Deep learning-based semiconductor material processing monitoring method
By quantifying the influence of microcrack structure on electric field distribution using deep learning methods, a coupling constraint relationship between microcrack and microcavity migration behavior is constructed, solving the path offset problem during microcavity migration and improving the stability and controllability of semiconductor processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- QINGDAO HUAXIN JINGDIAN TECH CO LTD
- Filing Date
- 2026-05-07
- Publication Date
- 2026-07-14
Smart Images
Figure CN122135854B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor processing technology, and in particular to a semiconductor material processing monitoring method based on deep learning. Background Technology
[0002] Microcavity structures within semiconductor materials have significant application value in the fabrication of microelectronic devices, optoelectronic devices, and high-precision functional materials. Current technologies typically employ high-energy beams such as femtosecond lasers to directly induce the formation of microcavity structures within semiconductor materials. Compared to traditional layer-by-layer stacking bonding processes, laser processing directly constructs microcavities within the bulk material, fundamentally avoiding multilayer alignment errors and misalignment defects caused by interfacial thermal mismatch, thus ensuring the integrity and hermeticity of the structure.
[0003] However, in actual laser processing, laser energy is highly concentrated in local areas, which not only forms microcavity structures in the target processing area, but also induces microcrack propagation and stress concentration effects in the surrounding areas, thereby causing additional thermal damage and structural destruction to non-target areas.
[0004] To mitigate the side effects of laser processing, existing technologies have introduced an electric field-induced microcavity migration mechanism. This mechanism uses an external electric field to drive the directional movement of a pre-formed microcavity within a semiconductor material, enabling the reconstruction and optimization of the microcavity within the target functional region. However, current microcavity migration processes typically optimize only the electric field driving conditions. In the electric field-induced microcavity migration process, the microcracks formed during laser processing not only exhibit clear spatial distribution characteristics and directionality but also generate significant stress concentration and localized electric field enhancement effects at the crack tips. This further affects the uniformity of the applied electric field distribution within the material, leading to electric field distortion. Simultaneously, the spatial location and extension direction of the microcracks constrain the microcavity migration path, causing the microcavity to deviate during migration.
[0005] Therefore, existing technologies neglect the potential impact of microcrack structures formed during the laser processing stage on subsequent electric field distribution and microcavity migration behavior, making it difficult to predict path deviation risks during microcavity migration and limiting the stability and controllability of high-precision semiconductor microstructure processing. Summary of the Invention
[0006] To overcome the defects and shortcomings of existing technologies, this invention provides a semiconductor material processing monitoring method based on deep learning. By quantifying the potential influence of microcrack structure on the electric field distribution during microcavity migration, it effectively improves the stability and controllability of semiconductor microcavity structure processing.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] In a first aspect, the present invention provides a semiconductor material processing monitoring method based on deep learning, comprising the following steps: S100, acquiring spatial distribution data of microcracks and stress response distribution data generated during the formation of a microcavity in a semiconductor material via laser processing, and simultaneously acquiring electric field distribution data driving the microcavity to migrate; S200, extracting the spatial distribution features of microcracks from the microcrack spatial distribution data, and analyzing the stress concentration degree at the microcrack tip in conjunction with the stress response distribution data; S300, extracting the electric field intensity distribution features from the electric field distribution data, and combining the microcrack spatial distribution features and the stress concentration degree at the microcrack tip... The following steps are taken: S400: Analyze the distortion effect of microcracks on the local electric field distribution and the electric field enhancement effect at the microcrack tip to determine the local field strength shift caused by electric field distortion; S500: Based on the angle between the microcrack extension direction and the microcavity migration direction, the distance between the microcavity migration path and the microcrack tip, and the local field strength shift caused by electric field distortion, construct the coupling constraint relationship between the microcrack structural features and the microcavity migration behavior; S500: Input the coupling constraint relationship into a deep learning network for nonlinear mapping to identify path deviation risks during the microcavity migration process and provide path deviation risk warnings.
[0009] Furthermore, the specific steps of step S200 include:
[0010] S210. Extract the spatial location of microcracks, the location of microcrack tips, and the direction of microcrack extension from the microcrack spatial distribution data as the spatial distribution characteristics of microcracks.
