Machine learning assisted plasma etching based cr2s3 thickness control method

Abstract

The application discloses a kind of Cr2S3 thickness control methods based on machine learning auxiliary and plasma etching, comprising the following steps: step 1, Cr2S3 nanosheet sample is prepared on the growth substrate of different materials, and sample is uniformly transferred to standardization substrate by transfer process;Step 2, the standardized thickness data set corresponding to the real thickness of optical image is constructed;Step 3, based on standardized thickness data set, machine learning regression model is trained, and high-precision mapping model of Cr2S3 optical image information and material thickness is established;Step 4, the Cr2S3 sample to be regulated thickness is placed in plasma etching equipment, and physical sputtering etching is carried out.The application first prepares sample using differentiated growth substrate, then uniformly transfers sample to standardization substrate, eliminates atomic force microscope measurement interference caused by surface roughness of original growth substrate, and ensures the accuracy of machine learning training data.

Description

Technical Field

[0001] This invention relates to the technical field of two-dimensional material preparation and characterization, and in particular to a method for controlling the thickness of Cr2S3 based on machine learning assistance and plasma etching. Background Technology

[0002] Two-dimensional magnetic materials have broad application prospects in spintronic devices, high-density storage, and quantum computing due to their unique physical properties. As a typical non-van der Waals layered magnetic semiconductor, chromium sulfide (Cr2S3) not only has excellent environmental stability, but its interlayer magnetic coupling strength, electronic transport characteristics, and band structure are also strongly dependent on the material thickness. For example, under the thin-film limit, Cr2S3 exhibits significant thickness-dependent magnetic phase transitions and enhanced photoelectric response characteristics. Therefore, achieving atomic-level precise control of the thickness of Cr2S3 is a prerequisite for exploring its intrinsic properties and constructing high-performance devices.

[0003] However, due to the strong covalent interactions between Cr2S3 layers, it is a typical non-van der Waals material, and tends to grow in an isotropic three-dimensional island-like mode during the growth process. At present, it is extremely difficult to directly prepare large-area, uniformly thick ultrathin nanosheets (especially single-layer or few-layer nanosheets) by traditional mechanical exfoliation or chemical vapor deposition (CVD) methods, and the yield is low. Therefore, developing a reliable post-growth thinning process is the key to obtaining high-quality ultrathin Cr2S3.

[0004] Among numerous thinning methods, argon plasma etching exhibits significant advantages over mechanical stripping and wet chemical etching. As a dry process, plasma etching avoids surface contamination and residues introduced by chemical solvents, providing a clean processing environment. Furthermore, by adjusting the radio frequency power and gas flow rate, plasma etching can achieve layer-by-layer, slow physical bombardment of the material surface, possessing the potential to achieve precise sub-nanometer thickness adjustment. It is currently one of the most compatible and controllable thinning technologies in semiconductor processes.

[0005] While plasma thinning technology offers significant advantages in material damage control and compatibility, transforming it into a precise and controllable standardized process hinges on the rapid determination of the etching endpoint and timely adjustment of process parameters. Currently, characterization of the thickness of two-dimensional materials largely relies on general methods such as atomic force microscopy (AFM), Raman spectroscopy, or transmission electron microscopy (TEM). However, these precision instrument-based detection methods typically suffer from cumbersome operation and long imaging cycles. In particular, AFM and TEM often require lengthy single scans or imaging sessions and impose stringent requirements on sample preparation and the identification of specific micro / nano regions. This lengthy data acquisition and analysis process leads to significant process feedback lag, making it difficult for researchers to obtain effective thickness information in a short time to guide subsequent processes during the iterative "etching-detection" cycle. This greatly reduces experimental throughput and preparation efficiency, limiting the application potential of plasma technology in atomic-level thinning.

[0006] Therefore, in order to solve the problem of process feedback lag caused by the low efficiency of traditional detection methods and to overcome the limitations of existing technologies in real-time monitoring, an intelligent solution is urgently needed. Summary of the Invention

[0007] The purpose of this invention is to solve the problems in the prior art by proposing a Cr2S3 thickness control method based on machine learning assistance and plasma etching, which can solve the above problems.

