betaFeedback

WorksPhysics and Astronomy

Machine learning analysis of structural data to predict electronic properties in near-surface InAs quantum wells

Abstract

Semiconductor crosshatch patterns in thin film heterostructures form as a result of strain relaxation processes and dislocation pile-ups during growth of lattice mismatched materials. Due to their connection with the internal misfit dislocation network, these crosshatch patterns are a complex fingerprint of internal strain relaxation and growth anisotropy. Therefore, this mesoscopic fingerprint not only describes the residual strain state of a near-surface quantum well, but also could provide an indicator of the quality of electron transport through the material. Here, we present a method utilizing computer vision and machine learning to analyze AFM crosshatch patterns that exhibits this correlation. Our analysis reveals optimized electron transport for moderate values of λλ (crosshatch wavelength) and εε (crosshatch height), roughly 1 μμm and 4 nm, respectively, that define the average waveform of the pattern. Simulated 2D AFM crosshatch patterns are used to train a machine learning model to correlate the crosshatch patterns to dislocation density. Furthermore, this model is used to evaluate the experimental AFM images and predict a dislocation density based on the crosshatch waveform. Predicted dislocation density, experimental AFM crosshatch data, and experimental transport characterization are used to train a final model to predict 2D electron gas mean free path. This model shows electron scattering is strongly correlated with elastic effects (e.g. dislocation scattering) below 200 nm λMFPλ_{MFP}.

Introduction

One of the most common features of strained-layer epitaxy is the surface crosshatch pattern that forms as the crystal undergoes misfit dislocation formation, glide, and pile-up [1, 2]. However, as ubiquitous as these crosshatch patterns are, there are few studies attempting to bridge the gap between the observed surface morphologies and material parameters of interest, such as strain state or carrier scattering lifetimes [3, 4]. In principle, internal elastic strain and the surface crosshatch pattern that forms as a result are intimately related via the elasticity tensor of the material system. Proper definition of this interdependence will enable researchers to more rapidly characterize and optimize complex heterostructure growth. For example, an atomic force microscopy (AFM) image of a surface takes over an order of magnitude less time to complete when compared to cryogenic 4-point transport measurements. Previous models from the late 1990s and early 2000s attempted to solve this problem considering either a single uniform strain value [4] or a single line of dislocations [3]. However, expansion to two-dimensional surfaces and the use of these models to correlate surface corrugation to the quality of electronic transport has yet to be unveiled.

Specifically, near-surface quantum well heterostructures within the last decade have been extensively studied for topological superconductivity [5, 6] and gate-tunable superconducting circuits for quantum information[7, 8, 9, 10, 11]. These studies utilize material stacks in which the quantum well region is placed close to the surface of the wafer. In this case, the active region of the wafer is well-separated from the initial mismatched interface by the growth of substantial metamorphic buffer layers [12, 13]. Compared to buried quantum well structures which are well known to have carrier scattering dominated by background and ionized impurities [14], modulation in the surface structure and surface dangling bonds have a much stronger impact on near-surface quantum well systems. As a result, understanding the surface strain state and strain fluctuations becomes of greater importance. Therefore, in these heterostructures it is of great significance to develop models with which we can quickly understand and predict how these materials will perform in cryogenic applications without the need for long fabrication, cooldown, and measurement cycles.

Here, we present a method of AFM image processing for MBE-grown near-surface InAs quantum well structures in which we relate the surface crosshatch pattern to quantum well transport quality. The surface crosshatch pattern is a direct result of misfit dislocation formation during the growth process and is a sign of strain relaxation [1, 2]. Anisotropy in this pattern indicates anisotropic adatom diffusion along <110>-type directions in the (001) plane, which corresponds to anisotropic transport as well. We analyze the room temperature AFM images and extract a parametrization of the physical waveforms on the surface that define the crosshatch and corroborate this with the low temperature electron mean free path. By using the measured electron mean free path of the quantum well, we examine the relationship between the surface crosshatch and resulting transport properties in order to draw conclusions regarding quantum well transport behavior from surface morphology. Simulated 2D crosshatch patterns are used to correlate surface crosshatch patterns to an internal dislocation density for the experimental AFM images. The predicted dislocation densities are then used to model electron scattering mean free path values that are compared against the experimentally measured values. This study serves as a proof of concept for the applicability of computer analysis in the context of AFM image processing in which expanding to the utilization of machine learning is highly desirable.

