betaFeedback

WorksPhysics and Astronomy

Identification and Structural Characterization of Twisted Atomically Thin Bilayer Materials by Deep Learning

Abstract

Two-dimensional materials are expected to play an important role in next-generation electronics and optoelectronic devices. Recently, twisted bilayer graphene and transition metal dichalcogenides have attracted significant attention due to their unique physical properties and potential applications. In this study we describe the use of optical microscopy to collect the color space of chemical vapor deposition (CVD) molybdenum disulfide (\mbox{MoS}_2), and the application of a semantic segmentation convolutional neural network (CNN) to accurately and rapidly identify thicknesses of \mbox{MoS}_2 flakes. A second CNN model is trained to provide precise predictions on the twist angle of CVD-grown bilayer flakes. This model harnessed a dataset comprising over 10,000 synthetic images, encompassing geometries spanning from hexagonal to triangular shapes. Subsequent validation of the deep learning predictions on twist angles was executed through the second harmonic generation and Raman spectroscopy. Our results introduce a scalable methodology for automated inspection of twisted atomically thin CVD-grown bilayer.

These authors contributed equally to this work

E-mail: [email protected]; [email protected]; [email protected]; [email protected]

Inspired by magic-angle graphene, [1, 2] twisted bilayer graphene and transition metal dichalcogenides (TMDs) have emerged as a promising platform for the study of moiré physics, encompassing a range of phenomena such as Hubbard physics, [3, 4] superconductivity, [5] or valley polarization. [6] The twist angle in TMDs bilayers can significantly alter their correlated electronic phases and their optical properties. [7] For example, in the twist angle (2∘≤θ<6∘2^\circ\le\theta< 6^\circ), low-frequency interlayer shear and layer breathing modes exhibit rapid change with the twist angle θ\theta. [8] Additionally, the formation of the moiré Brillouin zone introduces new energy subbands in twisted MoS2_2 bilayers with twist angles close to 0∘0^\circ or 60∘60^\circ, [9] or high-lying excitons in bilayer WSe2_2, which can be tuned over 235 meV by enforcing different twist angles in the range 0∘0^\circ to 60∘60^\circ. [10] The electric field control of the 2H bilayer MoS2_2 interlayer exciton at room temperature is possible due to the out-of-plane electric dipole. [11]

Chemical vapor deposition (CVD) can be used to fabricate bilayer graphene [12] and bilayer TMDs with different twist angles, [13, 14, 15, 16, 17] i.e. different stacking arrangements between the two layers. Typically, only 0∘0^\circ (AA stacking, or 3R) or 60∘60^\circ (AB stacking, or 2H) arrangements are possible for the second layer on bilayer TMDs, as these are energetically the more favourable. [16, 18] Reflectivity spectra have shown an A and B exciton energy difference of 49 meV between the 2H and 3R bilayer MoS2_2. [15] Reverse-flow chemical vapor epitaxy provided a way to controllable grow high-quality bilayer TMD single crystals with different growth temperatures. [16]

Second Harmonic Generation (SHG) spectroscopy is frequently employed to characterize the twist angles of TMDs. [19, 20] This reliable method determines the orientation of exfoliated flakes and subsequently enables the stacking of homo- or hetero- layers using a dry transfer technique. [21, 22, 23] In addition to SHG, differential reflectance spectroscopy can also be used for characterizing TMDs, specifically by identifying twist angles through the transition of interlayer excitons. [15, 24] Raman spectroscopy, particularly low wavenumber Raman spectroscopy, [25, 26] is another common method used for this purpose. [8, 16, 27, 28, 29] For cases where atomic precision is required, Transmission Electron Microscopy (TEM) can be employed to determine the twist angles of graphene. [30] Further augmenting the atomic precision of characterization techniques, the Scanning Tunneling Microscope (STM) offers local probing of twist angles with atomic level resolution. [31] Although accurate and reliable, these experimental techniques are costly, require specialised equipment and are time consuming. It is therefore desirable to develop alternative structural characterisation techniques that are cost effective, fast and easily implemented without compromising accuracy and reliability.

Although the thickness is normally confirmed by atomic force microscopy (AFM), Raman, etc, [32] optical contrast is frequently adopted by experience researcher in terms of the speed and simplicity based on the optical contrast. [33] Recent developments in artificial intelligence (AI) have led to the adoption of new techniques for processing microscopy image datasets of layer thicknesses, edges, dimensions, etc. [34, 35, 36] For example, autonomous robotic search and stacking of graphene flakes were proposed to detect up to 400 monolayer graphene samples in one hour. [34, 35] Unsupervised Machine Learning (ML) and Deep Learning (DL) techniques have been used to classify two-dimensional (2D) materials into different categories. [37, 38, 39, 40] In combination with an automatic optical microscope stage, the desired 2D materials can be searched automatically. [39] However, to the best of our knowledge, no automatic procedure for determining the twist angle in bilayer atomically thin materials (e.g. TMDs and graphene) has been described up to now. Presumably, the DL model may lack sufficient accuracy due to the absence of adequate experimental data for training.