[0011] S220. Calculate the mean stress response distribution inside the semiconductor using stress response distribution data, and extract the tip stress response value corresponding to the microcrack tip location. Use the ratio of the tip stress response value to the mean stress response distribution as the stress concentration degree to characterize the microcrack tip field enhancement characteristics.
[0012] Further, the extraction of the microcrack spatial location, microcrack tip location, and microcrack extension direction in step S210 includes: performing skeletonization processing on the microcrack contour and taking the principal axis direction of the fitted crack skeleton line as the microcrack extension direction; extracting the pixel spatial coordinates on the crack skeleton line to obtain the microcrack spatial location; and performing endpoint detection on the microcrack skeleton line and taking the spatial coordinates corresponding to the endpoint pixels of the microcrack skeleton line as the microcrack tip location.
[0013] Furthermore, the specific steps of step S300 include:
[0014] S310. Establish a three-dimensional spatial coordinate system with the geometric center of the semiconductor material as the origin, and map the electric field distribution data, the spatial location of the microcrack, and the location of the microcrack tip to the three-dimensional spatial coordinate system.
[0015] S320. Perform spatial discretization processing on the electric field distribution data, construct the electric field intensity distribution spatial field corresponding to the three-dimensional spatial coordinate system, and extract the electric field intensity value of each spatial sampling point;
[0016] S330. The microcrack coverage area and the microcrack tip coverage area are determined by the spatial distribution characteristics of microcracks, and the mean electric field intensity in the microcrack coverage area and the extreme electric field intensity in the microcrack tip coverage area are extracted by the spatial field of electric field intensity distribution.
[0017] S340. Using the average electric field intensity in the region without microcracks as the reference electric field intensity, calculate the difference between the average electric field intensity in the microcrack-covered region and the reference electric field intensity to obtain the electric field intensity offset, which is used to characterize the distortion effect of microcracks on the local electric field distribution.
[0018] S350. The ratio of the electric field extreme value offset in the area covered by the microcrack tip to the reference electric field strength is used as the electric field strength gain coefficient, wherein the electric field extreme value offset is the difference between the electric field strength extreme value in the area covered by the microcrack tip and the reference electric field strength.
[0019] Further, the determination of the microcrack coverage area and the microcrack tip coverage area in step S330 includes: taking the spatial location of the microcrack as the microcrack coverage area, and taking the stress concentration degree corresponding to the microcrack tip location as a range adjustment factor, and determining the microcrack tip coverage area in combination with the microcrack tip location.
[0020] Furthermore, the specific steps of step S400 include:
[0021] S410. In a three-dimensional spatial coordinate system, by vectorizing the microcavity migration path and the propagation direction of each microcrack, the angle between the microcavity migration direction vector and the propagation direction vector of each microcrack is calculated to obtain the angle constraint parameter set.
[0022] S420. By calculating the minimum Euclidean space distance between the microcavity migration path and the position of each microcrack tip, the distance constraint parameter set is obtained.
[0023] S430. Along the microcavity migration path, the electric field intensity offset and electric field intensity gain coefficient of each spatial sampling point are statistically analyzed to obtain the electric field distortion constraint parameter set.
[0024] S440. Normalize the angle constraint parameter set, distance constraint parameter set, and electric field distortion constraint parameter set to construct a coupled constraint relationship containing direction constraint term, distance constraint term, and electric field distortion constraint term, which is used to characterize the coupling relationship between microcrack structure features and microcavity migration behavior.
[0025] Furthermore, the specific steps of step S500 include:
[0026] S510. By performing feature encoding on the direction constraint term, distance constraint term, and electric field distortion constraint term in the coupling constraint relationship, a multi-dimensional feature vector is constructed, and the multi-dimensional feature vector is arranged according to the microcavity migration path sequence to generate input sample data.
[0027] S520. Construct a deep learning network that includes an input layer, a feature extraction layer, and a risk discrimination layer, and train the deep learning network using historical microcavity migration process data, wherein the historical microcavity migration process data includes labeled path offsets and corresponding risk level labels.
