A real-time data monitoring and early warning method for a VCSEL epitaxial wafer growth process

Abstract

The present application relates to the technical field of data processing, more particularly, the present application relates to a kind of VCSEL epitaxial wafer growth process real-time data monitoring early warning method, comprising: using MOCVD equipment obtains the refractive index of the material used in the current time epitaxial wafer in VCSEL epitaxial wafer growth process;Determine the growth rate of the current epitaxial wafer;Determine the mutation degree of the growth rate of the current epitaxial wafer;Determine the correction mutation degree of the growth rate of the current epitaxial wafer;Determine the early warning threshold of the current time;According to the size of the correction mutation degree and the early warning threshold, to realize the real-time data monitoring early warning of VCSEL epitaxial wafer growth process.The present application realizes growth rate accurate inversion and mutation intelligent identification, effectively distinguishes abnormal and normal process change, self-adapting adjustment early warning threshold, reduces false alarm rate, improves the accuracy and practicality of VCSEL epitaxial growth monitoring.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a real-time data monitoring and early warning method for the growth process of VCSEL epitaxial wafers. Background Technology

[0002] Vertical-cavity surface-emitting lasers (VCSELs) are core light source devices in cutting-edge technologies such as optical communication, 3D sensing, and lidar. Their performance is highly dependent on the quality of the epitaxial wafer composed of multiple layers of semiconductor materials. The quality of the epitaxial wafer, especially the thickness, composition, and interface quality of each material layer, directly determines the photoelectric performance, reliability, and yield of the final VCSEL chip. Metal-organic chemical vapor deposition (MOCVD) is currently the mainstream technology for manufacturing VCSEL epitaxial wafers. This process involves complex physicochemical reactions and requires extremely stringent stability of parameters such as temperature, pressure, and gas flow rate in the growth environment to ensure the thickness uniformity of the epitaxial wafer across the entire wafer surface and throughout the growth process.

[0003] In traditional MOCVD growth control, the monitoring methods for the epitaxial wafer growth process are relatively limited. They usually rely on manual monitoring by operators with experience, or offline detection methods such as X-ray diffraction and scanning electron microscopy to assess the quality of the finished epitaxial wafer after the entire growth process is completed. These traditional methods have significant lag and cannot detect and correct potential deviations in process parameters in real time during the growth process. For example, when the heater power fluctuates or the gas flow controller becomes unstable, the growth rate will change abnormally, causing the epitaxial layer thickness of the epitaxial wafer to deviate from the design value. Traditional methods often cannot detect this until several hours after the growth is completed. By then, the entire wafer or batch of epitaxial wafers may already be of poor quality, resulting in a waste of materials and time costs.

[0004] Therefore, by introducing online monitoring technology based on reflectance spectroscopy, the refractive index change of the material used in the epitaxial wafer is monitored in real time, and the real-time growth rate of the epitaxial wafer is derived. The stability of the growth rate is a direct reflection of the thickness uniformity of the epitaxial wafer. However, since the VCSEL structure contains multiple layers of materials with different compositions, such as GaAs and AlAs, when switching from one epitaxial material to another, the growth rate will undergo a normal and expected step change due to the difference in material physical properties. This change can be called a benign mutation. At the same time, abnormal fluctuations in the growth rate caused by unknown factors such as equipment instability or process parameter drift are abnormal mutations that require early warning. Existing monitoring methods are difficult to effectively distinguish between these two completely different mutations, often misjudging normal material process switching as abnormal, thus generating a large number of false alarms and reducing the reliability and practicality of real-time data monitoring and early warning of the VCSEL epitaxial wafer growth process. Summary of the Invention

[0005] To address the problem of inaccurate early warnings in existing technologies due to the inability to distinguish between normal epitaxial material switching and unstable actual equipment parameters, this invention proposes a real-time data monitoring and early warning method for the VCSEL epitaxial wafer growth process. This method includes the following steps:

[0006] The refractive index of the material used in the VCSEL epitaxial wafer at the current moment is obtained using an MOCVD device. Based on the refractive index and the dominant frequency of the refractive index data sequence corresponding to the current moment's preset time window, the growth rate of the epitaxial wafer at the current moment is determined. Based on the instantaneous fluctuations and continuous trends of the growth rate within the preset time window, the abrupt change in the growth rate of the epitaxial wafer at the current moment is determined. Based on the changes in the standard deviation of the refractive index within the preset time window, the abrupt change is corrected to obtain the corrected abrupt change in the growth rate of the epitaxial wafer at the current moment. Clustering is performed on the abrupt change in the growth rate of the epitaxial wafer at each moment (including the current moment and all previous historical moments) and the corrected abrupt change in the growth rate of the epitaxial wafer at each moment to determine the warning threshold for the current moment. Based on the magnitude of the corrected abrupt change and the warning threshold, real-time monitoring and early warning of the VCSEL epitaxial wafer growth process are achieved.