AFM Crosshatch Extraction and Evaluation

We first present our one-dimensional analysis of the crosshatch pattern. Assuming cubic symmetry of the surface, we would not expect drastically different behavior between the [110] and [1̄10] directions (parallel to the crosshatch peaks/valleys) within the (001) plane. Figure 1 presents the InAs quantum well heterostructure studied here, an exemplary AFM input image, a schematic of our model definitions, and an example of a 1D analysis relating the electron mobilities as a function of the crosshatch wavelength and amplitude. The structure of the near-surface InAs quantum well is presented in Figure 1a, with an examplary input AFM image seen in Figure 1b. For our AFM crosshatch analysis we define the “waveform” of the crosshatch pattern in 1D. We define the wavelength, λ\lambda, as the distance between peaks in the crosshatch and the amplitude, ε\varepsilon, is the height above the midpoint of the crosshatch. A schematic of this definition is presented in Figure 1c, which is an adaptation from the definition presented in Fitzgerald et al. in 1997 [4]. From the analysis we present in Figure 2 and Figure 3, we plot the 1D crosshatch definition presented here against quantum well 2D electron gas (2DEG) mobility for a series of 36 samples, seen in Figure 1d-Figure 1f. The μMAX\mu_{\mathrm{MAX}} and μLOW\mu_{\mathrm{LOW}} plots presented in Figure 1e and Figure 1f, respectively, are the mobilities calculated for the RxxR_{xx} and RyyR_{yy} magnetoresistance traces. The average mobility in plot Figure 1d is the average for each sample as calculated from the high and low values seen in Figure 1e,Figure 1f.

Schematic of near-surface InAs quantum well heterostructure, exemplary AFM image, crosshatch waveform model, and mobility plots

Figure 1. a Schematic of near-surface InAs quantum well heterostructure. b Exemplary 20x20 µm AFM image for a near-surface InAs quantum well sample JS629. c Schematic of crosshatch waveform model, adapted from [4]. d-f Plots of quantum well mobility as a function of the average crosshatch amplitude (εˉ\bar{\varepsilon}) and wavelength (λˉ\bar{\lambda}).

Deconvolution of AFM image

Figure 2. a Averaged color map for vertical crosshatch channel of sample JS629. b and c are the vertical and horizontal channels of sample JS629 after threshold processing to define the peaks and valleys of the waveform. d Overlaid vertical and horizontal channels from our AFM image process algorithm on top of the original rotated AFM image from Figure 1b.

Histograms of extracted AFM widths for the vertical and horizontal channels

Figure 3. Extraction of λ\lambda. Histogram plots in a and b for the vertical and horizontal channels, respectively, for the exemplary AFM image of sample JS629, presented in Figure 1a. The overlaid line in both plots is a Gaussian fit to the histogram from which we extract a center of mass value and a line width. These values are listed in the legend of this plot.

Simple intuition would lead one to believe that the optimal regime in terms of physical structure would be in the limit of εˉ→0\bar{\varepsilon} \to0 and λˉ→∞\bar{\lambda} \to\infty. However, we note a general trend in our analysis suggesting an optimal point in the AFM crosshatch pattern in the region of moderate values of λˉ\bar{\lambda} and εˉ\bar{\varepsilon}. Thus, to more completely evaluate the AFM image in terms of electronic transport behavior, we extend our analysis software to incorporate the second dimension of the crosshatch pattern and fit our parametrization of the crosshatch to a linear regression model.

The AFM analysis is initialized by inputting the raw 20x20 µm images and removing any residual borders from the normalization routines. The initial pixel dimension is then converted into physical length units (nm) based on the AFM scan size. After this step, AFM images are trimmed to standardized pixel densities. The AFM images are then rotated to align the surface crosshatch pattern such that the peaks and valleys are vertical (horizontal). To preserve as much as possible of the original data-set upon rotation, image resolution is increased, however, some data must be cut-off. For example, with a 512x512 pixel image we calculate a 4% maximum possible loss depending on the degree of rotation.