In this work, a systematic methodology is reported that demonstrates the potential of DL and image processing tools for determining the twist angles in CVD-grown bilayer TMDs and graphene. Specifically, we trained four different DL algorithms to identify the thickness of the CVD-grown flakes. The twist angle in individual bilayer TMDs flakes can be estimated by means of image processing tools, such as implemented in OpenCV. [41] This procedure, illustrated below, is nevertheless slow and not effective for large-scale sample analysis. To circumvent this problem, we have developed a second DL model to characterize the twist angle of individual flakes in an efficient way. All codes and datasets are open access and freely available, provided with user-friendly instructions. Our work aims to provide new tools designed to facilitate and make more effectively structural characterization of CVD-grown twisted TMDs samples, with extended applicability to CVD-grown graphene and hexagonal boron nitride (h-BN) and other CVD-grown two-dimensional materials.

Identification and analysis of optical micrographs of CVD-grown bilayer atomically thin materials using deep learning

Figure 1. Identification and analysis of optical micrographs of CVD-grown bilayer atomically thin materials (e.g. TMDs) using deep learning. Images of the TMDs are shown as the example. (a) Original optical micrographs (OMs) captured by an Optical Microscope. (b) Images processed and labeled using LabelMe, followed by training via convolutional neural network (CNN) employing various classification methods. (c) Typical outcomes of the TMD thickness derived from the processing in step (b). (d) The regression CNN model trained by the artificially generated dataset for twist-angle prediction. (e) Twist angles predicted by using the CNN model from (d) are shown at the left corner at each image. Scale bar: 10 μm\mu\mathrm{m}.

In our study, optical micrographs of CVD-grown bilayer atomically thin materials are captured as shown in Figure 1 (for MoS2\mathrm{MoS}_{2}) and supplementary for graphene[12] (Figure S15). These images are then processed and utilized to train a convolutional neural network (CNN) for the identification of flake thickness. Subsequently, a different CNN model, developed using a synthetic dataset, is used to predict the twist angles of the flakes, with these predictions displayed on the corresponding images. The workflow diagram of the process to identify the twist bilayer of TMDs is visually shown in the supplementary Figure S1.

We utilized deep learning techniques to determine the thickness of atomically thin flakes, harnessing a supervised neural network trained on manually labeled images. These atomically thin flakes were initially distinguished by optical contrast and subsequently verified through AFM and Raman spectroscopy, [13, 14], detailed in Supplementary Section 1.1.

Figure 2 showcases the adeptness of our segmentation deep learning (DL) models in classifying flakes into monolayers (1L), bilayers (2L), and thicker layers (TL) with remarkable precision, as further elaborated in Figure 2 and Figure S7. Figure 2(a) shows an original unprocessed microscopy image. Figure 2(b-Figure 2e) illustrate the manual labeling of background, monolayer, bilayer, and thick layers.

Segmentation techniques for classifying thickness in atomically thin CVD-grown bilayer flakes

Figure 2. Segmentation techniques for classifying thickness in atomically thin CVD-grown bilayer flakes. (a) Optical micrographs of a bilayer MoS2\mathrm{MoS}_{2} flake. (b-c) Detail the labeling for single and bilayer regions. (d) Final labeled result identifying thickness variations.(e) Explains the color labels, correlating colors to specific thickness levels. (f) Optical micrographs of another bilayer flake using as a test image. (g-n) Four CNN models and segmentation results employing DeepLabV3, FCN, LR-ASPP, and U-Net models respectively. Notably, the U-Net model excels in recognizing the contours of imperfect triangular flakes.Scale bar: 10 μ\mum.

In our evaluation, four DL models were rigorously tested: DeepLabV3, [42] Fully Convolutional Network (FCN), [43] MobileNetV3, [44] and U-Net, [45] each employing distinct segmentation strategies and architectures. For instance, DeepLabV3 integrates a backbone architecture for feature extraction with advanced techniques like dilated convolutions and spatial pyramid pooling. The FCN model, tailored for semantic segmentation, ensures the output size matches the input. Meanwhile, LR-ASPP, a variant of DeepLabV3’s ASPP, aims for efficiency in mobile and edge computing. Notably, U-Net, recognized for its effectiveness in biomedical image segmentation, features a distinctive U-shaped architecture. Our implementation utilized Python 3.8 and Pytorch 1.10, [46] with U-Net emerging as the most adept in capturing the nuances of shape, particularly evident in cases of distorted triangular overlays. Compared to the original micrographs, U-Net’s segmentation reveals enhanced detail in both monolayer and bilayer configurations (Figure 2(n)), affirming its selection as the primary model for our study. For a comprehensive understanding of the labeling process and additional model comparisons, readers are directed to the Supplementary Information (Supplementary 2.1).

Once the pixels in an experimental image have been classified, either by a human or by any of the trained DL models described in Sec. I of SI, and the resulting flakes color-coded according to their thickness, it is possible to determine the twist angle formed between an underlying single-layer and an overlying second layer using appropriate image analysis software, such as OpenCV. [41] It contains useful functionalities to e.g. determine contour lines enclosing a particular flake and measure the corresponding enclosed area, fit an approximate polygonal shape or find the minimum desired polygonal shape that encloses a given set of pixels in the image. Given that TMDs typically exhibit a triangular shape, we employ triangles as the chosen polygonal shape, leading to an estimation of the corner positions. By carrying out this process for each of the layers in a flake, it is possible to estimate the twist angles between each pair of layers from their corner positions by simple trigonometric calculation.