[0007] This invention achieves accurate inversion of epitaxial wafer growth rate by analyzing the dominant frequency of refractive index data sequence within a preset time window and combining it with real-time refractive index data. It fully leverages the real-time advantages of reflectance spectroscopy, providing an accurate data foundation for monitoring the growth process. By organically combining instantaneous fluctuations with continuous trends, a comprehensive evaluation model for abrupt changes is constructed, effectively identifying abnormal growth rate characteristics, distinguishing between instantaneous fluctuations and continuous changes, and improving the accuracy of abrupt change detection. By correcting for abrupt changes based on changes in the refractive index standard deviation, intelligent identification of normal process changes such as material switching is achieved. When the refractive index standard deviation changes significantly, it indicates a possible material switching, and the abrupt change is corrected accordingly, effectively avoiding misjudgments of benign abrupt changes. Through cluster analysis of the difference between historical abrupt changes and corrected abrupt changes, an adaptive warning threshold is determined, achieving intelligent threshold adjustment. This allows for dynamic adjustment of the warning standard based on the statistical characteristics of the actual growth process, improving the adaptability and accuracy of the monitoring system, effectively reducing the false alarm rate, and enhancing the reliability and practicality of monitoring and warning of VCSEL epitaxial wafer growth process.

[0008] Furthermore, the growth rate satisfies:

[0009] In the formula, This represents the growth rate of the epitaxial wafer at the current moment. The wavelength of the light source used in the spectral sensing equipment of the MOCVD machine. The refractive index of the material used in the epitaxial wafer at the current moment. The dominant frequency of the refractive index data sequence corresponding to the preset time window at the current moment.

[0010] This invention constructs a growth rate calculation model that includes wavelength, refractive index, and dominant frequency, achieving accurate inversion based on reflection spectroscopy technology and accurately reflecting the real-time growth status of epitaxial wafers. The extraction of dominant frequency can effectively capture the periodic variation characteristics in the refractive index data sequence, reflecting the dynamic characteristics of the growth process. The real-time change of refractive index directly reflects the change in material properties and, together with wavelength, constitutes the basic parameters for optical measurement. Through the direct correlation between optical parameters and growth rate, the real-time performance and accuracy of monitoring the VCSEL epitaxial wafer growth process are improved.

[0011] Furthermore, the method for obtaining the dominant frequency of the refractive index data sequence is as follows: based on the refractive index data of the materials used in the epitaxial wafer at all times within the preset time window to which the current time belongs, a refractive index data sequence corresponding to the time window to which the current time belongs is constructed, and the dominant frequency of the refractive index data sequence is identified by using the fast Fourier transform algorithm on the refractive index data sequence.

[0012] Furthermore, the degree of mutation satisfies:

[0013] In the formula, The abrupt change in the growth rate of the epitaxial wafer at the current moment. This represents the growth rate of the epitaxial wafer at the current moment. The first of the remaining times within the preset time window that does not include the current time. The growth rate of the epitaxial wafer at a given time point. This represents the number of times within the preset time window to which the current time belongs. The feature point corresponding to the first moment within the preset time window to which the current moment belongs. The feature point corresponding to the current time. for and The slope between For the standard normalized function, It is the absolute value symbol.

[0014] This invention achieves a comprehensive assessment of mutation degree by constructing a composite function containing a variance term and a slope term. The variance term reflects the deviation of the current growth rate from the historical rate within the time window, while the slope term reflects the changing trend between feature points. The combination of the two can comprehensively assess the mutation characteristics of the growth rate. The variance term effectively captures instantaneous fluctuations; when the current rate differs significantly from the historical average, the mutation degree increases accordingly. The slope term reflects the continuous changing trend and can identify continuous upward or downward trends. Normalization ensures the comparability of two indicators with different dimensions, and weighted averaging allows the mutation degree to balance the influence of instantaneous fluctuations and trend changes.