After rotation, our AFM analysis software produces two images corresponding to the vertical (Y) and horizontal (X) channels and then applies a de-noising routine to enhance the signal-to-noise ratio (SNR). The de-noising procedure averages the pixel values in each column across the entire image, preserving the vertical crosshatch and condensing the original 2D image into an effectively 1D line. This line is then re-expanded to the original pixel size for the purpose of comparison to the original AFM image. This produces a smoothed version of the vertical crosshatch “lines”, as seen in Figure 2a. This procedure is also done for the horizontal channel. We then apply a threshold such that height values are cut-off and replaced with empty pixels, examples of this for both the vertical and horizontal channels are shown in Figure 2b and c, respectively. The thresholding algorithm takes as an input the de-noised waveform that was created in the previous step and calculates the absolute midpoint between the maximum and minimum heights. Then, the local maxima (minima) are calculated for each peak (valley) above (below) the global midpoint. The threshold is applied to the image such that only information below the threshold value is kept. To check this result, we also present an overlay of the recreated crosshatch pattern on the original AFM image it was generated from in Figure 2d. In this image, we see good agreement between the “valleys” present in the crosshatch pattern and the red (vertical) and blue (horizontal) channels that our algorithm picks out.

Once the AFM images are processed and we obtain the thresholded channel plots as shown in Figure 2b and Figure 2c, we then extract the widths of each line and gap between lines. These values are plotted in separate histogram plots for the vertical and horizontal channels, examples of which are shown in Figure 3a and Figure 3b. The histogram shows extracted values for the AFM image presented in Figure 1b and corresponds to the de-convoluted crosshatch graphs presented in Figure 2b and Figure 2c. We then fit this histogram to a Gaussian distribution to extract a center of mass (CoM) value and the full-width at half max (FWHM). In the context of a real material evaluation, the CoM is related to the crosshatch density (residual strain state), whereas the FWHM value is then related to the fluctuations in this strain state.

To extract the height of the crosshatch, ε\varepsilon, we use the height information from the AFM images to map the color values of each pixel back to the physical value. For each line, the peak-to-valley heights for the individual X- and Y-crosshatch patterns together to get εx\varepsilon_x and εy\varepsilon_y, respectively. We define a mean value ε‾x,y\overline{\varepsilon}_{x,y} across all 512 lines in the AFM image, from which we calculate the standard deviation, σx,yε\sigma^{\varepsilon}_{x,y} for each AFM image. Thus, the resultant dataset for each AFM image contains λ\lambda, ε\varepsilon, σλ\sigma^\lambda, and σε\sigma^\varepsilon for X- and Y-channels for each sample analyzed in this method.

Linear Regression Model

We then fit the variables from the compiled dataset across all 36 samples to a multiple linear regression model using the stats package within base R statistical software (ver 4.3.3) [15]. The results of two of these model fits are presented in Figure 4. Figure 4a,Figure 4c are the regression model and calculated residuals for Model 1 and Figure 4b,Figure 4d are the same for Model 2. The difference between models is that for Model 2, the λ‾x\overline{\lambda}_x and λ‾y\overline{\lambda}_y variables are removed from the fit. The resulting linear regression equations for Model 1 and 2 are presented in the supplemental material. To generate the plots in Figure 4a, Figure 4b, the model λ‾MFP\overline{\lambda}_{MFP} value is calculated using the linear regression equations found in the supplemental material using the experimental dataset that is obtained through our computer vision algorithm. The R2R^2 and p-values for each model fit are presented in the insets in Figure 4a and Figure 4b.

Four-panel linear regression and residual plots

Figure 4. Linear regression model fitting. a Linear regression model fitting to the full dataset. The red line is the model fit. Calculated R2^2 and p-values are stated in the inset. b Linear regression model fitting with λx\lambda_x and λy\lambda_y removed from dataset. The red line is the model fit. c Calculated residuals for model 1. d Calculated residuals for model 2.