[0008] To achieve the above objectives, this invention proposes a Cr2S3 thickness control method based on machine learning assistance and plasma etching, comprising the following steps: Step 1: Prepare Cr2S3 nanosheet samples on growth substrates of different materials, and transfer the samples uniformly to a standardized substrate using a transfer process; Step 2: Perform optical microscopy imaging on the transferred sample, use AFM to obtain the real thickness label, and simultaneously extract the optical image features of the corresponding region to construct a standardized thickness dataset that corresponds one-to-one between the optical image and the real thickness. Step 3: Based on the standardized thickness dataset, perform feature engineering on the optical image features, and then train a machine learning regression model based on the processed features to establish a high-precision mapping model between Cr2S3 optical image information and material thickness. Step 4: Place the Cr2S3 sample with the thickness to be adjusted in the plasma etching equipment, set the argon flow rate, radio frequency power and sample position parameters, and perform physical sputtering etching. Step 5: Perform optical microscopy imaging on the etched sample, calculate the remaining thickness of the etched sample using the model trained in Step 3, and obtain the true etching rate under the set parameters by using the thickness prediction difference before and after multiple etching rounds and the corresponding etching time; then, under the same etching parameters, adjust the etching time according to the true etching rate until the target thickness is reached.

[0009] Preferably, in step 1, the growth substrate includes, but is not limited to, mica substrate, sapphire substrate and silicon dioxide / silicon substrate, and the standardized substrate is silicon dioxide / silicon substrate.

[0010] Preferably, in step 1, the transfer process is a polydimethylsiloxane-assisted transfer process.

[0011] Preferably, in step 2, the extracted optical image features include red, green and blue color values, contrast, hue, saturation and brightness.

[0012] Preferably, in step 3, the feature engineering process includes: extracting the RGB values ​​of the sample area and the background area and performing linearization processing; calculating optical contrast features, including the contrast of the R, G, and B channels and their square terms; calculating channel ratio features; and converting the RGB space into hue, saturation, and lightness (HSV) space to extract features.

[0013] Preferably, in step 3, the regression model uses a voting regressor, which consists of a weighted random forest regression and gradient boosting regression.

[0014] The beneficial effects of this invention are as follows: This invention first prepares Cr2S3 samples using differentiated growth substrates, and then uniformly transfers the samples to a standardized substrate, eliminating the interference of atomic force microscopy measurements caused by the surface roughness of the original growth substrate, avoiding test data deviations caused by substrate morphology defects, and ensuring the accuracy of machine learning training data. This invention uses a machine learning model to calibrate the rate error caused by substrate etching. In traditional physical sputtering etching, because the plasma simultaneously etches the underlying silicon dioxide / silicon substrate while etching the Cr2S3 sample, the step height change measured by AFM is much lower than the actual material thinning. However, the machine learning model constructed in this invention directly uses optical features to predict the sample thickness, avoiding the step height measurement artifacts caused by simultaneous substrate etching, revealing the true etching process, and improving the thickness control precision to the atomic level.

[0015] The features and advantages of the present invention will be described in detail through embodiments and in conjunction with the accompanying drawings. Attached Figure Description

[0016] Figure 1This is a schematic diagram showing the color changes of Cr2S3 nanosheet samples of different thicknesses according to the present invention.

[0017] Figure 2 The image shows a Raman spectrum of a Cr2S3 sample with a representative thickness.

[0018] Figure 3 A correlation analysis diagram showing the relationship between the thickness predicted by the machine learning model and the actual thickness measured by AFM.

[0019] Figure 4 This is a diagram of a plasma etching apparatus.

[0020] Figure 5 Five samples with different initial thicknesses (labeled S1-S5).

[0021] Figure 6 The diagram shows the morphological evolution during the plasma etching process; (a) shows the optical microscope images of sample S2 after 0 min, 10 min, and 20 min, respectively, and (b) shows the AFM morphology of sample S2 after 0 min, 10 min, and 20 min of etching, respectively, and the corresponding height contour lines.

[0022] Figure 7 This is a graph showing the change in thickness identified by the machine learning model and the thickness measured by AFM over time during plasma etching.

[0023] Figure 8 The Raman spectra of Cr2S3 samples (S2, S3, S5) are compared after 10 min and 20 min of etching.