Nevertheless, the use of image analysis software as described above has several disadvantages. First, it is not easily automated, requiring human intervention and thus resulting in a slow and tedious process that cannot be applied efficiently on a large scale. A better alternative is to train a DL model to directly predict the twist angles from the appropriately cropped experimental image of a twisted bilayer [Figure 1(d-e)]. Training such a model would require a large database of preprocessed experimental images for which the twist angles had been previously determined. However there is a more practical and expedient procedure, which consists of employing a synthetically generated image database. As can be seen in Figure 1 and Figure 2, experimental samples of MoS2_2 bilayer flakes typically consist of two roughly triangular shapes, with the rotation angles in the range of 0 to 60∘60^\circ. It is trivial to generate large numbers of synthetic images of rotated pairs of superimposed triangles. This adds the advantage that such images can be made to sample homogeneously all possible twist angles, while training on labelled experimental images would result in a bias towards the experimentally observed twist angles. It is even possible to generate synthetic images that mimic more faithfully the experimental ones, by e.g. cropping corners of the triangles, or introducing random noise to their edges.

In the process of generating the datasets for training the second neural network, a sequence of images is created, including double-layered polygons that morph sequentially from hexagons to truncated triangles and finally into triangles (as illustrated in Figure 3(a)). Formation of the outer polygon begins with the careful selection of random variables, such as central coordinates, side lengths, and rotational angles, promoting visual diversity. Once the outer polygon is defined, an inner polygon is positioned within it. The inner polygon’s location, rotation, and side length are determined through random selection, while ensuring its side length is equal to or less than the outer polygon’s, preserving their hierarchical nesting and maintaining distinct spatial separation between them.

Deep learning approach for recognizing twist angles in atomically thin bilayer flakes

Figure 3. Deep learning approach for recognizing twist angles in atomically thin bilayer flakes. (a) Synthetic datasets illustrating varying twist angles in uniformly colored MoS2_2 flakes post-segmentation. (b) ResNet CNN model training using the linear regression approach on the datasets from (a). (c) The detailed structure of ResNet with input training synthetic datasets and output twist angles of the datasets. (d) Prediction of twist angles for actual as-grown MoS2_2 bilayer samples using the trained CNN model.

The angle formed by a vertex of each polygon and its respective center is calculated, providing a measure to assess the angular difference between them. This angular variation is then incorporated into the filename of the stored image, serving as a perceptible geometric attribute label. Consequently, the generated datasets, enriched with clear geometric annotations and offering a wide variety of forms, become a valuable resource for training datasets in the subsequent recognition of twist angles in CVD-grown samples.

After that, a Residual Network (ResNet)[47] Convolutional Neural Network (see Figure 3) is trained to predict the twist angle by regression to a synthetically generated database of more than 10000 images. The input images are RGB images with a resolution of 512×512512 \times512 pixels obtained from Figure 3(a). In model design, the choice between using a single-layer or multi-layer approach for regression tasks hinges on the complexity of the data relationship

A direct 512 to 1 dimension reduction in the network might be suitable for simpler, linear problem, while a more gradual decrease, such as 512-256-128-1, tends to perform better for capturing complex, non-linear relationships in the data (as shown in Figure 3 (b) and (c)). In the latter approach, additional layers aid the model in learning nuanced data patterns, offering a potential boost in predictive accuracy for intricate problems. Upon being input into the ResNet network, the final angles are obtained, as illustrated in Figure 3 (b) and (c).

After training the CNN regression model, the real as-grown bilayer MoS2_2 micrographs after the first deep learning model are used as input, and the twist angle of the bilayer MoS2_2 are obtained through this second neural network. The capabilities of the model are illustrated in Figure 4. The identification of twist bilayer MoS2_2 using OpenCV is also demonstrated in the supplementary (Figure S8) and shown in the Figure 4 as a comparison for the DL methodology. It can be concluded that the CNN regression model we used here can identify the twist angle in bilayer MoS2_2 while being tolerant to shape irregularities. For bilayer graphene synthesized through CVD, it is evident that different edges can correspond to varying twist angles in the secondary layer [12] (Figure S15). Demonstrably, our methodology proficiently enables the identification of such twist angles within hexagonally shaped, CVD-grown bilayer graphene, offering insightful exploration into its structural intricacies, as depicted in Supplementary Figure S15.

Image not extracted

Figure 4. Performance evaluation of the twisted bilayer MoS2_2 identification CNN model. (a)–(h) present predicted angles for different flakes, showcasing the efficacy of the CNN-based model (indicated in the bottom-left corner of each subfigure) in comparison to angles obtained via OpenCV (displayed in the bottom-right corner). It is noteworthy that in numerous instances, OpenCV was unable to identify the angles, resulting in absent data in the bottom-right corner of some subfigures. (i) Histogram of the quantity of bilayer MoS2_2 at various twist angles. (j)–(u) Second harmonic generation (SHG) and thickness identification for corresponding samples with (v) serving as the legend. (w) Comparison between true and predicted twist angle angles θ\theta using the artificially generated test datasets and SHG dataset in (j)–(u), respectively.

To verify the identification of the twist angles of CVD-grown bilayer MoS2_2, SHG measurements are performed on these samples with various twist angles as shown in Figure 4 (j)-(Figure 4v). The true and predicted twist angle θ\theta using the artificially generated test datasets and SHG dataset is summarized in Figure 4 (w), showing the reasonable accuracy of our model.