When considering all variables in our dataset (Model 1), we calculate the coefficient of determination, R2R^2, to be 0.27, with a statistical significance (p-value) of 0.21. This is suggestive of some relationship in the dataset to our desired transport parameter, though due to the complexity of the dataset the direct relationship is unclear. However, when removing the λ\lambda values for both X- and Y-channels (Model 2), we calculate the R2R^2 value to be 0.26, i.e. roughly the same variance as in Model 1, except in the new model we calculate a p-value of 0.09. Thus, we conclude that there is a reasonably significant correlation between λ‾MFP\overline{\lambda}_{MFP} measured in transport and surface corrugation as defined by ϵ‾x,y\overline{\epsilon}_{x,y}, σx,yx\sigma^x_{x,y}, and σx,yy\sigma^y_{x,y} (Model 2). The low R2R^2 value is likely due to our small sample size of only 36 samples. Furthermore, the improvement from the removal of the λx,y\lambda_{x,y} variables from the linear model suggests that the dependance of transport quality on crosshatch wavelength is weak at best, but rather how non-uniform this crosshatch is.

However, it is clear a more in-depth model is required to better capture the correlation between the corrugated crosshatch surface and the resulting electronic transport quality.

2D Crosshatch Simulation

As seen in Figure 4, a linear regression model fails to accurately capture the complexity of AFM crosshatch patterns. This suggests that higher dimensionality solutions may be necessary to realize the relationship between the elastic behavior of the semiconductors and their electronic transport behavior. To realize such an analysis, we develop simulation tools to simulate two-dimensional surface crosshatch patterns. This allows for the creation of a large machine learning model with which to better understand the physical system.

Our AFM simulation utilizes the model for surface crosshatch originally proposed by Andrews et al. in 2002 [3]. The original model solves Hooke’s law in one dimension for an arbitrary elastic medium. The model is initialized by defining some starting density of dislocations along the a 1D line, and then integrating the strain field caused by the dislocations to a specified distance away from this starting point. This results in us obtaining an expression for the displacement field of the elastic medium at an arbitrary distance away. We first expanded the model to include two dimensions to generate a surface crosshatch in InAs from a given dislocation density. This model randomly places dislocations within a 20×2020 \times20 μm\mu\mathrm{m} region based on an initial user-defined dislocation density. The model equations are presented below in Equations 1 and 2 for calculating the displacement field in x and y, respectively.

ux=bπ[h2−(x−x0)⋅h(x−x0)2+(y−y0)2+h2−arctan⁡(x−x0h)](1)u_x = \frac{b}{\pi}\left[\frac{h^2-(x-x_0)\cdot h}{(x-x_0)^2+(y-y_0)^2+h^2}-\arctan\left(\frac{x-x_0}{h}\right)\right] \tag*{(1)}
uy=bπ[h2−(y−y0)⋅h(x−x0)2+(y−y0)2+h2−arctan⁡(y−y0h)](2)u_y = \frac{b}{\pi}\left[\frac{h^2-(y-y_0)\cdot h}{(x-x_0)^2+(y-y_0)^2+h^2}-\arctan\left(\frac{y-y_0}{h}\right)\right] \tag*{(2)}

Where bb is the Burgers vector, hh is the effective depth from the dislocation line, and x0,y0x_0,y_0 are the dislocation positions along x and y, respectively. These equations allow us to compute the displacement field at any arbitrary (x,y)(x,y) point in the material, considering the influence of dislocations spread across a 2D plane. For the purposes of this model, we assume the dislocations to be randomly placed in the 2D plane as well as non-interacting. An examplary simulated crosshatch pattern is presented in Figure 5a. We also present the experimental AFM image presented in Figure 1b again, now side-by-side with the 2D version of the simulated crosshatch pattern for comparison. We note good quantitative agreement regarding lambda values, however, the height at the dislocation density presented here is roughly one order of magnitude lower than the experimental data.