[0024] In the diagram: 1. Radio frequency coil; 2. Argon plasma; 3. Vacuum pump; 4. Quartz tube; 5. Substrate. Detailed Implementation

[0025] A method for controlling the thickness of Cr2S3 based on machine learning assistance and plasma etching includes the following steps: Step 1: Using chemical vapor deposition, vertically grown Cr2S3 nanosheet samples were prepared on various substrates such as mica and sapphire. By using a polydimethylsiloxane (PDMS)-assisted transfer process, Cr2S3 samples from different growth substrates were uniformly transferred to a silicon wafer with an oxide layer (290nm SiO2) to eliminate the influence of optical background differences and surface roughness of the growth substrates on measurement accuracy. Step 2: On the SiO2 / Si substrate, images of the sample are captured using an optical microscope. Figure 1 The color of the sample changes regularly with thickness; Figure 2The Raman spectrum shows the frequency shift of the characteristic peak of Cr2S3 with thickness.

[0026] On the transferred silica / silicon substrate, flat and fold-free Cr2S3 nanosheets were selected, and their step height was measured using AFM as the true label. The optical red, green and blue color values ​​(RGB) and contrast features of the corresponding regions were extracted to construct a standardized dataset of "optical image-true thickness" containing 85 samples.

[0027] Feature extraction is performed on the dataset, including red, green and blue color values ​​(RGB), contrast, and hue, saturation and brightness (HSV), and an ensemble learning-based regression model is trained.

[0028] The regression model employs a voting regressor, which consists of a weighted combination of random forest regression and gradient boosting regression.

[0029] like Figure 3 The correlation analysis plot shown indicates that the thickness values ​​predicted by the machine learning model are in high agreement with the actual thickness measured by AFM. The data points are closely distributed near the diagonal of y=x, and the coefficient of determination R0 is high. 2 The accuracy reached 0.948, indicating that the model has the ability to replace AFM for fast and high-precision thickness identification.

[0030] Controlled thinning of samples is performed using an inductively coupled plasma system. A schematic diagram of the plasma thinning device is shown below. Figure 4 As shown.

[0031] The plasma thinning device mainly consists of a quartz reaction chamber, an external radio frequency coil, a gas flow controller, and a vacuum pump. The Cr2S3 sample to be processed is placed in the reaction chamber.

[0032] Process parameters: High-purity argon (Ar) gas was introduced into the experiment, with a flow rate of 16 sccm, a cavity pressure of 1 mTorr, and a radio frequency power of 150 W.

[0033] Under these conditions, an argon plasma beam excited by an RF coil bombards the sample surface, achieving physical sputtering thinning.

[0034] Verify the universality and controllability of plasma thinning process on samples with different initial thicknesses. Sample selection and initial characterization: First, five representative Cr2S3 samples (labeled S1-S5) were selected for the experiment; such as Figure 5As shown, these five samples exhibit distinctly different geometric shapes and grayscale characteristics under an optical microscope, intuitively reflecting their varying initial thickness distributions (S1: 21.4 nm, S2: 21.7 nm, S3: 26.3 nm, S4: 26.6 nm, S5: 30.7 nm). This diversity of samples provides a solid experimental basis for verifying the consistency of etching rates.

[0035] Morphological evolution monitoring: To further investigate the specific evolution of morphology and thickness during the etching process, sample S2 was selected as a typical example from the above samples, and a step-by-step etching experiment was conducted on it to demonstrate this dynamic evolution process, such as... Figure 6 As shown in (a), with the increase of etching time, the imaging features of sample S2 under an optical microscope show a significant evolution of grayscale from light to dark. At 0 min, the sample was at its initial thickness (approximately 21.7 nm), exhibiting a clear and uniform light gray color with sharp edges. After 10 minutes of etching, the hue changed to medium gray, indicating that the material thickness was being effectively reduced. When the etching process reached 20 minutes, the sample further turned into dark gray, and its hue contrasted sharply with the substrate background.

[0036] This intuitive evolution from light gray to dark gray clearly records the entire process of Cr2S3 nanosheets being thinned layer by layer by physical sputtering, and the sample maintained its complete geometry throughout the process without breakage or edge chipping.

