RHEED diffraction pattern analysis method based on convolutional neural network

Abstract

The invention relates to an RHEED diffraction pattern analysis method based on a convolutional neural network. The method comprises the following steps: acquiring an RHEED diffraction pattern and a growth temperature of a surface state of a to-be-detected sample in real time when the to-be-detected sample rotates to a set angle; analyzing the RHEED diffraction pattern by using a convolutional neural network to obtain a reconstruction type of the to-be-tested sample at the current growth temperature; and changing the growth temperature and repeating the above steps to obtain the reconstruction types of the sample to be tested at different growth temperatures, thereby obtaining the corresponding relationship between the growth temperature and the reconstruction type of the sample to be tested. According to the method, the RHEED diffraction pattern can be efficiently and rapidly analyzed, and the reconstruction type of the material surface can be accurately identified, so that the corresponding relation between the growth temperature and the reconstruction type is established. Meanwhile, the optimal growth temperature is obtained by observing the change of the surface reconstruction type of the material, so that the accuracy of the semiconductor epitaxial growth temperature can be improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor manufacturing technology, and in particular to a RHEED diffraction pattern analysis method based on a convolutional neural network. Background Art

[0002] During the growth of MBE materials, by real-time monitoring of the diffraction pattern on the surface, the RHEED system can provide important information about the evolution of the surface structure during the film growth process. This information is crucial for adjusting growth parameters and controlling material structure and quality. The application of the RHEED system is not limited to monitoring the growth of thin films, but can also be used to observe the growth of nanostructures, such as observing the growth state of quantum dots and quantum wells. At the same time, it can also help identify the deoxidation state of the material, heterogeneous interfaces, and the crystal structure of the film, providing important information for the characterization of the material growth process. Compared with traditional surface analysis techniques, the RHEED system has the advantages of real-time and non-destructive, and can effectively monitor the microscopic evolution of the film surface structure. Through this system, researchers can more accurately control the film growth process, improve the quality and uniformity of materials, and provide strong support for device preparation.

[0003] Traditional RHEED systems rely on manual judgment and experience to analyze diffraction fringes. Since the sample generally rotates at a speed of 15-30 revolutions per minute, the RHEED diffraction fringes change quickly and the bright spots change greatly, and naked eye observation has certain limitations. Especially in the growth process of InAs quantum dot lasers, temperature is very important. It not only affects the quality of material growth, but also directly affects the emission wavelength and luminous intensity of quantum dot lasers. These two points are important indicators of quantum dot lasers. According to growth experience, every one degree Celsius deviation of the growth temperature from the optimal growth temperature will bring about a wavelength change of about 0.5nm. If the optimal growth temperature is judged manually, it is necessary to observe the RHEED fluorescent screen when the sample holder is stationary, which is seriously affected by subjective factors and it is difficult to ensure accuracy. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a RHEED diffraction pattern analysis method based on a convolutional neural network, which can accurately identify the reconstruction type of the material surface and improve the accuracy of the semiconductor epitaxial growth temperature.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a RHEED diffraction pattern analysis method based on a convolutional neural network, comprising the following steps:

[0006] Real-time acquisition of the RHEED diffraction pattern and growth temperature of the surface state of the sample to be tested when the sample to be tested is rotated to a set angle;

[0007] Analyzing the RHEED diffraction pattern using a convolutional neural network to obtain the reconstruction type of the sample to be tested at the current growth temperature;

[0008] The growth temperature is changed and the above steps are repeated to obtain the reconstruction types of the sample to be tested at multiple different growth temperatures, and then obtain the corresponding relationship between the growth temperature and the reconstruction type of the sample to be tested.

[0009] Furthermore, it also includes determining whether the reconstruction type of the sample to be tested changes during the change of the growth temperature. If a change occurs, the current growth temperature is used as the optimal growth temperature, and the epitaxial growth parameters are adjusted based on the optimal growth temperature.

[0010] Furthermore, the method also includes a step of calculating the deposition thickness of the sample to be tested based on the time consumed when the reconstruction type of the sample to be tested changes twice.

[0011] Furthermore, the sample is kept rotating during the detection process.

[0012] Furthermore, the sample rotation angle is determined by a photoelectric switch.

[0013] Furthermore, the neural network is trained using a plurality of RHEED images of different reconstruction types.

[0014] Beneficial Effects

[0015] Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention uses a convolutional neural network to analyze the RHEED images of the sample surface at different growth temperatures in real time, and obtains the comparative relationship between the growth temperature and the reconstruction type of the sample; at the same time, in the process of changing the growth temperature, by observing whether a change in the reconstruction type occurs on the sample surface, if a change occurs, the growth temperature at the time of the change is used as the optimal growth temperature, thereby achieving precise control of the semiconductor epitaxial growth temperature, which not only improves the epitaxial growth efficiency, but also significantly improves the predictability and stability of semiconductor epitaxial performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of a flow chart of an embodiment of the present invention;

[0017] Figure 2 is a system schematic diagram of an embodiment of the present invention;

[0018] Figure 3 It is a reconstruction type diagram of the surface (2×4) of a GaAs (001) substrate according to an embodiment of the present invention;

[0019] Figure 4It is a c(4×4) reconstruction type diagram of the GaAs(001) substrate surface according to an embodiment of the present invention;

[0020] Figure 5 is a wavelength diagram of a quantum dot laser grown twice through an embodiment of the present invention;

[0021] Figure 6 is a convolutional neural network output result diagram according to an embodiment of the present invention;

[0022] Figure 7 It is a structural diagram of a convolutional neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0024] An embodiment of the present invention relates to a RHEED diffraction pattern analysis method based on a convolutional neural network, which is divided into a training phase and an analysis phase.