One possible application of our methodology could be used to select more samples for probing their optical properties correlations. The strain, defects and doping of the inhomogeneity of the individual bilayer CVD-grown flakes could influence the Raman spectra, making the interpretation of Raman signal challenging. [48, 49, 50, 51, 52, 53, 54] Figure 5 (a) shows the Raman spectra of the CVD-grown bilayer MoS₂ samples with twist angles ranging from 0-60°.

Moiré phonons in twisted CVD-grown bilayer MoS₂

Figure 5. Moiré phonons in twisted CVD-grown bilayer MoS₂. (a) Raman spectra of CVD grown bilayer MoS₂ with the excitation energy (EexE_{\mathrm{ex}}) of 2.54 eV. Peaks assigned to moiré phonons modes[28] including FTA, FLA and FA1′^{\prime}_{1} modes are marked by red, blue and gray arrows, respectively. The dashed lines corresponds to the peak position of second-order Raman modes and the E′E^{\prime} and A1′A^{\prime}_{1} modes are also marked out. (b, c) Comparison between the calculated (Cal, lines) and experimental (Exp, circles) frequencies of twist-angle-dependent moiré phonons including (b) FTA, FLA and (c) FA1′^{\prime}_{1} phonons.

The three vertical dashed lines correspond to the peak positions of the second-order Raman modes E′(M)LO2E^{\prime}(\mathrm{M})\mathrm{LO}_{2}-LA(M) (∼150\sim150 cm−1^{-1}), A1′(M)A^{\prime}_{1}(\mathrm{M})-LA(M) (∼178\sim178 cm−1^{-1}) and TA(K) (∼190\sim190 cm−1^{-1}).[55] In addition, several branches of moiré phonons are observed in the Raman spectra. Moiré phonons refer to zone-center phonons in twisted bilayer MoS₂, which are folded from the off-center phonons in monolayer MoS₂ due to the periodic moiré potential and thus exhibit shift in peak frequency with the change of twist angle. [28] The arrows indicates the assigned moiré phonon modes including folded TA (FTA), folded LA (FLA) and folded A1′A_{1}^{\prime} (FA1′_{1}^{\prime}) modes, which exhibit high frequency sensitivity to the twist angle, similar to the previous results. [28] Figure 5 (b)-(Figure 5c) summarized the experimental and calculated frequencies of the three moiré phonon modes for comparison. The experimental peak positions of moiré phonons agree well with the theoretical ones [28] based on twist angle of twisted bilayer MoS2_2 and the phonon dispersion of monolayer MoS2_2. Note that the frequencies of FA1′_{1}^{\prime} mode on the CVD-grown bilayer MoS2_2 diverging relatively from the theoretical curve in the range of 24.6° to 34.7° may be arisen from different reasons, where strain is very likely as E′ and A1′A_{1}^{\prime} modes also show behaviors of strain-induced shifts. [51] We did not discuss the results of twist angle at 0-5° and 55-60° in detail, because the case is very complicated due to the phonon renormalization induced by lattice relaxation. [8]

Another possible application of our methodology could be used for optimization of the large-scale growth for bilayer atomically materials (as shown in Figure S17). In addition, the exfoliated flakes with regular shape could also be the next possible applications of our model (as show in Figure S18). [56, 57] This thorough work-flow acts as a clear guide, detailing the methodology utilized in the study for the precise identification and analysis of twisted bilayers in TMDs grown via CVD. This method can also be adapted to include graphene (as shown in Figure S15) and hBN, as well as their heterostructures.

In conclusion, we have demonstrated a robust and efficient methodology for the identification and analysis of twisted bilayers in TMDs using deep learning and OpenCV techniques. Our approach, which combines optical micrographs, deep learning models, and OpenCV, provides accurate and comprehensive predictions of thickness properties and twist angles of bilayer TMDs. The comparison of various deep learning models revealed that the U-Net model exhibits superior performance in terms of global accuracy, mean intersection over union, and processing speed.

Our methodology can be extended to other two-dimensional materials grown by CVD, including both homojunctions and heterojunctions, highlighting its versatility and broad applicability in the field of 2D materials analysis. Our datasets and codes are made freely available as a service to the community. We hope that, by facilitating and automating the structural analysis of TMDs, this work will contribute to further advancements in the field of TMDs and 2D materials in general, thus also the rapid growing Autonomous lab using AI.[58]

Methods

A description of the data pipeline, data preparation, deep learning training of the semantic segmentation model, twist angle identification by OpenCV can be found in SI.

Author Contributions

Y. X. supervised the project; Y. X. and E. R. H. conceived the idea. H. Y., R. H., H. W., and X. H. contributed equally to this work; H. Y. performed the coding with the help of Y. X. and R. H.; H. Y. and Y. Z. X. prepared the datasets with assistance from H. C. and M. G.; H. Y., X. H., K. H., and Y. Z. prepared the figures with the help of Y. X., R. H., and E. R. H.; H. W., Y. Z., and X. Z. measured the Raman spectra and SHG, and analyzed the data under the supervision of P. T.; Y. X., R. H., and E. R. H. wrote the manuscript.