[0015] Furthermore, the feature points are obtained by using the temporal sequence of all moments within the preset time window to which the current moment belongs as the horizontal axis and the growth rate of the epitaxial wafer at all moments within the preset time window to which the current moment belongs as the vertical axis to construct the feature points corresponding to all moments within the preset time window to which the current moment belongs.

[0016] Furthermore, the modified mutation degree satisfies:

[0017] In the formula, This represents the corrected abrupt change in the growth rate of the epitaxial wafer at the current moment. The abrupt change in the growth rate of the epitaxial wafer at the current moment. The standard deviation of the refractive index of the material used in the epitaxial wafer at all times within the preset time window to which the current time belongs. This represents the standard deviation of the refractive index of the material used in the epitaxial wafer at all times within the preset time window preceding the current time. For hyperparameters, It is a function with maximum value. It is a natural exponential function. It is an absolute value function.

[0018] This invention constructs an adaptive adjustment mechanism for correcting abrupt changes by introducing an exponential decay term for the change in refractive index standard deviation. This mechanism can intelligently adjust the weight of the abrupt change according to the changes in refractive index fluctuations, effectively distinguishing between material switching and abnormal fluctuations. When the refractive index standard deviation changes significantly, the exponential term approaches 0, and the corrected abrupt change decreases accordingly, indicating that a normal material switching may have occurred, thus avoiding misjudgment as an anomaly. When the change in refractive index standard deviation is small, the exponential term approaches 1, and the corrected abrupt change approaches the original abrupt change, maintaining sensitivity to abnormal fluctuations.

[0019] Furthermore, the preset number of clusters for the clustering is 2.

[0020] Further, determining the warning threshold at the current moment includes: constructing a feature dataset based on the difference; clustering the feature dataset using a clustering algorithm to obtain two clusters; designating the cluster with the fewest data points among the two clusters as the suspected benign mutation cluster; and determining the maximum value among the corrected mutation degrees corresponding to all data points within the suspected benign mutation cluster as the warning threshold at the current moment.

[0021] This invention constructs a feature dataset of the difference between mutation degree and corrected mutation degree, enabling statistical analysis of historical monitoring data. A clustering algorithm is used to divide the differential data into two clusters, effectively distinguishing the statistical characteristics of benign mutations and normal / abnormal clusters. Clusters with fewer data points are identified as suspected benign mutation clusters. Reasonable cluster classification based on the relative scarcity of benign mutations avoids the complexity of pre-labeling training data, achieving unsupervised adaptive threshold determination. By selecting the maximum corrected mutation degree within a benign mutation cluster as the warning threshold, it ensures that normal process switching will not trigger false alarms.

[0022] Furthermore, the clustering adopts the k-means clustering algorithm.

[0023] Furthermore, the method for real-time monitoring and early warning of VCSEL epitaxial wafer growth process includes: in response to the correction mutation degree being greater than the early warning threshold, determining that there is a risk in the epitaxial wafer growth process at the current moment, and triggering an early warning prompt, thereby completing the real-time data monitoring and early warning of the VCSEL epitaxial wafer growth process.

[0024] The present invention has the following beneficial effects:

[0025] (1) This method overcomes the limitation of existing monitoring methods in distinguishing between benign and anomalous mutations by using a two-layer assessment of mutation degree and corrected mutation degree to achieve accurate identification. The mutation degree integrates the instantaneous fluctuations and continuous trends of growth rate. Benign mutations are usually characterized by step-like but stable changes, while anomalous mutations are accompanied by irregular and continuous fluctuations. The corrected mutation degree further incorporates changes in the standard deviation of refractive index to enhance the characteristic differences between the two types of mutations. This layered identification mechanism essentially distinguishes the physical characteristics of the two types of mutations, reduces the misjudgment of normal material switching as anomalous, and lowers the false alarm rate.

[0026] (2) By clustering the difference between historical mutation degree and corrected mutation degree, the warning threshold at the current moment is determined, replacing the traditional fixed threshold setting. It adapts to different stages of VCSEL epitaxial growth, changes in equipment status and process batch fluctuations, so that the threshold always matches the normal fluctuation baseline of the current growth environment. For example, in the high stability material growth stage, the threshold is automatically narrowed to sensitively capture small anomalies. In the stage of frequent material switching, the threshold is appropriately widened to tolerate normal process fluctuations, ensuring that the warning neither misses real anomalies nor responds to false signals.

