Big data rendering method and wafer detection data rendering system

Abstract

The invention provides a big data rendering method and a wafer detection data rendering system, and belongs to the field of image data processing of semiconductor detection, and the big data rendering method comprises the steps of data source loading analysis, data screening and data de-duplication, drawing as required, and rendering of ten millions of data. The system comprises a data loading module, a coordinate conversion and partitioning module, a data processing module, a node selection module and a rendering module. According to the method, the rendered data volume is effectively reduced through the spatial filtering algorithm and the data downsampling technology, the risk of jamming caused by direct drawing of a large data volume by the UI thread is reduced, meanwhile, the software rendering speed is increased, and popularization and application in the data processing field such as wafer detection are facilitated.

Description

Technical Field

[0001] This invention belongs to the field of image data processing for semiconductor inspection, specifically relating to a big data rendering method and a wafer inspection data rendering system. Background Technology

[0002] In the field of wafer inspection, users create recipes based on wafer inspection software and perform inspections according to these recipes, generating corresponding result files. However, the size of the recipe increases with the number of tests and ROIs (Regions of Interest) it contains, and the number of defects in the inspection results increases as the threshold setting in the tests decreases. Furthermore, in actual work, recipes often contain over 1 million ROIs, and result files occasionally contain over 1 million defects. However, loading and rendering such large amounts of data using traditional GDI+ for individual drawing or UI thread-driven full refresh rendering results in extremely poor performance stability, leading to UI lag or even complete system crashes, especially for software requiring real-time updates to the interface and image operations. Therefore, a method for processing tens of millions of data points is needed to reduce the data in UI thread rendering, thereby improving the speed of large-scale data rendering and software response rate, and ultimately enhancing overall inspection efficiency. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a big data rendering method and a wafer inspection data rendering system, which can solve the above-mentioned problems.

[0004] Design Principle: The user's field of vision and human eye resolution are limited during operation. When rendering images with massive data volumes in real time, the pixel size of the unit area visible to the user essentially shrinks, and the human eye and brain cannot perceive length differences of about 1µm. When a user zooms in or out on an image, their actual physical field of vision decreases, and the pixel size of the unit area increases. This invention utilizes this characteristic to propose a method that can improve the processing speed of tens of millions of data points. It can effectively reduce the data in the UI thread rendering and significantly improve the speed and response rate of software rendering large amounts of data. The overall solution is as follows: A method for rendering large amounts of data in software is designed. Through spatial filtering algorithms and data downsampling techniques, the amount of data rendered is effectively reduced, lowering the risk of the UI thread freezing due to directly rendering large amounts of data, while simultaneously improving software rendering speed. The specific solution is as follows.

[0005] A big data rendering method based on spatial filtering algorithm and data downsampling is proposed. The big data rendering method includes: S1, data source loading and parsing; loading test data and other data separately from the data source during the import process; S2, data filtering and deduplication; reducing the data volume according to the spatial filtering algorithm and the characteristics of whether the data is occluded; S3, on-demand drawing; identifying the data to be drawn within the field of view; S4, rendering of tens of millions of data points; loading the first screen and subsequent loading after scaling the field of view, and then downsampling according to the defect ratio based on the dataset obtained by on-demand drawing for the final UI rendering.

[0006] Furthermore, the data source loading and parsing includes: S11, data separation and loading, separating recipe information from test information, and loading and parsing the data of a specific test only when a user accesses it; S12, creating a recipe object, dynamically creating a recipe data object based on reflection.

[0007] Furthermore, data filtering and deduplication include: S21, coordinate system transformation, which transforms the coordinates from physical coordinates to display coordinates based on the spatial distribution rules of the data coordinates; S22, constructing a data pyramid, which divides the data into blocks based on spatial range, data quantity, and resolution requirements, and constructs a multi-layered data pyramid structure from the blocks; S23, data deduplication, which removes duplicate data from each layer, removes data from occluded or repeated locations, and downsamples the data according to different data types.

[0008] Furthermore, when constructing the data pyramid, a spatial filtering algorithm is used to divide the data into blocks according to the requirements of spatial range, data quantity and resolution, and then recursively divide it into n blocks. The data pyramid structure is arranged from top to bottom with resolution from low to high.

[0009] Furthermore, data rendering includes: S41, obtaining the data set to be drawn, determining the data set to be drawn based on the current resolution when the first screen is loaded; S42, uniform sampling, using a defect ratio downsampling strategy to project ROI / Defect data to a low-resolution area to achieve uniform sampling; S43, UI rendering, drawing the UI thread based on the obtained data set.