Crosshatch surface modeling with panels a–g

Figure 5. Crosshatch surface modeling. a Simulated crosshatch surface for a starting density of 1.4×109/cm21.4 \times10^{9}/\mathrm{cm}^{2}. b and c present exemplary experimental (“original”) and the 2D simulated AFM images, respectively, for comparison. The simulated image presented here is the 2D representation of the 3D crosshatch surface presented in a. d and e: λˉx\bar{\lambda}_{x} and λˉy\bar{\lambda}_{y}, respectively, as a function of dislocation density from our crosshatch simulations. f and g: εˉx\bar{\varepsilon}_{x} and εˉy\bar{\varepsilon}_{y} as a function of dislocation density from our crosshatch simulations. In d-g, the red lines represent the average value across the repeat simulations for each dislocation density while the grey shaded area represents the first standard deviation.

Machine Learning Analysis

When generating the simulated dataset, we simulate the surface 1.51.5 μm\mu\mathrm{m} away from the misfit interface for each dislocation density, consistent with the thickness of the material stack used in our experiments. For simplicity, we assume all dislocations form at the interface with the InP substrate. The simulation was run for 1000 iterations per dislocation density, for 18 different densities ranging from 3×107/cm23 \times10^{7}/\mathrm{cm}^{2} to 3×109/cm23 \times10^{9}/\mathrm{cm}^{2}. With this, we then extract the simulated λ\lambda and ε\varepsilon values for each simulation. The simulated dataset is plotted in Figure 5b,Figure 5c for λˉx,y\bar{\lambda}_{x,y} and Figure 5d,Figure 5e for εˉx,y\bar{\varepsilon}_{x,y}, respectively.

The parameters extracted from the AFM simulations – crosshatch wavelength λˉ\bar{\lambda} and amplitude εˉ\bar{\varepsilon} – are calculated the same as our previous analysis (schematic shown in Figure 1c). We used these parameters to train a Random Forest machine learning model, which predicted the dislocation density from the AFM-derived λˉ\bar{\lambda} and εˉ\bar{\varepsilon} values. The best hyperparameters for the model were determined through cross-validation and yielded a test R2R^2 value of 0.976, indicating our model captures nearly all of the observed variance in predicting dislocation density. More details about the model can be found in the supplemental materials.

We then take the model that defines ρdisloc.(λˉ,ϵˉ)\rho_{\mathrm{disloc.}}(\bar{\lambda},\bar{\epsilon}) and use it to correlate the experimentally measured values of λˉ\bar{\lambda} and ϵˉ\bar{\epsilon} in order to evaluate the dislocation density of our experimentally analyzed AFM surfaces. To evaluate the effectiveness of our model in evaluating the experimental dataset, we predict the dislocation densities for each of the 36 original samples. An example of this comparison is presented in Table I.

MetricSimulatedExperimentalDeviation
RaR_a0.94 Å1.02 Å7.8%
Mean Height6.92 Å7.10 Å2.5%
ϵ\epsilon1.21 Å1.28 Å5.5%

Table I. Comparison of simulated crosshatch parameters and experimental crosshatch parameters for a dislocation density of 5.97×108 cm−25.97 \times10^{8}\,\mathrm{cm}^{-2}. The experimental dislocation density was determined by the model trained on the simulated dataset.

Where RaR_a is the calculated surface roughness and the mean height is the average valley-peak height for the crosshatch pattern. We note that while our model is rudimentary in the assumption of only one mismatched interface, we find that it suitably captures the behavior of the real system.

The experimental ρdisloc.exp.\rho_{\mathrm{disloc.}}^{\mathrm{exp.}} values are then correlated with the experimental electron mean free path values, λˉMFP\bar{\lambda}_{\mathrm{MFP}}, in order to generate a predictive model for λˉMFP\bar{\lambda}_{\mathrm{MFP}}. The results of this model are presented in Figure 6.

Electron mean free path model

Figure 6. Electron mean free path model. Machine learning random forest model fitting the experimental electron mean free path to the dislocation model generated in this study. The red line represents perfect agreement between the machine learning model and the experimentally measured quantities.

The divergence near 200 nm from the model predictions that indicates the physics observed at larger mean free path values is not accurately captured by our model. Rather, the dependence of λˉMFP\bar{\lambda}_{\mathrm{MFP}} on λˉx,y\bar{\lambda}_{x,y}, ϵˉx,y\bar{\epsilon}_{x,y}, and ρdisloc.\rho_{\mathrm{disloc.}} is not strongly defined. We emphasize here that the model only accounts for information regarding the structure of the material, and does not consider any form of electronic scattering. This result indicates that this 200 nm λˉMFP\bar{\lambda}_{\mathrm{MFP}} is a crossover point where the electron scattering goes from being dominated by dislocation scattering induced by elastic strain and plastic deformation caused by epitaxial growth, to being dominated by ionized impurity and electron-electron scattering.