Acknowledgements

This work received funding from the National Natural Science Foundation of China (NSFC) Grants (No. 62011530438, No. 61704129, No.12204472, No. 12127807) are acknowledged. This work was partially supported by the Key Research and Development Program of Shaanxi (Program No.2021KW-02), Fundamental Research Funds for the Central Universities (QTZX23026), and the fund of the State Key Laboratory of Solidification Processing in Northwestern Polytechnical University (grant no. SKLSP201612). The work of ERH is supported by MCIN/AEI/ 10.13039/501100011033/FEDER, UE through projects PID2022-139776NB-C66 and TED2021-132267B-I00.

Supporting Information

This material is available free of charge via the Internet at ACS Publications. It includes detailed descriptions of dataset preparation, semantic segmentation results, synthetic dataset generation, verification of twisted bilayer graphene, and Raman spectra of CVD-grown twisted bilayer MoS2_2.

All scripts used in this study are available on GitHub.[59] Currently, the source code is provided as a supplementary file for reviewers and readers to examine. Upon acceptance of the manuscript, the code will be fully open-sourced at the following repository: https://github.com/YongXie-ICMM/Twist2DNet.

Each subfolder in the supplementary package corresponds directly to the respective sections described in this Supporting Information.

(1) Cao, Y.; Fatemi, V.; Fang, S.; Watanabe, K.; Taniguchi, T.; Kaxiras, E.; Jarillo-Herrero, P. Unconventional superconductivity in magic-angle graphene superlattices. Nature 2018, 556, 43–50.

(2) Cao, Y.; Fatemi, V.; Demir, A.; Fang, S.; Tomarken, S. L.; Luo, J. Y.; Sanchez-Yamagishi, J. D.; Watanabe, K.; Taniguchi, T.; Kaxiras, E.; Ashoori, R. C.; Jarillo-Herrero, P. Correlated insulator behaviour at half-filling in magic-angle graphene superlattices. Nature 2018, 556, 80–84.

(3) Tang, Y.; Li, L.; Li, T.; Xu, Y.; Liu, S.; Barmak, K.; Watanabe, K.; Taniguchi, T.;

TOC Graphic

Graphical summary showing a fully convolutional model for thickness identification and segmentation classification, followed by a residual network model for twist angles regression, trained on synthetic datasets

Figure 1 (PDF p. 25).