[0028] S530. Input the input sample data into the trained deep learning network, perform nonlinear mapping on the multidimensional feature vector through the feature extraction layer, and output the corresponding path offset risk probability value in the risk discrimination layer.
[0029] S540. Based on the comparison between the path deviation risk probability value and the preset risk threshold, if the path deviation risk probability value is less than or equal to the preset risk threshold, no path deviation risk warning is issued; if the path deviation risk probability value is greater than the preset risk threshold, a path deviation risk warning is issued, and the path deviation risk probability value is mapped to the microcavity migration path in the three-dimensional spatial coordinate system to generate corresponding risk distribution information for visual warning output of the microcavity migration process.
[0030] In a second aspect, the present invention provides an electronic device, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor, and the processor executes a deep learning-based semiconductor material processing monitoring method by calling the computer program stored in the memory.
[0031] Thirdly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform a deep learning-based semiconductor material processing monitoring method.
[0032] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0033] 1. This invention establishes a quantitative correlation between microcrack structure and local electric field distortion, and further combines the stress concentration effect at the microcrack tip to modify the electric field enhancement behavior, thereby effectively improving the process controllability and consistency of the entire process of microcavity fabrication and migration in semiconductor materials.
[0034] 2. This invention constructs a multidimensional coupled constraint relationship that includes directional constraints, distance constraints, and electric field distortion constraints, and introduces a deep learning network to perform nonlinear mapping and prediction of microcavity migration path offset risks. This enables risk identification and early warning of the microcavity migration process, thereby reducing the processing cost and scrap rate of semiconductor micro / nano structure devices. Attached Figure Description
[0035] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:
[0036] Figure 1 This is a flowchart illustrating the semiconductor material processing monitoring method based on deep learning provided in an embodiment of the present invention.
[0037] Figure 2 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0038] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0039] Please see Figure 1 , Figure 1 This is a schematic diagram of the overall process of the semiconductor material processing monitoring method based on deep learning provided in this embodiment of the invention, which specifically includes the following steps:
[0040] S100. Acquire spatial distribution data of microcracks and stress response distribution data generated during the formation of microcavities in semiconductor materials via laser processing. Simultaneously acquire electric field distribution data that drives the microcavities to migrate. The spatial distribution data of microcracks can be acquired using a scanning laser infrared thermal imaging system. The stress response distribution data can be acquired using Raman spectroscopy stress measurement technology or photoluminescence (PL) stress characterization technology. The electric field distribution data is acquired by an electric field sensor with a sampling frequency of not less than 100Hz, an electric field strength measurement range of 0-1000V / m, and a measurement accuracy of ±1V / m.
[0041] S200. Extract the spatial distribution characteristics of microcracks from the microcrack spatial distribution data, and analyze the stress concentration at the microcrack tip by combining the stress response distribution data.
[0042] Microcracks, as defect structures within semiconductor materials, directly alter the normal distribution of the electric field within the semiconductor material, causing localized electric field distortion. Microcracks themselves are dielectric defects, disrupting the uniformity of the electric field within the semiconductor material. When an electric field acts on the semiconductor material, the electric field lines in the microcrack-covered area bend, converge, or diverge, leading to abnormal fluctuations in the local electric field intensity. Simultaneously, the extension direction of the microcrack affects the direction of the electric field distortion. Microcracks parallel to the electric field direction exacerbate the convergence of electric field lines, while those perpendicular to the direction hinder the transmission of electric field lines. The spatial location of the microcrack determines the extent of the electric field distortion. Electric field distortion caused by microcracks near the microcavity migration path directly affects the microcavity migration trajectory, increasing the risk of path deviation. Therefore, the spatial distribution characteristics of microcracks directly determine the degree and spatial distribution pattern of the electric field distortion. The specific steps of step S200 include:
[0043] S210. By extracting the spatial location, tip location, and extension direction of microcracks from the microcrack spatial distribution data, these can be used as spatial distribution characteristics of microcracks. This provides a quantitative basis for subsequent analysis of the distortion effect of microcracks on the electric field distribution, calculation of the angle between the microcavity migration direction and the crack direction, and assessment of the distance risk between the microcavity and the crack tip.