[0027] (3) Correction of mutation degree is calibrated by changing the standard deviation of refractive index to effectively filter out pseudo mutation signals caused by measurement noise and environmental interference, making the characterization of growth rate mutation closer to the real physical process, ensuring early detection of abnormalities such as equipment instability and process parameter drift. For example, when the growth rate deviates slightly but continuously due to abnormal furnace temperature, the correction of mutation degree can identify and trigger an early warning, giving maintenance personnel time to adjust, avoiding quality problems such as uneven epitaxial wafer thickness and composition deviation, and ensuring the process stability of VCSEL epitaxial growth.

[0028] (4) After reducing false alarms, the early warning information of the monitoring system is more valuable and enhances the operators' trust in the system. At the same time, it avoids unnecessary shutdowns for inspections due to misjudgment, such as interrupting the material switching process for equipment inspection, reducing interference with the normal production rhythm. Accurate abnormal early warnings can also guide targeted maintenance, such as adjusting only the drift process parameters without a complete shutdown, reducing production interruption time and maintenance costs, and improving the production efficiency and yield of VCSEL epitaxial wafers. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the steps of a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of the data sequence of material refractive index samples for a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention.

[0031] Figure 3 This is a schematic diagram of the main frequency of the refractive index data sequence corresponding to each time window of a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention.

[0032] Figure 4 This is a schematic diagram of the growth rate of the epitaxial wafer at various times in a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention.

[0033] Figure 5 This is a schematic diagram of the abrupt change in the growth rate of an epitaxial wafer at various times and a corrected abrupt change diagram in a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention.

[0034] Figure 6 This invention relates to a real-time data monitoring and early warning method for the growth process of VCSEL epitaxial wafers. A schematic diagram of the stable clustering results at time points.

[0035] Figure 7 This invention relates to a real-time data monitoring and early warning method for the growth process of VCSEL epitaxial wafers. A schematic diagram of the stable clustering results at time points.

[0036] Figure 8 This is a schematic diagram illustrating the early warning identification of a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below. The described embodiments are only a part of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0039] Please see Figure 1 The diagram illustrates a flowchart of a real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to an embodiment of the present invention. The method includes the following steps:

[0040] S01: Use MOCVD equipment to obtain the refractive index of the material used in the VCSEL epitaxial wafer at the current moment during the growth process.

[0041] It should be noted that VCSEL epitaxial wafer growth is a complex physicochemical process, and the final product quality is determined by multiple factors such as process thermodynamics, kinetics, and fluid dynamics. In order to analyze the true state of the epitaxial wafer growth process, it is necessary to acquire growth process parameters simultaneously, so as to provide a comprehensive data foundation for subsequent analysis.

[0042] Specifically, for each epitaxial wafer growth process, the wavelength of the light source used by the spectral sensing device integrated on the MOCVD equipment is obtained. For example, if the spectral sensing device is a reflectometer, the most commonly used light source is a helium-neon laser with a wavelength of 632.8 nanometers. The reflectometer is then used to obtain the refractive index of the material used in the VCSEL epitaxial wafer at the above wavelength in real time. For example, the refractive index of GaAs material is about 3.8, while the refractive index of AlAs material is about 3.0.

[0043] The implementers can set the data collection frequency according to the specific implementation situation, for example, once per second.

[0044] To demonstrate the operability of this invention, this embodiment provides a quantization calculation process, see [link to documentation]. Figure 2 This is a simplified data set of VCSEL epitaxial wafer growth process data, including the material refractive index at 40 time points. Figure 2 It is evident that the processing involves three types of epitaxial materials; in this embodiment, the most common method is to use... and Taking two time points as examples, we will proceed with the subsequent calculations.

[0045] S02: Determine the growth rate of the epitaxial wafer at the current moment.

[0046] It should be noted that when the epitaxial layer of the epitaxial wafer grows on the substrate, the probe light will be reflected at the top surface of the epitaxial layer and at the interface between the epitaxial layer and the underlying material. These two reflected beams will interfere, and as the layer thickness increases, the interference effect will show a periodic change in intensity, forming an oscillation signal in the time domain. One complete cycle of this oscillation corresponds to the physical thickness of the epitaxial layer. Therefore, by capturing this oscillation cycle, the growth rate of the epitaxial wafer can be calculated.

[0047] The growth rate of the epitaxial wafer at the current moment is determined based on the refractive index and the dominant frequency of the refractive index data sequence corresponding to the preset time window to which the current moment belongs.