Conclusion

In conclusion, we present here a new method for AFM image analysis using computer image processing to quantitatively parametrize AFM crosshatch patterns and correlate the extracted parametrization with material quality. Using computer vision thresholding, we deconvolute crosshatch images into separate X- and Y- channels and calculate average crosshatch wavelengths, λˉx,y\bar{\lambda}_{x,y}, and standard deviations, σx,yλ\sigma^{\lambda}_{x,y}, for both channels. AFM height data is used to calculate the average height variation, ϵˉx,y\bar{\epsilon}_{x,y}, and the standard deviations, σx,yϵ\sigma^{\epsilon}_{x,y}, for both channels. We present newly developed simulation tools to generate AFM crosshatch patterns from a

known dislocation density. Using this simulation, we are able to generate large datasets with which to train a machine learning model that relates the crosshatch λˉ\bar{\lambda} and ϵˉ\bar{\epsilon} values to an initial dislocation density. From this model we relate the experimental AFM image to a predicted dislocation density in order to correlate electron mean free path, λˉMFP\bar{\lambda}_{\mathrm{MFP}}, to a dislocation density, ρdisloc.\rho_{\mathrm{disloc.}}. We finally present a model based on predicted dislocation density that attempts to predict λˉMFP\bar{\lambda}_{\mathrm{MFP}} based solely on strain-induced displacement fields. We observe a divergence between the experimental values and predicted values near λˉMFP=200\bar{\lambda}_{\mathrm{MFP}} = 200 nm that points to this point as a crossover between transport that is dominated by scattering induced by strain-fields and ionized impurity scattering.

Experimental Details

Material Growth: InAs near-surface quantum wells are grown in a Varian Gen II molecular beam epitaxy (MBE) system on 50.8 mm InP (001) wafers. The InAs quantum well structures are grown on a strain-relaxed graded buffer layer with top-layer composition of In0.81_{0.81}Al0.19_{0.19}As. The quantum well is grown to be 4 mm thick and is separated from the wafer surface by a 10 nm thick In0.81_{0.81}Ga0.19_{0.19}As barrier layer. More details and specifics about this structure growth can be found in reference [12].

Atomic Force Microscopy Imaging: Atomic force microscopy imaging is done with a NanoMagnetics Instruments AFM operating in a dynamical tapping mode. Preliminary AFM image normalization and processing is done using the Gwyddion software package [16].

Quantum Hall Characterization: Quantum Hall transport is measured in an Oxford TelsatronPT 1.5 K He4 cryostat. The 2-in wafers are diced into roughly 5×55 \times5 mm pieces which then are etched in Transene Type-D (Transene Company) aluminum etchant to remove the top Al layer. Samples are then cleaned in solvents, rinsed in de-ionized water, and then blow-dried with dry nitrogen. Measurements are done in a Van der Pauw wiring configuration.

Supplemental Information

The supplemental material document contains further details regarding the linear regression model used and the machine learning model data curation, model selection, and training.

Acknowledgments

We thank Stefano Martiniani for fruitful discussions. The authors would like to acknowledge support for this project from the Office of Naval Research award number N00014-22-1-2764.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The AFM analysis code is available on github: https://github.com/ShabaniLab.