[0010] This invention also provides a wafer inspection data rendering system, comprising: a data loading module for loading test data on demand and creating recipe data objects through a reflection mechanism; a coordinate transformation and segmentation module for converting physical coordinates into display coordinates and spatially segmenting the data; a data processing module for constructing a pyramid structure and deduplicating and downsampling the data; a node selection module for selecting corresponding pyramid level nodes according to changes in the field of view; and a rendering module for completing the visualization rendering of the data according to the current resolution and sampling strategy.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: the present application can effectively reduce the amount of data to be rendered through spatial filtering algorithm and data downsampling technology, reduce the risk of UI thread freezing due to direct rendering of large amounts of data, and improve software rendering speed, which is convenient for promotion and application in data processing fields such as wafer quantity inspection. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the big data rendering method of the present invention; Figure 2 This is a schematic diagram of downsampling based on the defect ratio. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0014] The test data in wafer inspection software recipes can contain millions of ROI data points. How to render over 10 million image data points in seconds without the human eye noticing, while ensuring data integrity, requires sampling based on the actual field of view and resolution. The overall flow of the tens of millions of data rendering method proposed in this invention is as follows: Figure 1 Specifically, the big data rendering method based on spatial filtering algorithms and data downsampling includes the following steps.

[0015] S1. Data Source Loading and Parsing: This step loads Test data and other data separately from the data source during the recipe import process. Specifically, data source loading and parsing includes the following:

[0016] S11. Data Separation and Loading: Separate recipe information from test information, and load and parse the data of a test only when the user accesses a specific test. This step separates the information in the recipe from the information in the test. When importing the recipe into the software, not all data is loaded. Loading and parsing only occurs after the user enters a test.

[0017] S12. Create a recipe object, dynamically creating a recipe data object based on reflection.

[0018] S2. Data filtering and deduplication; reducing data volume based on spatial filtering algorithms and whether data is occluded; specifically, data filtering and deduplication include the following.

[0019] S21. Coordinate system transformation: Based on the rules of data coordinate distribution in space, the coordinates are transformed from physical coordinates to display coordinates.

[0020] S22. Construct a data pyramid by dividing the data into blocks based on spatial range, data quantity, and resolution requirements, and then constructing a multi-layered data pyramid structure from the blocks. When constructing the data pyramid, a spatial filtering algorithm is used to divide the data into blocks according to the spatial range, data quantity, and resolution requirements, and then recursively divide it into n blocks. The data pyramid structure is arranged from top to bottom with resolution from low to high.

[0021] Specifically, the data forms a top-down pyramid structure (representing resolution from low to high from top to bottom), with different resolution levels from top to bottom. The top layer is low-resolution overview data, and the bottom layer is high-resolution detailed data.

[0022] S23. Data Deduplication: Perform deduplication on data at each layer, removing data from occluded or duplicate locations, and downsampling according to different data types. Deduplication includes occlusion removal and duplicate point merging or removal operations based on spatial coordinates.

[0023] Overall, this step addresses the spatial distribution rules of data coordinates, transforming them from physical coordinates to display coordinates. Using a spatial filtering algorithm, the data is divided into blocks based on spatial range, data quantity, and resolution requirements, then recursively divided into n blocks, forming a top-down pyramid structure (representing resolution from low to high). Then, due to the different sizes of the data (ROI and Defect), there will be overlap during rendering, so occluded data and data in the same location need to be removed. Finally, the data type is downsampled.

[0024] S3. On-demand drawing; identify the data to be drawn within the field of view. After the processing in the previous step, each layer of data selects the corresponding nodes from the above n blocks based on the display field of view size. When the field of view is zoomed in, only the most representative upper-level nodes are retrieved; when the field of view is zoomed out, the bottom-level nodes within that field of view are retrieved. Specifically, on-demand drawing includes: selecting the corresponding level nodes from the data pyramid structure according to the current field of view size to achieve adaptive data loading under field of view zooming in and out.

[0025] This step follows the data pyramid structure after the previous step. For each layer of data, based on the display field of view, corresponding nodes are selected from n blocks. When the field of view is magnified, only the most representative upper-level nodes are retrieved; when the field of view is shrunk, the lower-level nodes within that field of view are retrieved.

[0026] S4, rendering of tens of millions of data points; initial screen loading and subsequent loading after zooming in and out, based on the dataset obtained by drawing on demand, and then downsampling according to the defect ratio, to perform the final UI rendering.

[0027] When loading the first screen of ROI / Defects, we can obtain the set of data to be drawn based on the current display resolution. To avoid a large amount of redundant data in areas with dense ROI / Defects, we can downsample according to the defect ratio, projecting the ROI / Defects onto a smaller resolution to ensure sampling uniformity, such as... Figure 2 As shown. Finally, the UI thread is drawn based on the obtained data set.

[0028] The rendering of tens of millions of data points includes initial screen loading and subsequent loading after zooming in. The final UI rendering is performed based on the dataset obtained by drawing on demand and then downsampling according to the defect ratio. The specific steps are as follows.

[0029] S41. Obtain the data set to be drawn. Determine the data set to be drawn based on the current resolution when the first screen is loaded.

[0030] S42. Uniform sampling: A defect ratio downsampling strategy is adopted to project ROI / Defect data onto a low-resolution area to achieve uniform sampling. The defect ratio downsampling strategy can adjust the sampling density according to the distribution of the number of defects per unit area to avoid rendering overload in local dense areas.

[0031] S43, UI rendering: The UI thread is drawn based on the obtained data set. The UI thread drawing process uses an asynchronous loading and caching mechanism to ensure interface smoothness and data integrity.

[0032] The present invention also provides a wafer inspection data rendering system, which includes the following modules.

[0033] The data loading module is used to load Test data on demand and create recipe data objects through reflection.

[0034] The coordinate transformation and partitioning module is used to convert physical coordinates into display coordinates and to partition the data into spatial blocks.

[0035] The data processing module is used to construct the pyramid structure and perform deduplication and downsampling on the data.

[0036] The node selection module is used to select the corresponding pyramid level node based on changes in the field of view.

[0037] The rendering module is used to complete the visualization rendering of data based on the current resolution and sampling strategy.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A big data rendering method based on spatial filtering algorithm and data downsampling, characterized in that, Big data rendering methods include: S1. Data source loading and parsing; during the import process, the recipe loads Test data and other data separately from the data source. S2. Data filtering and deduplication; reducing data volume based on spatial filtering algorithms and whether data is occluded; S3. Draw on demand; identify the data that needs to be drawn within the field of view; S4, rendering of tens of millions of data points; initial screen loading and subsequent loading after zooming in and out, based on the dataset obtained by drawing on demand, and then downsampling according to the defect ratio, to perform the final UI rendering.

2. The big data rendering method according to claim 1, characterized in that, Data source loading and parsing includes: S11. Data separation and loading: Separate recipe information from test information, and load and parse the data of a specific test only when a user accesses that specific test. S12. Create a recipe object, dynamically creating a recipe data object based on reflection.

3. The big data rendering method according to claim 1, characterized in that, Data filtering and deduplication include: S21. Coordinate system transformation: Based on the rules of data coordinate distribution in space, the coordinates are transformed from physical coordinates to the display coordinate system. S22. Construct a data pyramid by dividing the data into blocks based on spatial range, data quantity, and resolution requirements, and then constructing a multi-layered data pyramid structure from the divided data. S23. Data deduplication: Perform deduplication on the data of each layer, remove data in occluded or duplicated positions, and downsample according to different data types.

4. The big data rendering method according to claim 3, characterized in that, When constructing the data pyramid, a spatial filtering algorithm is used to divide the data into blocks according to the requirements of spatial range, data quantity and resolution, and then recursively divide it into n blocks. The data pyramid structure is arranged from top to bottom with resolution from low to high.

5. The big data rendering method according to claim 3, characterized in that, Deduplication includes occlusion culling and duplicate point merging or removal operations based on spatial coordinates.

6. The big data rendering method according to claim 1, characterized in that, On-demand drawing includes: selecting corresponding level nodes from the data pyramid structure based on the current field of view size, enabling adaptive data loading as the field of view is zoomed in or out.

7. The big data rendering method according to claim 1, characterized in that, Data rendering includes: S41. Obtain the data set to be drawn, and determine the data set to be drawn based on the current resolution when the first screen is loaded; S42. Uniform sampling: A defect ratio downsampling strategy is adopted to project ROI / Defect data to a low-resolution area to achieve uniform sampling. S43, UI rendering: Draws the UI thread based on the obtained data set.

8. The big data rendering method according to claim 7, characterized in that, The defect ratio downsampling strategy refers to adjusting the sampling density based on the distribution of the number of defects per unit area to avoid rendering overload in densely populated areas.

9. The big data rendering method according to claim 7, characterized in that, The UI thread rendering process employs an asynchronous loading and caching mechanism to ensure interface smoothness and data integrity.

10. A wafer inspection data rendering system, characterized in that, The system includes: The data loading module is used to load Test data on demand and create recipe data objects through reflection. The coordinate transformation and segmentation module is used to convert physical coordinates into display coordinates and spatially segment the data; The data processing module is used to construct the pyramid structure and perform deduplication and downsampling on the data; The node selection module is used to select the corresponding pyramid level node based on changes in the field of view. The rendering module is used to complete the visualization rendering of data based on the current resolution and sampling strategy.

Images