[0044] S220. Calculate the mean stress response distribution inside the semiconductor using stress response distribution data, and extract the tip stress response value corresponding to the microcrack tip location. The ratio of the tip stress response value to the mean stress response distribution is used as the stress concentration degree to characterize the field enhancement characteristics of the microcrack tip. The stress concentration degree at the microcrack tip will significantly enhance the local electric field strength, producing an electric field enhancement effect, which in turn exacerbates the distortion of the electric field distribution. The stress concentration at the microcrack tip forms a local stress field. This stress field changes the dielectric constant distribution of the semiconductor material, causing changes in the dielectric properties of the tip region. This results in a large number of electric field lines converging in the tip region, causing a sharp increase in electric field strength and forming an electric field enhancement effect. The greater the stress concentration degree, the stronger the tip stress field, the more significant the change in dielectric properties, the more obvious the electric field enhancement effect, and the more severe the electric field distortion. Therefore, the electric field enhancement effect at the microcrack tip will directly affect the electric field driving force during the microcavity migration process, causing the microcavity migration trajectory to deviate.
[0045] Step S210 extracts the spatial location, tip location, and extension direction of the microcrack, including: skeletonizing the microcrack contour and using the principal axis of the fitted crack skeleton line as the extension direction; extracting the pixel spatial coordinates on the crack skeleton line to obtain the spatial location of the microcrack; and detecting the endpoints of the microcrack skeleton line and using the spatial coordinates corresponding to the endpoint pixels as the tip location of the microcrack. Specifically, the Zhang-Suen skeleton extraction algorithm is used to iteratively refine the microcrack contour until the crack region converges to a crack skeleton of single pixel width. The process involves: 1) Laying a linear model of the crack skeleton line using the least squares method; 2) Determining the direction of the principal axis of the fitted line as the microcrack propagation direction; 3) Extracting the three-dimensional spatial coordinates of all pixels on the crack skeleton line point by point; 4) Using an endpoint detection algorithm based on neighborhood voxel counting to identify the endpoints of the crack skeleton line, including checking the number of skeleton voxels in the 26-neighborhood of each voxel on the skeleton line. If the number of neighboring skeleton voxels of a voxel is equal to 1, then the voxel is determined to be the endpoint of the skeleton line, and the three-dimensional spatial coordinates corresponding to the endpoint pixel are taken as the microcrack tip position.
[0046] S300. Extract the electric field intensity distribution characteristics from the electric field distribution data, combine the spatial distribution characteristics of microcracks and the stress concentration degree at the microcrack tip, analyze the distortion effect of microcracks on the local electric field distribution and the electric field enhancement effect at the microcrack tip, and determine the local field intensity shift caused by electric field distortion.
[0047] By constructing a unified three-dimensional spatial coordinate system, the electric field intensity distribution is transformed from an independent physical quantity into a coupled response quantity affected by the cracked structure. This allows the electric field distribution to not only reflect the externally applied driving field but also simultaneously embody the medium discontinuity effect and tip reinforcement effect caused by the microcrack. By quantifying the degree of electric field distortion caused by the microcrack and the tip reinforcement effect, the local field intensity shift caused by the structural defect is determined, providing key physical inputs and constraints for subsequent coupling constraint relationship construction and path offset modeling. The specific steps of step S300 include:
[0048] S310. Establish a three-dimensional spatial coordinate system with the geometric center of the semiconductor material as the origin, and map the electric field distribution data, the spatial location of the microcrack, and the location of the microcrack tip into the three-dimensional spatial coordinate system. Specifically, establish a three-dimensional spatial coordinate system with the geometric center of the semiconductor material as the origin, where the x-axis is parallel to the length direction of the semiconductor material, the y-axis is parallel to the width direction of the semiconductor material, and the z-axis is parallel to the thickness direction of the semiconductor material (i.e., perpendicular to the material surface). Extract the physical location coordinates and corresponding electric field intensity values of each sampling point in the electric field distribution data. Convert the physical location coordinates of each sampling point into coordinates in the three-dimensional spatial coordinate system, while retaining the corresponding electric field intensity values, forming a three-dimensional coordinate-electric field intensity mapping dataset. Similarly, convert the physical coordinates of the microcrack spatial location and the microcrack tip location into coordinates in the three-dimensional spatial coordinate system.