[0048] The implementers can set the size of the time window according to the specific implementation situation. For example, a time window containing 60 moments with the current moment as the last moment can use the first 60 moments initially collected as pre-collected data, which is only used to build the time window for subsequent monitoring and is not used for subsequent calculations. In this embodiment, the size of the time window is set to 5 for the convenience of calculation.

[0049] Specifically, the main frequency of the refractive index data sequence is obtained as follows:

[0050] Based on the refractive index data of the materials used in the epitaxial wafers at all times within the preset time window to which the current time belongs, a refractive index data sequence corresponding to the time window to which the current time belongs is constructed. The dominant frequency of the refractive index data sequence is then identified using a Fast Fourier Transform algorithm. (See also...) Figure 3 This represents the dominant frequency of the refractive index data sequence corresponding to the time window at each of the 40 time points obtained using the Fast Fourier Transform algorithm.

[0051] Specifically, the growth rate satisfies:

[0052] ;

[0053] In the formula, This represents the growth rate of the epitaxial wafer at the current moment. The wavelength of the light source used in the spectral sensing equipment of the MOCVD machine. The refractive index of the material used in the epitaxial wafer at the current moment. The dominant frequency of the refractive index data sequence corresponding to the preset time window at the current moment.

[0054] in, The increase in physical thickness of the epitaxial layer of the epitaxial wafer, which reflects the oscillation of the refractive index data sequence within a preset time window at the current moment for one complete cycle, is calculated by dividing the increase in thickness within one cycle by the time taken for one complete cycle. The growth rate of the epitaxial wafer at the current moment, i.e. The larger the value, the thicker the epitaxial membrane can be grown at the current moment; conversely, the smaller the value, the less likely it is to grow.

[0055] For example, calculation and The growth rates of the epitaxial wafer at the two time points are as follows:

[0056] ;

[0057] ;

[0058] See Figure 4 The growth rate of the epitaxial wafer at 40 time points, from Figure 2 It can be seen The difference in growth rate caused by changes in the epitaxial wafer material is a benign abrupt change in the process, while Figure 4 There was a certain deviation, so the following calculation steps need to be continued.

[0059] S03: Determine the abrupt change in the growth rate of the epitaxial wafer at the current moment.

[0060] It should be noted that any instability in the growth environment will ultimately manifest in changes in the growth rate. To ensure quality, the growth process of a normal epitaxial wafer needs to be stable, and its growth rate should also be stable. If the growth rate jumps significantly beyond its recent average level, it indicates that the current epitaxial wafer growth parameters may be drifting, leading to a breakdown in growth stability, deviation of the epitaxial wafer thickness from the preset target value, and disruption of thickness uniformity, thus failing to meet the accuracy requirements of the epitaxial layer design. Therefore, by comparing the changes in the growth rate of the epitaxial wafer at the current moment with those at other moments within the preset window, the degree of abrupt change in the growth rate of the epitaxial wafer at the current moment can be evaluated.

[0061] The abrupt change rate of the epitaxial wafer's growth rate at the current moment is determined based on the instantaneous fluctuation and continuous change trend of the growth rate within the preset time window to which the current moment belongs.

[0062] Specifically, obtain the feature points corresponding to all times within the preset time window to which the current time belongs:

[0063] Using the temporal sequence of all moments within the preset time window to which the current moment belongs as the horizontal axis and the growth rate of the epitaxial wafer at all moments within the preset time window to which the current moment belongs as the vertical axis, feature points corresponding to all moments within the preset time window to which the current moment belongs are constructed.

[0064] Specifically, the degree of mutation satisfies:

[0065] ;

[0066] In the formula, The abrupt change in the growth rate of the epitaxial wafer at the current moment. This represents the growth rate of the epitaxial wafer at the current moment. The first of the remaining times within the preset time window that does not include the current time. The growth rate of the epitaxial wafer at a given time point. This represents the number of times within the preset time window to which the current time belongs. The feature point corresponding to the first moment within the preset time window to which the current moment belongs. The feature point corresponding to the current time. for and The slope between For the standard normalized function, It is the absolute value symbol.