References

References

  1. [1]R. A. Burmeister, G. P. Pighini, and P. E. Greene, Large Area Epitaxial Growth of GaAs1−xPx for Display Applications, Trans. Am. Inst. Min. Metall. Pet. Eng. Inc. 245, 587 (1969).
  2. [2]K. H. Chang, R. Gibala, D. J. Srolovitz, P. K. Battacharya, and J. F. Mansfield, Crosshatched surface morphology in strained iii-v semiconductor films, J. Appl. Phys. 67, 4093 (1990).DOI
  3. [3]A. M. Andrews, J. S. Speck, A. E. Romanov, M. Bobeth, and W. Pompe, Modeling cross-hatch surface morphology in growing mismatched layers, J. Appl. Phys. 91, 1933 (2002).DOI
  4. [4]E. A. Fitzgerald, S. B. Samavedam, Y. H. Xie, and L. M. Giovane, Influence of strain on semiconductor thin film epitaxy, J. Vac. Sci. Technol. A 15, 1048 (1997).DOI
  5. [5]J. Shabani, M. Kjaergaard, H. J. Suominen, Y. Kim, F. Nichele, K. P. Kroustrup, T. Stankevic, R. M. Lutchyn, P. Krogstrup, R. Feidenhans’l, S. Kraemer, C. Nayak, M. Troyer, C. M. Marcus, and C. J. Palmstrøm, Two-dimensional epitaxial superconductor-semiconductor heterostructures: A platform for topological superconducting networks, Phys. Rev. B 93, 155402 (2016).DOI
  6. [6]S. M. Frolov, M. J. Manfra, and J. D. Sau, Topological superconductivity in hybrid devices, Nat. Phys. 16, 718 (2020).DOI
  7. [7]L. Casparis, M. R. Connolly, M. Kjaergaard, N. J. Pearson, A. Kringhøj, T. W. Larsen, F. Kuemmeth, T. Wang, C. Thomas, S. Gronin, G. C. Gardner, M. J. Manfra, C. M. Marcus, and K. D. Petersson, Superconducting gatemon qubit based on a proximitized two-dimensional electron gas, Nat. Nanotechnol. 13, 915 (2018).DOI
  8. [8]G. Scappucci, C. Kloeffel, F. A. Zwanenburg, D. Loss, M. Myronov, J.-J. Zhang, S. De Franceschi, G. Katsaros, and M. Veldhorst, The germanium quantum information route, Nat. Rev. Mater. 6, 926 (2021).
  9. [9]M. Hinderling, S. C. ten Kate, M. Coraiola, D. Z. Haxell, M. Stiefel, M. Mergenthaler, S. Paredes, S. W. Bedell, D. Sabonis, and F. Nichele, Direct microwave spectroscopy of andreev bound states in planar ge josephson junctions, arXiv:2403.03800 (2024).DOI
  10. [10]A. Tosato, V. Levajac, J.-Y. Wang, C. J. Boor, F. Borsoi, M. Botifoll, C. N. Borja, S. Martí-Sánchez, J. Arbiol, A. Sammak, M. Veldhorst, and G. Scappucci, Hard superconducting gap in germanium, Commun. Mater. 4 (2023).DOI
  11. [11]W. M. Strickland, L. J. Baker, J. Lee, K. Dindial, B. H. Elfeky, P. J. Strohbeen, M. Hatefipour, P. Yu, I. Levy, J. Issokson, V. E. Manucharyan, and J. Shabani, Characterizing losses in inas two-dimensional electron gas-based gatemon qubits, Phys. Rev. Research 6, 023094 (2024).arxiv.org/abs/2309.17273
  12. [12]W. M. Strickland, M. Hatefipour, D. Langone, S. M. Farzaneh, and J. Shabani, Controlling fermi level pinning in near-surface inas quantum wells, Appl. Phys. Lett. 121, 092104 (2022).DOI
  13. [13]L. E. A. Stehouwer, A. Tosato, D. D. Esposti, D. Costa, M. Veldhorst, A. Sammak, and G. Scappucci, Germanium wafers for strained quantum wells with low disorder, Appl. Phys. Lett. 123, 092101 (2023).DOI
  14. [14]Y. J. Chung, A. Gupta, K. W. Baldwin, K. W. West, M. Shayegan, and L. N. Pfeiffer, Understanding limits to mobility in ultrahigh-mobility gaas two-dimensional electron systems: 100 million cm²/vs and beyond, Phys. Rev. B 106, 075134 (2022).DOI
  15. [15]R Core Team, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria (2024).
  16. [16]D. Nečas and P. Klapetek, Gwyddion: an open-source software for spm data analysis, Cent. Eur. J. Phys. 10, 181 (2012).DOI

Paper details

vlnc:W_n5ud2z3y
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