[0049] S320. Spatial discretization processing is performed on the electric field distribution data to construct a spatial field of electric field intensity distribution corresponding to a three-dimensional spatial coordinate system. The electric field intensity value of each spatial sampling point is extracted. Specifically, based on the three-dimensional spatial coordinate system established in step S310, uniformly distributed three-dimensional spatial sampling points are generated according to a preset sampling step size (e.g., 0.05μm). For each spatial sampling point, if the sampling point is the original sampling point of the electric field distribution data, its corresponding electric field intensity value is directly extracted; if the sampling point is not the original sampling point, the Kriging interpolation method (interpolation accuracy not less than 95%) is used to take the average electric field intensity value of the 5 nearest original electric field sampling points around the sampling point as the electric field intensity value of the sampling point, ensuring that each spatial sampling point has a corresponding electric field intensity value.
[0050] S330. The microcrack coverage area and the microcrack tip coverage area are determined by the spatial distribution characteristics of microcracks, and the mean electric field intensity in the microcrack coverage area and the extreme electric field intensity in the microcrack tip coverage area are extracted by the spatial field of electric field intensity distribution.
[0051] S340. Using the average electric field intensity in the region without microcracks as the reference electric field intensity, calculate the difference between the average electric field intensity in the microcrack-covered region and the reference electric field intensity to obtain the electric field intensity offset, which is used to characterize the distortion effect of microcracks on the local electric field distribution.
[0052] S350. The ratio of the electric field extreme value offset in the area covered by the microcrack tip to the reference electric field strength is used as the electric field strength gain coefficient, wherein the electric field extreme value offset is the difference between the electric field strength extreme value in the area covered by the microcrack tip and the reference electric field strength.
[0053] Step S330, determining the microcrack coverage area and the microcrack tip coverage area, includes: using the spatial location of the microcrack as the microcrack coverage area, and using the stress concentration degree corresponding to the microcrack tip location as a range adjustment factor; combining the microcrack tip location to determine the microcrack tip coverage area. Specifically, the microcavity radius is used as the reference radius of the microcrack tip's influence range, the stress concentration degree corresponding to the microcrack tip location is used as the range adjustment factor, and the product of the reference radius and the range adjustment factor is used as the actual radius of the microcrack tip coverage area; a three-dimensional spherical region is constructed with the microcrack tip location as the center and the actual radius as the sphere radius, and this three-dimensional spherical region is the microcrack tip coverage area.
[0054] S400. Based on the angle relationship between the microcrack extension direction and the microcavity migration direction, the distance relationship between the microcavity migration path and the microcrack tip, and the local field strength shift caused by electric field distortion, construct the coupling constraint relationship between the microcrack structural features and the microcavity migration behavior.
[0055] By uniformly modeling the spatial relationship between microcrack structural features and microcavity migration paths, as well as the local field strength changes caused by electric field distortion, a coupled constraint relationship is constructed that integrates directional constraints, distance constraints, and electric field distortion constraints. This unifies the anisotropic effects induced by cracks, the near-field interference effect at the tip, and the non-uniform distortion effect of the electric field into the same constraint space, providing input features with clear physical meaning for subsequent deep learning networks. The specific steps of step S400 include:
[0056] S410. In the three-dimensional spatial coordinate system, by vectorizing the microcavity migration path and the propagation direction of each microcrack, the angle between the microcavity migration direction vector and the propagation direction vector of each microcrack is calculated to obtain the angle constraint parameter set. Specifically, based on the three-dimensional spatial coordinate system established in S310, a continuous spatial coordinate sequence of the microcavity migration path is obtained as the microcavity migration path sequence; the angle between the microcavity migration direction vector and the propagation direction vector of each migration path point in the microcavity migration path sequence is traversed to obtain the angle constraint parameter set.
[0057] S420. By calculating the minimum Euclidean space distance between the microcavity migration path and the position of each microcrack tip, the distance constraint parameter set is obtained.