[0067] in, This value reflects the difference between the growth rate of the epitaxial wafer at the current moment and other moments within its preset time window. A larger value indicates a greater difference between the current growth rate and other growth rates within the preset time window, suggesting a drastic instantaneous fluctuation or jump in the growth process at that moment. Considering the existence of continuous and unstable segments in the actual growth rate, this leads to… The value may decrease, making it insensitive to continuous unstable segments. Therefore, the slope between the two furthest feature points within the preset time window to which the current moment belongs is introduced. That is, the slope between the first moment and the current moment reflects the overall trend of the growth rate within the preset time window to which the current moment belongs. The larger this value is, the more significant and continuous unstable the growth rate shows throughout the time window, and the greater the possibility of a sudden change in the growth rate of the epitaxial wafer at the current moment. In summary, if The larger the value and The larger the value, the greater the abrupt change in the growth rate of the epitaxial wafer at the current moment, thus avoiding the possibility of missed judgments in the evaluation of a single index.

[0068] For example, calculation and The abrupt change in the growth rate of the epitaxial wafer at the two time points is as follows:

[0069] The growth rate data of all epitaxial wafers within the preset time window are as follows: ,but The feature point corresponding to the first moment within the preset time window is , The feature point corresponding to time is ;

[0070] The first part of the relation: ;

[0071] The second part of the relation: ;

[0072] but .

[0073] for The specific data regarding the time will not be elaborated upon here;

[0074] The first part of the relation: ;

[0075] The second part of the relation: ;

[0076] but .

[0077] S04: Determine the correction abruptness of the growth rate of the epitaxial wafer at the current moment.

[0078] It should be noted that, considering the multi-layer material changes during the epitaxial growth process within the MOCVD equipment cavity, the degree of change in growth rate corresponding to benign mutations caused by material switching in the normal process can be quite large, leading to frequent invalid alarms issued by the system during normal operation. Therefore, it is necessary to distinguish between benign mutations and abnormal mutations. Benign mutations are usually caused by material differences. Thus, by constructing a correction factor that can identify changes in multiple different layers, benign mutations can be suppressed, thereby reducing false alarms.

[0079] Based on the change in the standard deviation of the refractive index within the preset time window at the current moment, the abrupt change is corrected to obtain the corrected abrupt change in the growth rate of the epitaxial wafer at the current moment.

[0080] Specifically, the modified mutation degree satisfies:

[0081] ;

[0082] In the formula, This represents the corrected abrupt change in the growth rate of the epitaxial wafer at the current moment. The abrupt change in the growth rate of the epitaxial wafer at the current moment. The standard deviation of the refractive index of the material used in the epitaxial wafer at all times within the preset time window to which the current time belongs. This represents the standard deviation of the refractive index of the material used in the epitaxial wafer at all times within the preset time window preceding the current time. For hyperparameters, It is a function with maximum value. It is a natural exponential function. It is an absolute value function.

[0083] Implementers can set hyperparameters according to the specific implementation situation, for example, 0.001. The existence of hyperparameters is to prevent... This can lead to situations where the formula becomes meaningless.

[0084] in, A correction factor reflecting the abrupt change in the growth rate of the epitaxial wafer at the current moment. The closer this value is to 0, the more stable the refractive index change of the epitaxial wafer growing within the preset window at the current moment. This means that the abrupt change in the growth rate of the epitaxial wafer at the current moment is unrelated to material switching and is more likely to be a genuine process anomaly. Therefore, for those who do not suppress the original abrupt change, i.e. The closer it gets to 0, The closer the value is to 1, the more it maintains or approaches the original abrupt change rate; conversely, the closer the value is to 1, the more it indicates that the material type has changed significantly within the preset time window at the current moment. This means that the abrupt change in the growth rate of the epitaxial wafer at the current moment is likely caused by a change in process materials, and therefore, its abrupt change rate needs to be suppressed. The closer it gets to 1, Approaching 0 reduces the abrupt change in the growth rate of the epitaxial wafer at the current moment.

[0085] For example, calculation and The corrected abrupt change in the growth rate of the epitaxial wafer at the two time points is as follows:

[0086] The refractive index data of the materials used in the epitaxial wafers at all times within the preset time window are as follows: The corresponding standard deviation is ;

[0087] If the refractive index data of the material used for the epitaxial wafer at all times within the preset time window are {3.795, 3.795, 3.802, 3.781, 3.783}, then the standard deviation is... ;

[0088] Correction factors in the relation: ;

[0089] but .

[0090] for The specific data regarding the time will not be elaborated upon here;

[0091] Correction factors in the relation: ;

[0092] but .

[0093] S05: Determine the warning threshold for the current moment.