[0025] The training phase includes the following steps:

[0026] Use photoelectric switches to determine the sample rotation angle;

[0027] Use high-speed photography to obtain precise images of the RHEED system as the reconstruction occurs;

[0028] Label the RHEED system’s precise images into training and validation sets;

[0029] The convolutional neural network is trained using the training set. The structure of the convolutional neural network is as follows: Figure 7 shown.

[0030] like Figure 1 As shown, the analysis phase includes the following steps:

[0031] Real-time acquisition of the RHEED diffraction pattern and growth temperature of the sample surface state when the sample is rotated to a set angle;

[0032] The RHEED diffraction pattern is analyzed using a convolutional neural network to obtain the reconstruction type of the sample to be tested at the current growth temperature;

[0033] The growth temperature is changed and the above steps are repeated to obtain the reconstruction types of the sample to be tested at multiple different growth temperatures, and then the corresponding relationship between the growth temperature and the reconstruction type of the sample to be tested is obtained.

[0034] During the process of growth temperature change, it is determined in real time whether the reconstruction type of the sample to be tested changes. If a change occurs, the current growth temperature is used as the optimal growth temperature, and the epitaxial growth parameters are adjusted based on the optimal growth temperature.

[0035] In some embodiments, the deposition thickness of the sample to be tested may be calculated based on the time consumed when the reconstruction type of the sample to be tested changes twice.

[0036] The present implementation is further described below through specific tests.

[0037] like Figure 2 As shown, a dual-chamber solid-state source MBE P600 device from DCA of Finland is used, which is equipped with a KSA 400 RHEED system, a 1.8 million pixel 660 frame high-speed camera is selected for image acquisition, and a high-performance host computer is used for data processing.

[0038] Using GaAs (001) substrate as the test sample, the valve opening of the arsenic source furnace was changed to observe the temperature when the surface reconstruction of gallium arsenide changed from (2×4) to c(4×4). When (2×4) reconstruction and c(4×4) reconstruction appeared, the pattern was collected using a high-speed photography platform and sent to the convolutional neural network model for data processing. Both reconstruction patterns were successfully identified and the growth temperature of the current device was successfully read. In some embodiments, the growth temperature of the current sample holder can be obtained by a visual recognition camera.

[0039] In addition, since quantum dot lasers are very sensitive to growth temperature, a logic judgment module can be designed to accurately determine the optimal growth temperature of quantum dot lasers. This logic judgment module can obtain the reconstruction type of the sample to be tested in real time and determine whether a reconstruction transition occurs. By determining the transition point temperature of GaAs reconstruction during each growth process, the thermocouple temperature when the RHEED pattern changes is found. This method can accurately determine the timing of the RHEED pattern transition and read the thermocouple temperature at this time, which is used as the optimal temperature for the growth of quantum dot lasers.

[0040] like Figure 2 As shown in the figure, during the growth process, a high-speed camera is used to shoot the RHEED screen in real time, and each frame of the image is sent to the processing program. The module includes a convolutional neural network part and a data processing part. The convolutional neural network is used to identify which reconstruction type the image belongs to, and the output is as follows: Figure 6 The test results are shown. During this process, the system will determine the type of reconstruction while slowly changing the sample rack temperature. During the temperature change process, the reconstruction type will change. Once the reconstruction type of the sample changes, for example, Figure 3 The (2×4) shown is reconfigured into Figure 4 c(4×4) shown, the system will detect that the reconstruction type has changed, and the processing program will record the temperature at this time. The temperature here is provided in real time by the visual monitoring module. Finally, the program outputs an optimal growth temperature, which is used as the optimal temperature for the growth of quantum dot lasers. The currently obtained optimal growth temperature is automatically written into the growth script of the MBE growth control machine as a growth parameter, and the subsequent growth process automatically grows according to the growth script.

[0041] After the growth is completed, Figure 5 The quantum dot laser samples grown in the two experiments were tested for photoluminescence spectra, and it was found that the peak wavelength deviation of the quantum dot laser grown each time was very small, and the peak emission wavelengths were located at 1309.7nm and 1311.3nm respectively, and the emission intensity was also very close. This proves that even if the source furnace state and substrate quality will change slightly during each growth process, we can still determine the optimal growth temperature more accurately, and repeated experiments can still achieve the above effect. It can be seen that the invention can achieve the expected effect and has good repeatability and stability.

Claims

1. A RHEED diffraction pattern analysis method based on convolutional neural network, characterized in that: The following steps are involved: Real-time acquisition of the RHEED diffraction pattern and growth temperature of the surface state of the sample to be tested when the sample to be tested is rotated to a set angle; Analyzing the RHEED diffraction pattern using a convolutional neural network to obtain the reconstruction type of the sample to be tested at the current growth temperature; The growth temperature is changed and the above steps are repeated to obtain the reconstruction types of the sample to be tested at multiple different growth temperatures, and then the corresponding relationship between the growth temperature and the reconstruction type of the sample to be tested is obtained.

2. The method according to claim 1, characterized in that It also includes judging whether the reconstruction type of the sample to be tested changes according to the output result of the convolutional neural network during the process of the growth temperature changing. If a change occurs, the current growth temperature is used as the optimal growth temperature, and the epitaxial growth parameters are adjusted based on the optimal growth temperature.

3. The method according to claim 2, characterized in that The method also includes a step of calculating the deposition thickness of the sample to be tested based on the time consumed when the reconstruction type of the sample to be tested changes twice.

4. The method according to claim 1, characterized in that The sample was kept rotating during the detection process.

5. The method according to claim 1, characterized in that The sample rotation angle is determined by a photoelectric switch.

6. The method according to claim 1, characterized in that The neural network is trained using a plurality of RHEED images of different reconstruction types.

Images