[0058] S430. Along the microcavity migration path, the electric field intensity offset and electric field intensity gain coefficient of each spatial sampling point are statistically analyzed to obtain the electric field distortion constraint parameter set.
[0059] S440. Normalize the angle constraint parameter set, distance constraint parameter set, and electric field distortion constraint parameter set to construct a coupled constraint relationship containing direction constraint terms, distance constraint terms, and electric field distortion constraint terms. This relationship is used to characterize the coupling relationship between microcrack structure features and microcavity migration behavior. Specifically, the constraint parameters in the angle constraint parameter set, distance constraint parameter set, and electric field distortion constraint parameter set are respectively subjected to Min-Max normalization so that the value range of each constraint term is unified to [0,1].
[0060] S500: Input the coupling constraint relationship into the deep learning network for nonlinear mapping, identify the path deviation risk in the microcavity migration process, and provide path deviation risk warning.
[0061] By transforming multi-source coupling constraints into structured inputs to a deep learning network, and combining nonlinear mapping to learn the complex mapping relationship between microcrack structural features and microcavity migration path offset, the automatic identification and prediction of potential path offset risks during migration is achieved. The specific steps of step S500 include:
[0062] S510. By performing feature encoding on the direction constraint term, distance constraint term, and electric field distortion constraint term in the coupling constraint relationship, a multi-dimensional feature vector is constructed, and the multi-dimensional feature vector is arranged according to the microcavity migration path sequence to generate input sample data.
[0063] S520. Construct a deep learning network comprising an input layer, a feature extraction layer, and a risk discrimination layer. Train the deep learning network using historical microcavity migration process data. The historical microcavity migration process data includes labeled path offsets and corresponding risk level labels. Specifically, the input layer has 4 neurons, and the ReLU activation function is used to receive and initially process the input features. The feature extraction layer contains 3 hidden layers: the first hidden layer has 32 neurons, the second hidden layer has 16 neurons, and the third hidden layer has 8 neurons, all using the ReLU activation function to progressively extract and non-linearly map the input features and uncover the intrinsic correlation between features. A Dropout layer is used between each hidden layer to prevent overfitting, with a Dropout probability of 0.2. The risk discrimination layer has 1 neuron, and the Sigmoid activation function is used to output the microcavity migration path offset risk probability value (range [0,1]). The closer the probability value is to 1, the higher the path offset risk; the closer it is to 0, the lower the path offset risk.
[0064] S530. Input the input sample data into the trained deep learning network, perform nonlinear mapping on the multidimensional feature vector through the feature extraction layer, and output the corresponding path offset risk probability value in the risk discrimination layer. Specifically, the deep learning network training includes: (1) Data preparation stage: collect at least 100 sets of historical microcavity migration process data as training set. Each set of data contains complete microcavity migration path sampling point feature data, corresponding actual path offset, and manually labeled risk level labels, and is divided into training set, validation set, and test set in a ratio of 7:2:1; (2) Network training parameter settings: set the optimizer to Adam optimizer, set the initial learning rate to 0.001, and adaptively decay with each training round (decay coefficient is 0.95, decaying once every 10 rounds); the loss function adopts the cross-entropy loss function to measure The error between the network output risk probability value and the actual risk level label; the training rounds are set to 100 rounds, the batch size is set to 32, and the early stop strategy is set as follows: if the validation set loss does not decrease for 10 consecutive rounds, then stop training to avoid overfitting; (3) Network training process: input the training set into the constructed deep learning network, perform forward propagation calculation, and obtain the network output risk probability value; calculate the error between the output value and the actual label through the loss function, and use the backpropagation algorithm to update the weights and biases of each layer of the network; after each training round, use the validation set to verify the network performance and adjust the training parameters; after training, use the test set to test the network performance to ensure that the network accuracy is not less than 90% and the recall is not less than 88%. If the requirements are not met, supplement historical data and retrain until the performance requirements are met and the trained deep learning network is obtained;
[0065] S540. Based on the comparison between the path deviation risk probability value and the preset risk threshold, if the path deviation risk probability value is less than or equal to the preset risk threshold, no path deviation risk warning is issued; if the path deviation risk probability value is greater than the preset risk threshold, a path deviation risk warning is issued, and the path deviation risk probability value is mapped to the microcavity migration path in the three-dimensional spatial coordinate system to generate corresponding risk distribution information for visual warning output of the microcavity migration process.