[0094] It should be noted that through the above steps, the value of benign mutations will be significantly reduced, while the corresponding changes after correction for normal and abnormal mutations are not significant. Therefore, by analyzing the changes in mutation values ​​before and after correction at multiple time points in real time, dynamic clustering with a preset cluster size of 2 is performed. The maximum value of the corrected mutation degree of the epitaxial wafer growth rate at the corresponding time point is extracted from the data points within the clusters with a small number of clusters (since benign mutations are often located at the material exchange boundary of the layer, the number is small) as the warning threshold, thereby avoiding the unreliability of a globally fixed threshold.

[0095] Cluster the difference between the abrupt change in the growth rate of the epitaxial wafer at the current moment and the corrected abrupt change in the growth rate of the epitaxial wafer at each moment, and determine the warning threshold at the current moment.

[0096] Specifically, determining the warning threshold at the current moment includes:

[0097] Based on the aforementioned differences, a feature dataset is constructed;

[0098] Clustering algorithms, such as k-means clustering, are used to cluster feature datasets, with a preset... This yields two clusters;

[0099] The cluster with the fewest data points among two clusters is denoted as the suspected benign mutation cluster;

[0100] The maximum value of the corrected mutation degree corresponding to all data points within the suspected benign mutation cluster is determined as the warning threshold at the current moment.

[0101] like Figure 5 As shown in the figure, both the original mutation rate and the revised mutation rate are plotted. The original mutation rate is highly sensitive to all changes. It can be seen that, regardless of... and A benign mutation, or in and At the locations of anomalous mutations, very high peak values ​​were observed, indicating that it can effectively detect changes but cannot distinguish the nature of those changes; adjusting for mutation degree reveals that... and At this point, because the invention detected a synchronous change in the material's refractive index, the purple peak was successfully suppressed, resulting in a lower value for the green line. and Since the refractive index of the material is stable, this invention does not suppress it, so the green line retains the original peak, thus distinguishing between benign and aberrant mutations.

[0102] For example, calculation and The warning thresholds at the two time points are as follows:

[0103] right Clustering is performed on the differences between the given time and all previous historical times. The stable clustering results are as follows: Figure 6 As shown, the purple cluster is located on the left, with an eigenvalue close to 0, which represents the value from the beginning to the end. All data points with little difference in mutation degree before and after correction are considered to be non-benign mutations; the yellow cluster on the right represents data points with larger eigenvalues, corresponding to the current analysis. The time interval (marked by a purple asterisk) is significantly corrected because it represents a benign mutation, resulting in a higher difference in mutation degree before and after correction, distinguishing it from previous data points. Since the number of data points in the yellow cluster is far less than that in the purple cluster, the yellow cluster is identified as a suspected benign mutation cluster. At this point, the dynamic threshold will be set based on the data points within this cluster. Value to set, as The warning threshold at any given time.

[0104] for At any given time, the stable clustering results are as follows: Figure 7 As shown, the current analysis Time (marked by a purple asterisk), due to The time point is an unsuppressed outlier. and The values ​​are all high, so the eigenvalues ​​are very low. Therefore, It was correctly classified into the yellow normal / abnormal cluster, but not into the purple benign cluster; when obtaining the warning threshold, the maximum value of the corrected mutation degree within the benign cluster was used as the threshold, which can clearly identify abnormal mutations.

[0105] S06: Based on the magnitude of the modified mutation degree and the warning threshold, real-time monitoring and early warning of VCSEL epitaxial wafer growth process can be achieved.

[0106] Specifically, the method for real-time data monitoring and early warning of the VCSEL epitaxial wafer growth process includes:

[0107] If the corrected mutation rate is greater than the warning threshold, it indicates that the growth process of the epitaxial wafer has deviated from the range of normal fluctuations and benign mutations at the current moment. This suggests that the current growth process parameters may be unstable, which may lead to uneven epitaxial wafer thickness and a decrease in the final device performance. Therefore, it is determined that there is a risk in the epitaxial wafer growth process at the current moment, and a warning is triggered to prompt relevant personnel to conduct an inspection and complete the real-time data monitoring and warning of the VCSEL epitaxial wafer growth process.