References

  1. [1]Cao, Y.; Fatemi, V.; Fang, S.; Watanabe, K.; Taniguchi, T.; Kaxiras, E.; Jarillo-Herrero, P. Unconventional superconductivity in magic-angle graphene superlattices. Nature 2018, 556, 43–50.DOI
  2. [2]Cao, Y.; Fatemi, V.; Demir, A.; Fang, S.; Tomarken, S. L.; Luo, J. Y.; Sanchez-Yamagishi, J. D.; Watanabe, K.; Taniguchi, T.; Kaxiras, E.; Ashoori, R. C.; Jarillo-Herrero, P. Correlated insulator behaviour at half-filling in magic-angle graphene superlattices. Nature 2018, 556, 80–84.
  3. [3]Tang, Y.; Li, L.; Li, T.; Xu, Y.; Liu, S.; Barmak, K.; Watanabe, K.; Taniguchi, T.; MacDonald, A. H.; Shan, J.; Mak, K. F. Simulation of Hubbard model physics in WSe₂/WS₂ moiré superlattices. Nature 2020, 579, 353–358.DOI
  4. [4]Wu, F.; Lovorn, T.; Tutuc, E.; MacDonald, A. H. Hubbard model physics in transition metal dichalcogenide moiré bands. Phys. Rev. Lett. 2018, 121, 026402.DOI
  5. [5]Wang, L. et al. Correlated electronic phases in twisted bilayer transition metal dichalcogenides. Nat. Mater. 2020, 19, 861–866.DOI
  6. [6]Devakul, T.; Crépel, V.; Zhang, Y.; Fu, L. Magic in twisted transition metal dichalcogenide bilayers. Nat. Commun. 2021, 12, 6730.DOI
  7. [7]Lau, C. N.; Bockrath, M. W.; Mak, K. F.; Zhang, F. Reproducibility in the fabrication and physics of moiré materials. Nature 2022, 602, 41–50.DOI
  8. [8]Quan, J. et al. Phonon renormalization in reconstructed MoS₂ moiré superlattices. Nat. Mater. 2021, 20, 1100–1105.DOI
  9. [9]Marcellina, E.; Liu, X.; Hu, Z.; Fieramosca, A.; Huang, Y.; Du, W.; Liu, S.; Zhao, J.; Watanabe, K.; Taniguchi, T.; Xiong, Q. Evidence for moiré trions in twisted MoSe₂ homobilayers. Nano Lett. 2021, 21, 4461–4468.DOI
  10. [10]Lin, K.-Q.; Faria Junior, P. E.; Bauer, J. M.; Peng, B.; Monserrat, B.; Gmitra, M.; Fabian, J.; Bange, S.; Lupton, J. M. Twist-angle engineering of excitonic quantum interference and optical nonlinearities in stacked 2D semiconductors. Nat. Commun. 2021, 12, 1553.DOI
  11. [11]Peimyoo, N.; Deilmann, T.; Withers, F.; Escolar, J.; Nutting, D.; Taniguchi, T.; Watanabe, K.; Taghizadeh, A.; Craciun, M. F.; Thygesen, K. S.; Russo, S. Electrical tuning of optically active interlayer excitons in bilayer MoS₂. Nat. Nanotechnol. 2021, 16, 888–893.
  12. [12]Sun, L.; Wang, Z.; Wang, Y.; Zhao, L.; Li, Y.; Chen, B.; Huang, S.; Zhang, S.; Wang, W.; Pei, D.; Fang, H.; Zhong, S.; Liu, H.; Zhang, J.; et al. Hetero-site nucleation for growing twisted bilayer graphene with a wide range of twist angles. Nat. Commun. 2021, 12, 2391.DOI
  13. [13]Xie, Y.; Wang, Z.; Zhan, Y.; Zhang, P.; Wu, R.; Jiang, T.; Wu, S.; Wang, H.; Zhao, Y.; Nan, T.; Ma, X. Controllable growth of monolayer MoS₂ by chemical vapor deposition via close MoO₂ precursor for electrical and optical applications. Nanotechnology 2017, 28, 084001.DOI
  14. [14]Wang, Z.; Xie, Y.; Wang, H.; Wu, R.; Nan, T.; Zhan, Y.; Sun, J.; Jiang, T.; Zhao, Y.; Lei, Y.; Yang, M.; Wang, W.; Zhu, Q.; Ma, X.; Hao, Y. NaCl-assisted one-step growth of MoS₂–WS₂ in-plane heterostructures. Nanotechnology 2017, 28, 325602.DOI
  15. [15]Paradisanos, I.; Shree, S.; George, A.; Leisgang, N.; Robert, C.; Watanabe, K.; Taniguchi, T.; Warburton, R. J.; Turchanin, A.; Marie, X.; Gerber, I. C.; Urbaszek, B. Controlling interlayer excitons in MoS₂ layers grown by chemical vapor deposition. Nat. Commun. 2020, 11, 2391.DOI
  16. [16]Zhang, X.; Nan, H.; Xiao, S.; Wan, X.; Gu, X.; Du, A.; Ni, Z.; Ostrikov, K. Transition metal dichalcogenides bilayer single crystals by reverse-flow chemical vapor epitaxy. Nat. Commun. 2019, 10, 598.DOI
  17. [17]Wang, Z.; Sun, J.; Wang, H.; Lei, Y.; Xie, Y.; Wang, G.; Zhao, Y.; Li, X.; Xu, H.; Yang, X.; Feng, L.; Ma, X. 2H/1T′ phase WS₂(1 − x)Te₂x alloys grown by chemical vapor deposition with tunable band structures. Applied Surface Science 2020, 504, 144371.DOI
  18. [18]Dumcenco, D.; Ovchinnikov, D.; Marinov, K.; Lazic, P.; Gibertini, M.; Marzari, N.; Sanchez, O. L.; Kung, Y.-C.; Krasnozhon, D.; Chen, M.-W.; Bertolazzi, S.; Gillet, P.; Morral, A. F. i.; Radenovic, A.; Kis, A. Large-area epitaxial monolayer MoS₂. ACS Nano 2015, 9, 4611–4620.
  19. [19]Yin, X.; Ye, Z.; Chenet, D. A.; Ye, Y.; O’Brien, K.; Hone, J. C.; Zhang, X. Edge Nonlinear Optics on a MoS₂ Atomic Monolayer. Science 2014, 344, 488–490.
  20. [20]Kumar, N.; Najmaei, S.; Cui, Q.; Ceballos, F.; Ajayan, P. M.; Lou, J.; Zhao, H. Second harmonic microscopy of monolayer MoS₂. Phys. Rev. B 2013, 87, 161403.DOI