[0066] Please refer to Figure 2 The present invention also provides an electronic device 300, including a memory 310, a processor 320, and a communication bus 330; the memory 310 and the processor 320 are connected via the communication bus 330. The memory 310 stores a semiconductor material processing monitoring method based on deep learning, which can be loaded by the processor 320 and executed as provided in the above embodiments.
[0067] The memory 310 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 310 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the deep learning-based semiconductor material processing monitoring method provided in the above embodiments, etc. The data storage area may store data involved in the deep learning-based semiconductor material processing monitoring method provided in the above embodiments, etc.
[0068] Processor 320 may include one or more processing cores. Processor 320 executes instructions, programs, code sets, or instruction sets stored in memory 310, and calls data stored in memory 310 to perform various functions and process data according to the present invention. Processor 320 may be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that, for different devices, the electronic devices used to implement the functions of processor 320 may also be other types, and the embodiments of the present invention do not specifically limit this.
[0069] The communication bus 330 may include a path for transmitting information between the aforementioned components. The communication bus 330 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 330 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 The symbol is represented by a single double arrow, but this does not mean that there is only one bus or one type of bus.
[0070] This invention provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in the above embodiments, which is a deep learning-based semiconductor material processing monitoring method.
[0071] In this embodiment of the invention, the computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. The computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, the computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), lectern random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory stick, floppy disk, optical disk, magnetic disk, mechanical encoding device, or any combination thereof.
[0072] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0073] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to the technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions claimed in this invention.
Claims
1. A semiconductor material processing monitoring method based on deep learning, characterized in that, Including the following steps: S100: Acquire spatial distribution data of microcracks and stress response distribution data generated during the formation of microcavities in semiconductor materials by laser processing, and simultaneously acquire electric field distribution data that drives the microcavities to migrate. S200. Extract the spatial distribution characteristics of microcracks from the microcrack spatial distribution data, and analyze the stress concentration at the microcrack tip by combining the stress response distribution data. S300. Extract the electric field intensity distribution characteristics from the electric field distribution data, combine the spatial distribution characteristics of microcracks and the stress concentration degree at the microcrack tip, analyze the distortion effect of microcracks on the local electric field distribution and the electric field enhancement effect at the microcrack tip, and determine the local field intensity shift caused by electric field distortion. S400. Based on the angle relationship between the microcrack extension direction and the microcavity migration direction, the distance relationship between the microcavity migration path and the microcrack tip, and the local field strength shift caused by electric field distortion, construct the coupling constraint relationship between the microcrack structural features and the microcavity migration behavior. S500: Input the coupling constraint relationship into the deep learning network for nonlinear mapping, identify the path deviation risk in the microcavity migration process, and provide path deviation risk warning. The specific steps of step S300 include: S310. Establish a three-dimensional spatial coordinate system with the geometric center of the semiconductor material as the origin, and map the electric field distribution data, the spatial location of the microcrack, and the location of the microcrack tip to the three-dimensional spatial coordinate system. S320. Perform spatial discretization processing on the electric field distribution data, construct the electric field intensity distribution spatial field corresponding to the three-dimensional spatial coordinate system, and extract the electric field intensity value of each spatial sampling point; S330. The microcrack coverage area and the microcrack tip coverage area are determined by the spatial distribution characteristics of microcracks, and the mean electric field intensity in the microcrack coverage area and the extreme electric field intensity in the microcrack tip coverage area are extracted by the spatial field of electric field intensity distribution. S340. Using the average electric field intensity in the region without microcracks as the reference electric field intensity, calculate the difference between the average electric field intensity in the microcrack-covered region and the reference electric field intensity to obtain the electric field intensity offset, which is used to characterize the distortion effect of microcracks on the local electric field distribution. S350. The ratio of the electric field extreme value offset in the area covered by the microcrack tip to the reference electric field intensity is used as the electric field intensity gain coefficient, wherein the electric field extreme value offset is the difference between the electric field intensity extreme value in the area covered by the microcrack tip and the reference electric field intensity. The specific steps of step S400 include: S410. In a three-dimensional spatial coordinate system, by vectorizing the microcavity migration path and the propagation direction of each microcrack, the angle between the microcavity migration direction vector and the propagation direction vector of each microcrack is calculated to obtain the angle constraint parameter set. S420. By calculating the minimum Euclidean space distance between the microcavity migration path and the position of each microcrack tip, the distance constraint parameter set is obtained. S430. Along the microcavity migration path, the electric field intensity offset and electric field intensity gain coefficient of each spatial sampling point are statistically analyzed to obtain the electric field distortion constraint parameter set. S440. Normalize the angle constraint parameter set, distance constraint parameter set, and electric field distortion constraint parameter set to construct a coupled constraint relationship containing direction constraint term, distance constraint term, and electric field distortion constraint term, which is used to characterize the coupling relationship between microcrack structure features and microcavity migration behavior.