[0037] like Figure 6 (b) shows the AFM topography diagram and the corresponding height profile, which further quantifies this process: 0 min: AFM scan showed that the sample surface was flat, and the height difference of the measured cross section was approximately 21.7 nm; 10 min: After the first stage of etching, the sample thickness was uniformly reduced to about 17.7 nm, and no obvious defects were observed on the surface; 20 min: The sample thickness was further reduced to approximately 13.7 nm in a controllable manner.

[0038] AFM measurement data confirms that plasma etching can achieve thinning of layered materials.

[0039] Etching rate measurement: Using a trained machine learning model, optical images of all samples S1-S5 at different etching time points are input into the model for real-time thickness identification, and thickness variation curves over time are plotted. Figure 7 As shown.

[0040] The results show that the thickness of all samples exhibits an excellent decreasing trend with etching time. Under the current process conditions (150 W, 1 mTorr), the etching rate of Cr2S3 is constant at about 0.7 nm / min. This stable etching rate indicates that the present invention can achieve atomic-level precise control of the thickness of Cr2S3 by precisely controlling the etching time.

[0041] To verify whether plasma physical sputtering causes damage to the material, Raman spectroscopy analysis was performed on samples after etching for different times.

[0042] like Figure 8 As shown, the Raman spectra of samples S2, S3, and S5 after etching for 10 min and 20 min were compared. The results show that although the sample thickness was significantly reduced, all samples still clearly retained the typical Raman characteristic peaks of Cr2S3 after etching, without obvious defect impurities. The positions of the characteristic peaks shifted regularly with decreasing thickness, which is consistent with... Figure 2 The patterns observed in the intrinsic samples are consistent; this indicates that the argon plasma thinning process used in this invention is a non-destructive physical process that effectively maintains the crystal integrity of Cr2S3 while achieving atomically controllable thinning.

[0043] The above embodiments are illustrative of the present invention and are not intended to limit the present invention. Any simple modifications to the present invention are within the scope of protection of the present invention.

Claims

1. A method for controlling the thickness of Cr2S3 based on machine learning assistance and plasma etching, characterized in that, Includes the following steps: Step 1: Prepare Cr2S3 nanosheet samples on growth substrates of different materials, and transfer the samples uniformly to a standardized substrate using a transfer process; Step 2: Perform optical microscopy imaging on the transferred sample, use AFM to obtain the real thickness label, and simultaneously extract the optical image features of the corresponding region to construct a standardized thickness dataset that corresponds one-to-one between the optical image and the real thickness. Step 3: Based on the standardized thickness dataset, perform feature engineering on the optical image features, and then train a machine learning regression model based on the processed features to establish a high-precision mapping model between Cr2S3 optical image information and material thickness. Step 4: Place the Cr2S3 sample with the thickness to be adjusted in the plasma etching equipment, set the argon flow rate, radio frequency power and sample position parameters, and perform physical sputtering etching. Step 5: Perform optical microscopy imaging on the etched sample, calculate the remaining thickness of the etched sample using the model trained in Step 3, and obtain the true etching rate under the set parameters by using the thickness prediction difference before and after multiple etching rounds and the corresponding etching time; then, under the same etching parameters, adjust the etching time according to the true etching rate until the target thickness is reached.

2. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: In step 1, the growth substrate includes, but is not limited to, mica substrate, sapphire substrate and silicon dioxide / silicon substrate, and the standardized substrate is silicon dioxide / silicon substrate.

3. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: In step 1, the transfer process is a polydimethylsiloxane-assisted transfer process.

4. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: In step 2, the extracted optical image features include red, green and blue color values, contrast, hue, saturation and brightness.

5. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: In step 3, the feature engineering process includes: extracting the RGB values ​​of the sample area and the background area and performing linearization; calculating optical contrast features, including the contrast of the R, G, and B channels and their squared terms; calculating channel ratio features; and converting the RGB space into hue, saturation, and brightness spaces to extract features.

6. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: In step 3, the regression model uses a voting regressor, which consists of a weighted random forest regression and gradient boosting regression.

7. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: The specific parameters for plasma etching are: argon flow rate of 16 sccm, cavity pressure of 1 mTorr, and radio frequency power of 150 W.

8. The Cr2S3 thickness control method based on machine learning assistance and plasma etching as described in claim 1, characterized in that: The etching rate of Cr2S3 was constant at 0.7 nm / min.

Images