  21. [21]Castellanos-Gomez, A.; Buscema, M.; Molenaar, R.; Singh, V.; Janssen, L.; Van Der Zant, H. S.; Steele, G. A. Deterministic transfer of two-dimensional materials by all-dry viscoelastic stamping. 2D Mater. 2014, 1, 011002.
  22. [22]Wang, L.; Meric, I.; Huang, P.; Gao, Q.; Gao, Y.; Tran, H.; Taniguchi, T.; Watanabe, K.; Campos, L.; Muller, D.; Guo, J.; Kim, P.; Hone, J.; Shepard, K.; Dean, C. One-dimensional electrical contact to a two-dimensional material. Science 2013, 342, 614–617.DOI
  23. [23]Cheng, J.; Jiang, T.; Ji, Q.; Zhang, Y.; Li, Z.; Shan, Y.; Zhang, Y.; Gong, X.; Liu, W.; Wu, S. Kinetic nature of grain boundary formation in as-grown MoS₂ monolayers. Adv. Mater. 2015, 27, 4069–4074.DOI
  24. [24]Carrascoso, F.; Lin, D.-Y.; Frisenda, R.; Castellanos-Gomez, A. Biaxial strain tuning of interlayer excitons in bilayer MoS₂. J. Phys. Materials 2019, 3, 015003.DOI
  25. [25]Tan, P. H.; Han, W. P.; Zhao, W. J.; Wu, Z. H.; Chang, K.; Wang, H.; Wang, Y. F.; Bonini, N.; Marzari, N.; Pugno, N.; Savini, G.; Lombardo, A.; Ferrari, A. C. The shear mode of multilayer graphene. Nature Materials 2012, 11, 294–300.DOI
  26. [26]Zhang, X.; Han, W. P.; Wu, J. B.; Milana, S.; Lu, Y.; Li, Q. Q.; Ferrari, A. C.; Tan, P. H. Raman spectroscopy of shear and layer breathing modes in multilayer MoS₂. Phys. Rev. B 2013, 87, 115413.DOI
  27. [27]van der Zande, A. M.; Kunstmann, J.; Chernikov, A.; Chenet, D. A.; You, Y.; Zhang, X.; Huang, P. Y.; Berkelbach, T. C.; Wang, L.; Zhang, F.; Hybertsen, M. S.; Muller, D. A.; Reichman, D. R.; Heinz, T. F.; Hone, J. C. Tailoring the Electronic Structure in Bilayer Molybdenum Disulfide via Interlayer Twist. Nano Lett. 2014, 14, 3869–3875.DOI
  28. [28]Lin, M.-L.; Tan, Q.-H.; Wu, J.-B.; Chen, X.-S.; Wang, J.-H.; Pan, Y.-H.; Zhang, X.; Cong, X.; Zhang, J.; Ji, W.; Hu, P.-A.; Liu, K.-H.; Tan, P.-H. Moiré Phonons in Twisted Bilayer MoS₂. ACS Nano 2018, 12, 8770–8780.DOI
  29. [29]Wu, H.; Lin, M.-L.; Zhou, Y.; Zhang, X.; Tan, P.-H. Analyzing Fundamental Properties of Two-Dimensional Materials by Raman Spectroscopy from Microscale to Nanoscale. Analytical Chemistry 2023, 95, 10821–10838.DOI
  30. [30]Yoo, H.; Engelke, R.; Carr, S.; Fang, S.; Zhang, K.; Cazeaux, P.; Sung, S. H.; Hovden, R.; Tsen, A. W.; Taniguchi, T.; Watanabe, K.; Yi, G.-C.; Kim, M.; Luskin, M.; et al. Atomic and electronic reconstruction at the van der Waals interface in twisted bilayer graphene. Nat. Mater. 2019, 18, 448–453.DOI
  31. [31]Liu, H.; Zheng, H.; Yang, F.; Jiao, L.; Chen, J.; Ho, W.; Gao, C.; Jia, J.; Xie, M. Line and point defects in MoSe₂ bilayer studied by scanning tunneling microscopy and spectroscopy. ACS Nano 2015, 9, 6619–6625.DOI
  32. [32]Li, H.; Zhang, Q.; Yap, C. C. R.; Tay, B. K.; Edwin, T. H. T.; Olivier, A.; Baillargeat, D. From bulk to monolayer MoS₂: evolution of Raman scattering. Advanced Functional Materials 2012, 22, 1385–1390.DOI
  33. [33]Castellanos-Gomez, A.; Agraït, N.; Rubio-Bollinger, G. Optical identification of atomically thin dichalcogenide crystals. Applied Physics Letters 2010, 96, 213116.
  34. [34]Frisenda, R.; Castellanos-Gomez, A. Robotic assembly of artificial nanomaterials. Nat. Nanotechnol. 2018, 13, 441–442.DOI
  35. [35]Masubuchi, S.; Morimoto, M.; Morikawa, S.; Onodera, M.; Asakawa, Y.; Watanabe, K.; Taniguchi, T.; Machida, T. Autonomous robotic searching and assembly of two-dimensional crystals to build van der Waals superlattices. Nat. Commun. 2018, 9, 1413.DOI
  36. [36]Lin, X.; Si, Z.; Fu, W.; Yang, J.; Guo, S.; Cao, Y.; Zhang, J.; Wang, X.; Liu, P.; Jiang, K.; Zhao, W. Intelligent identification of two-dimensional nanostructures by machine-learning optical microscopy. Nano Res. 2018, 11, 6316–6324.DOI
  37. [37]Masubuchi, S.; Machida, T. Classifying optical microscope images of exfoliated graphene flakes by data-driven machine learning. npj 2D Mater. Appl. 2019, 3, 4.DOI
  38. [38]Han, B.; Lin, Y.; Yang, Y.; Mao, N.; Li, W.; Wang, H.; Yasuda, K.; Wang, X.; Fatemi, V.; Zhou, L.; Wang, J. I.-J.; Ma, Q.; Cao, Y.; Rodan-Legrain, D.; et al. Deep-learning-enabled fast optical identification and characterization of 2D materials. Adv. Mater. 2020, 32, 2000953.DOI