2. The semiconductor material processing monitoring method based on deep learning according to claim 1, characterized in that, The specific steps of step S200 include: S210. Extract the spatial location of microcracks, the location of microcrack tips, and the direction of microcrack extension from the microcrack spatial distribution data as the spatial distribution characteristics of microcracks. S220. Calculate the mean stress response distribution inside the semiconductor using stress response distribution data, and extract the tip stress response value corresponding to the microcrack tip location. Use the ratio of the tip stress response value to the mean stress response distribution as the stress concentration degree to characterize the microcrack tip field enhancement characteristics.
3. The semiconductor material processing monitoring method based on deep learning according to claim 2, characterized in that, The step S210 of extracting the spatial location of the microcrack, the location of the microcrack tip, and the direction of microcrack extension includes: performing skeletonization processing on the microcrack contour and taking the principal axis direction of the fitted crack skeleton line as the direction of microcrack extension; obtaining the spatial location of the microcrack by extracting the pixel spatial coordinates on the crack skeleton line; and performing endpoint detection on the microcrack skeleton line and taking the spatial coordinates corresponding to the endpoint pixels of the microcrack skeleton line as the location of the microcrack tip.
4. The semiconductor material processing monitoring method based on deep learning according to claim 1, characterized in that, The step S330 of determining the microcrack coverage area and the microcrack tip coverage area includes: taking the spatial location of the microcrack as the microcrack coverage area, taking the stress concentration degree corresponding to the microcrack tip location as a range adjustment factor, and determining the microcrack tip coverage area in combination with the microcrack tip location.
5. The semiconductor material processing monitoring method based on deep learning according to claim 1, characterized in that, The specific steps of step S500 include: S510. By performing feature encoding on the direction constraint term, distance constraint term, and electric field distortion constraint term in the coupling constraint relationship, a multi-dimensional feature vector is constructed, and the multi-dimensional feature vector is arranged according to the microcavity migration path sequence to generate input sample data. S520. Construct a deep learning network that includes an input layer, a feature extraction layer, and a risk discrimination layer, and train the deep learning network using historical microcavity migration process data, wherein the historical microcavity migration process data includes labeled path offsets and corresponding risk level labels. S530. Input the input sample data into the trained deep learning network, perform nonlinear mapping on the multidimensional feature vector through the feature extraction layer, and output the corresponding path offset risk probability value in the risk discrimination layer. S540. Based on the comparison between the path deviation risk probability value and the preset risk threshold, if the path deviation risk probability value is less than or equal to the preset risk threshold, no path deviation risk warning is issued; if the path deviation risk probability value is greater than the preset risk threshold, a path deviation risk warning is issued, and the path deviation risk probability value is mapped to the microcavity migration path in the three-dimensional spatial coordinate system to generate corresponding risk distribution information for visual warning output of the microcavity migration process.
6. An electronic device, comprising: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; characterized in that the processor executes the deep learning-based semiconductor material processing monitoring method as described in any one of claims 1-5 by calling the computer program stored in the memory.
7. A computer-readable storage medium, characterized in that, The device stores instructions that, when executed on a computer, cause the computer to perform the deep learning-based semiconductor material processing monitoring method as described in any one of claims 1-5.