[0108] like Figure 8 As shown in the figure, this is the final warning recognition diagram, which shows the data points of the corrected abrupt change degree, the dynamically generated warning threshold, and the triggered alarm. and None of the benign mutations exceeded their corresponding dynamically set warning thresholds, and no false alarms were generated. and Abnormal mutations, The values ​​exceeded their own warning thresholds, so the system triggered alarms at these two data points, thereby monitoring the actual process anomalies.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer, characterized in that, include: The refractive index of the material used in the VCSEL epitaxial wafer at the current moment during the VCSEL epitaxial wafer growth process was obtained using an MOCVD device. The growth rate of the epitaxial wafer at the current moment is determined based on the refractive index and the dominant frequency of the refractive index data sequence corresponding to the preset time window to which the current moment belongs. Based on the instantaneous fluctuations and continuous trends of the growth rate within the preset time window at the current moment, the abrupt change degree of the growth rate of the epitaxial wafer at the current moment is determined; the abrupt change degree satisfies: In the formula, The abrupt change in the growth rate of the epitaxial wafer at the current moment. This represents the growth rate of the epitaxial wafer at the current moment. The first of the remaining times within the preset time window that does not include the current time. The growth rate of the epitaxial wafer at a given time point. This represents the number of times within the preset time window to which the current time belongs. The feature point corresponding to the first moment within the preset time window to which the current moment belongs. The feature point corresponding to the current time. for and The slope between For the standard normalized function, The absolute value symbol is used. The feature points are obtained by using the time sequence of all times within the preset time window to which the current time belongs as the horizontal axis and the growth rate of the epitaxial wafer at all times within the preset time window to which the current time belongs as the vertical axis to construct the feature points corresponding to all times within the preset time window to which the current time belongs. Based on the change in the standard deviation of the refractive index within the preset time window at the current moment, the abrupt change is corrected to obtain the corrected abrupt change in the growth rate of the epitaxial wafer at the current moment. Cluster the difference between the abrupt change in the growth rate of the epitaxial wafer at the current moment and the corrected abrupt change in the growth rate of the epitaxial wafer at each moment, and determine the warning threshold at the current moment. By adjusting the degree of mutation and the warning threshold, real-time monitoring and early warning of VCSEL epitaxial wafer growth process can be achieved.

2. The real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 1, characterized in that, The growth rate satisfies: ； In the formula, This represents the growth rate of the epitaxial wafer at the current moment. The wavelength of the light source used in the spectral sensing equipment of the MOCVD machine. The refractive index of the material used in the epitaxial wafer at the current moment. The dominant frequency of the refractive index data sequence corresponding to the preset time window at the current moment.

3. A real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 1 or 2, characterized in that, The method for obtaining the dominant frequency of the refractive index data sequence is as follows: Based on the refractive index data of the materials used in the epitaxial wafers at all times within the preset time window to which the current time belongs, a refractive index data sequence corresponding to the time window to which the current time belongs is constructed. The main frequency of the refractive index data sequence is identified by using the fast Fourier transform algorithm.

4. The real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 1, characterized in that, The corrected mutation degree satisfies: ； In the formula, This represents the corrected abrupt change in the growth rate of the epitaxial wafer at the current moment. The abrupt change in the growth rate of the epitaxial wafer at the current moment. The standard deviation of the refractive index of the material used in the epitaxial wafer at all times within the preset time window to which the current time belongs. This represents the standard deviation of the refractive index of the material used in the epitaxial wafer at all times within the preset time window preceding the current time. For hyperparameters, It is a function with maximum value. It is a natural exponential function. It is an absolute value function.

5. The real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 1, characterized in that, The preset number of clusters for the clustering is 2.

6. The real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 5, characterized in that, Determining the warning threshold at the current moment includes: Based on the aforementioned differences, a feature dataset is constructed; Clustering algorithms are used to cluster the feature dataset, resulting in two clusters; The cluster with the fewest data points among two clusters is denoted as the suspected benign mutation cluster; The maximum value of the corrected mutation degree corresponding to all data points within the suspected benign mutation cluster is determined as the warning threshold at the current moment.

7. A real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 1 or 6, characterized in that, The clustering was performed using the k-means clustering algorithm.

8. The real-time data monitoring and early warning method for the growth process of a VCSEL epitaxial wafer according to claim 1, characterized in that, The method for real-time monitoring and early warning of VCSEL epitaxial wafer growth process includes: In response to the correction mutation degree being greater than the warning threshold, it is determined that there is a risk in the epitaxial wafer growth process at the current moment, and a warning is triggered to complete the real-time data monitoring and warning of the VCSEL epitaxial wafer growth process.

Images