  39. [39]Masubuchi, S.; Watanabe, E.; Seo, Y.; Okazaki, S.; Sasagawa, T.; Watanabe, K.; Taniguchi, T.; Machida, T. Deep-learning-based image segmentation integrated with optical microscopy for automatically searching for two-dimensional materials. npj 2D Mater. Appl. 2020, 4, 3.DOI
  40. [40]Sterbentz, R. M.; Haley, K. L.; Island, J. O. Universal image segmentation for optical identification of 2D materials. Sci. Rep. 2021, 11, 5808.
  41. [41]Open Source Computer Vision Library. https://opencv.org/, Accessed: February 10, 2024.
  42. [42]Chen, L.-C.; Papandreou, G.; Schroff, F.; Adam, H. Rethinking atrous convolution for semantic image segmentation. 2017; accessed on December 06, 2023.arxiv.org/abs/1706.05587
  43. [43]Long, J.; Shelhamer, E.; Darrell, T. Fully convolutional networks for semantic segmentation. Proceedings of the IEEE conference on computer vision and pattern recognition. 2015; pp 3431–3440.
  44. [44]Howard, A.; Sandler, M.; Chu, G.; Chen, L.-C.; Chen, B.; Tan, M.; Wang, W.; Zhu, Y.; Pang, R.; Vasudevan, V.; Le, Q. V.; Adam, H. Searching for mobilenetv3. Proceedings of the IEEE/CVF international conference on computer vision. 2019; pp 1314–1324.arxiv.org/abs/1905.02244
  45. [45]Ronneberger, O.; Fischer, P.; Brox, T. U-net: Convolutional networks for biomedical image segmentation. Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. 2015; pp 234–241.arxiv.org/abs/1505.04597
  46. [46]TorchVision maintainers and contributors TorchVision: PyTorch’s Computer Vision Library. https://github.com/pytorch/vision, 2016; Accessed: February 10, 2024.
  47. [47]He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition. 2016; pp 770–778.arxiv.org/abs/1512.03385
  48. [48]Lau, C. N.; Bockrath, M. W.; Mak, K. F.; Zhang, F. Reproducibility in the fabrication and physics of moiré materials. Nature 2022, 602, 41–50.DOI
  49. [49]Lin, K.-Q.; Holler, J.; Bauer, J. M.; Parzefall, P.; Scheuck, M.; Peng, B.; Korn, T.; Bange, S.; Lupton, J. M.; Schüller, C. Large-scale mapping of moiré superlattices by hyperspectral Raman imaging. Advanced Materials 2021, 33, 2008333.DOI
  50. [50]Lu, A.-Y.; Martins, L. G. P.; Shen, P.-C.; Chen, Z.; Park, J.-H.; Xue, M.; Han, J.; Mao, N.; Chiu, M.-H.; Palacios, T.; Tung, V.; Kong, J. Unraveling the Correlation between Raman and Photoluminescence in Monolayer MoS₂ through Machine-Learning Models. Advanced Materials 2022, 34, 2202911.DOI
  51. [51]Yang, R.; Yousuf, S. E. H.; Lee, J.; Zhang, P.; Liu, Z.; Feng, P. X.-L. Raman spectroscopic probe for nonlinear MoS₂ nanoelectromechanical resonators. Nano Letters 2022, 22, 5780–5787.DOI
  52. [52]Xie, Y.; Lee, J.; Wang, Y.; Feng, P. X.-L. Straining and Tuning Atomic Layer Nanoelectromechanical Resonators via Comb-Drive MEMS Actuators. Advanced Materials Technologies 2021, 6, 2000794.DOI
  53. [53]Hu, W.; Wang, Y.; He, K.; He, X.; Bai, Y.; Liu, C.; Zhou, N.; Wang, H.; Li, P.; Ma, X.; Xie, Y. Straining of atomically thin WSe₂ crystals: Suppressing slippage by thermal annealing. Journal of Applied Physics 2022, 132, 085104.DOI
  54. [54]Sun, Y.; Yin, S.; Peng, R.; Liang, J.; Cong, X.; Li, Y.; Li, C.; Wang, B.; Lin, M.-L.; Tan, P.-H.; Wan, C.; Liu, K. Abnormal out-of-plane vibrational Raman mode in electrochemically intercalated multilayer MoS₂. Nano Letters 2023, 23, 5342–5349.DOI
  55. [55]Zhang, X.; Qiao, X.-F.; Shi, W.; Wu, J.-B.; Jiang, D.-S.; Tan, P.-H. Phonon and Raman scattering of two-dimensional transition metal dichalcogenides from monolayer, multilayer to bulk material. Chem. Soc. Rev. 2015, 44, 2757–2785.
  56. [56]Palai, S. K.; Dyksik, M.; Sokolowski, N.; Ciorga, M.; Sánchez Viso, E.; Xie, Y.; Schubert, A.; Taniguchi, T.; Watanabe, K.; Maude, D. K.; Surrente, A.; Baranowski, M.; Castellanos-Gomez, A.; Munnuera, C.; Plochocka, P. Approaching the Intrinsic Properties of Moiré Structures Using Atomic Force Microscopy Ironing. Nano Letters 2023, 23, 4749–4755.DOI
  57. [57]Zhang, N.; Surrente, A.; Baranowski, M.; Maude, D. K.; Gant, P.; Castellanos-Gomez, A.; Plochocka, P. Moiré intralayer excitons in a MoSe₂/MoS₂ heterostructure. Nano letters 2018, 18, 7651–7657.DOI
  58. [58]Ren, Z.; Ren, Z.; Zhang, Z.; Buonassisi, T.; Li, J. Autonomous experiments using active learning and AI. Nature Reviews Materials 2023, 8, 563–564.DOI
  59. [59]Xie, Y. Twist2DNet. https://github.com/YongXie-ICMM/Twist2DNet.git, 2024; Accessed: February 10, 2024.

Paper details

vlnc:W_f23sgddd
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents