA solution was prepared as follows: 99% hydrolyzed PVA powder (purchased from Sigma Aldrich) was dissolved in water heated to 90C while stirring for 2 hours. Various concentrations were tested with the most success found for a 20 wt.% PVA solution. The PVA solution was pipetted onto the PMMA coated wafer then spin coated at 3000 rpm for 60 seconds. The wafer was then baked on a hot plate at 180C for 90 seconds. ALD alumina was grown on the PVA coated wafers with a Beneq TFS 500 flow-style reactor. The temperature used for deposition was 150C and the chamber pressure was ∼2 Torr with 250 ms TMA pulse, 4 s N purge, 300 ms H2O pulse, 6 s N purge. For each alumina deposition on the sacrificial PVA substrates, a silicon wafer was included in the reaction chamber, and positioned nearby. The silicon wafer served as a witness substrate to monitor the ALD growth. The thickness, stoichiometric analysis and electrical characterization of these witness films are reported in the supplementary material. PMMA (Kayaku 950 A11) was pipetted on the alumina-coated wafer and then spin coated at 1000 rpm for 60 seconds to achieve a ∼5 μm layer thickness. The wafer was then baked on a hot plate at 180C for 90 seconds, driving off the anisole solvent and hardening the PMMA. Double-sided tape was used to adhere the wafer backside to a glass slide for stability. Thermal release tape was then used to exfoliate the stack. The thermal release tape (with exfoliated stack attached) was then processed with O plasma operated at 100 W for 5 minutes at 20 sccm. The stack was then immediately dry laminated onto the target substrate (e.g. a silicon wafer cleaned using the previously described procedure). The target substrate was then heated to 120C to release the tape. The top PMMA layer was then dissolved in acetone at room temperature for 5 minutes. The wafer was immediately dried with gaseous N. Finally, the sample was processed again with O plasma operated at 100 W for 2 minutes at 20 sccm to remove remaining PMMA residue. Characterization Optical microscopy was performed using a Keyence laser microscope with a 5× objective. The SEM-EDS was performed on a Zeiss Gemini 300 using a 10 keV accelerating voltage and a working distance of 8.0 mm. Electron images were generated using an InLens secondary electron detector, and the EDS maps were generated using an Oxford Ultim max 170 detector with a dwell time of 4.2 ms. Raman spectroscopy was performed using a Horiba LabRAM microscope with a 532 nm laser and 50× objective. TEM imaging was performed using a JEOL JEM 2100F S/TEM. TEM images were formed on a Gatan OneView CMOS camera with an exposure time of 1 second without camera binning. TEM samples were prepared by polishing with a tripod polisher followed by final thinning using a Gatan PIPS II ion polishing system. Included in the supplementary materials, the DSC was performed with a Perkin Elmer PYRIS Diamond DSC. The variable angle spectroscopic ellipsometry measurements were taken with a J.A. Woollam M-2000DI Ellipsometer. The X-ray photoelectron spectroscopy characterization was performed with a Kratos Axis 165 X-ray Photoelectron Spectrometer with a monochromatic Al Kα x-ray source. Supplementary Information Supplementary Information. Acknowledgements The authors thankfully acknowledge Julia Schexnayder and Mohamad Al-Sheikhly for their polymer expertise and generous use of laboratory equipment. Author contributions G.P.B, N.A.B, A.T.H., and A.L.F conceived the experiments, G.P.B. conducted sample fabrication, materials characterization, and electrical measurements. A.T.H. and J.C. conducted TEM characterization. N.A.B, J.C., R.J.P., A.T.H., and A.L.F. advised on experiments and manuscript preparation. All authors reviewed the manuscript. Funding The authors declare no funding for this research. Data availability The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Gillian P. Boyce, Email: gpboyce@umd.edu Adam L. Friedman, Email: afriedman@lps.umd.edu References Associated Data Supplementary Materials Supplementary Information. Data Availability Statement The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.