Systems, devices, and methods for detecting and classifying patterns of heat maps

Abstract

This application relates to systems, apparatus, and methods for detecting and classifying patterns in heatmaps. The data can be converted into an image by encoding non-imaging data into a heatmap. Image analysis techniques can be used to analyze the resulting heatmap using one or more artificial intelligence models to provide inferences. These inferences can indicate the recognition and / or classification of patterns in the heatmap.

Description

Technical Field

[0001] This application relates to systems, apparatus, and methods for detecting and classifying patterns in heat maps. Background Technology

[0002] In manufacturing and production applications, various materials and products can be inspected to detect defects. Some or all of the raw materials, partially processed materials, products, etc., may be unusable due to defects. Therefore, identifying such defects in wafers can be important to control the use and / or distribution of defective products. Summary of the Invention

[0003] This application relates to a system comprising: at least one processor; and at least one non-transitory medium accessible to the processor, encoded with instructions that, when executed, cause the system to: encode a numerical dataset into a heatmap, wherein individual data points of the numerical dataset correspond to corresponding spatial locations, and individual pixels of the heatmap correspond to individual data points; and implement an artificial intelligence (AI) model configured to provide output including an indication of the presence of a pattern in the heatmap.

[0004] Another aspect of this application relates to a method comprising: encoding numerical data into a heatmap, wherein the numerical data includes a plurality of data points corresponding to a plurality of spatial locations, wherein the heatmap includes a plurality of pixels, wherein the plurality of pixels correspond to the plurality of data points; and providing output from an artificial intelligence (AI) model at least in part based on the heatmap, wherein the output includes an indication of the presence of a pattern in the heatmap.

[0005] Another aspect of this application relates to a system comprising: at least one processor; and at least one non-transitory medium accessible to the processor, encoded with instructions that, when executed, cause the system to: encode a first portion of a numerical dataset into a first heatmap, wherein the numerical dataset includes a plurality of data points corresponding to a plurality of spatial locations, and wherein the first heatmap includes a first plurality of pixels, wherein the first plurality of pixels correspond to the plurality of data points of the first portion; encode a second portion of the numerical dataset into a second heatmap, wherein the second heatmap includes a second plurality of pixels, wherein the second plurality of pixels correspond to the plurality of data points of the second portion; and implement a model configured to provide output including an indication of the presence of a pattern in at least one of the first heatmap or the second heatmap.

[0006] Another aspect of this application relates to a method comprising: encoding a first portion of a numerical dataset into a first heatmap, wherein the numerical dataset includes a plurality of data points corresponding to a plurality of spatial locations, and wherein the first heatmap includes a first plurality of pixels corresponding to the plurality of data points of the first portion; encoding a second portion of the numerical dataset into a second heatmap, wherein the second heatmap includes a second plurality of pixels corresponding to the plurality of data points of the second portion; and providing output from a model based at least in part on the first heatmap and the second heatmap, wherein the output includes an indication of the presence of a pattern in at least one of the first heatmap or the second heatmap. Attached Figure Description

[0007] Figure 1 A heat map of an example according to at least one embodiment of the present disclosure.

[0008] Figure 2 Example thermal images of a semiconductor wafer according to at least one embodiment of the present disclosure.

[0009] Figure 3 A diagram illustrating an overview of a data flow according to at least one embodiment of the present disclosure.

[0010] Figure 4 Example thermal images of a semiconductor wafer according to at least one embodiment of the present disclosure are shown.

[0011] Figure 5 A flowchart is provided to give an overview of a method according to at least one embodiment of the present disclosure.

[0012] Figure 6 This is an illustration of an example application of output data according to at least one embodiment of the present disclosure.

[0013] Figure 7 This is an example of multiple heatmaps generated from the same dataset according to at least one embodiment of this disclosure.

[0014] Figure 8 This is an example of multiple heat maps generated from different datasets of the same physical location according to at least one embodiment of this disclosure.

[0015] Figure 9 This is a block diagram of a model for analyzing data according to at least one embodiment of the present disclosure.

[0016] Figure 10 This is a block diagram of a model for analyzing data according to at least one embodiment of the present disclosure.

[0017] Figure 11 This is a schematic illustration of a computing system arranged according to at least one embodiment of the present disclosure.

[0018] Figure 12This is a flowchart of a method according to at least one embodiment of the present disclosure.

[0019] Figure 13 This is a flowchart of a method according to at least one embodiment of the present disclosure. Detailed Implementation

[0020] According to embodiments of this disclosure, one or more types of data can be encoded into a "heatmap" to generate an image that can be used to identify defects in the manufacturing or design process. One or more artificial intelligence (AI) techniques can be used to analyze the image to provide inferences. Inferences may include pattern recognition and / or classification (e.g., anomalies, defects). For example, test data from dies on a wafer can be encoded into a heatmap, where individual pixels of the heatmap represent the corresponding die on the wafer. The heatmap can be analyzed by AI to detect and / or classify defects on the wafer. In some instances, the output of the AI ​​can be used to grade and / or select individual dies. In some instances, the AI ​​can be used to generate pixel masks to indicate the grade of dies and / or the dies to be selected.

[0021] As used herein, AI refers to methods and techniques that enable computing systems to perform one or more tasks. AI may include, but is not limited to, machine learning models such as convolutional neural networks, recurrent neural networks, and support vector machines. AI models may be trained to make inferences (e.g., predictions, classifications) based on input data. These inferences may be provided as output data. In some applications, AI models may further provide confidence levels (e.g., scores) that indicate the probability that the inference is correct.

[0022] AI has made significant progress in extracting information from images. For example, AI has been used to analyze optical images (e.g., images acquired by digital cameras) to detect and / or classify objects, such as for guiding autonomous vehicles on city streets. AI has also been used to analyze non-optical images, such as ultrasound and MRI images of anatomical structures, to detect and / or classify lesions.

[0023] The manufacturing process also leverages the advantages of AI advancements in image analysis. For example, optical images of semiconductor wafers are acquired at various points during the manufacturing process. These images are analyzed using various AI models to detect and / or classify defects in the wafer. AI can be used to analyze hundreds or thousands of wafer images in a fraction of the time required by humans. In some applications, AI can detect and / or classify defects in the wafer more accurately and / or consistently based on images. Semiconductor manufacturers typically perform multiple quality control tests on wafers. However, these tests may not provide data in the form of images (e.g., impedance measurements, computational speed). This data may not benefit from AI advancements in image analysis (e.g., computer vision), potentially limiting the inferences that can be made from the data.

[0024] According to embodiments of this disclosure, data (e.g., numerical data) that may contain multiple data points and is not typically associated with an image (e.g., a dataset) can be encoded as a "heatmap". A heatmap can be generated by assigning pixels with intensity and / or hue (e.g., color) to each data point based on the value of the data points. In some embodiments, the location of pixels in the heatmap may be at least partially based on physical location, such as the location where data is acquired (e.g., measured). Many of the examples described herein relate to semiconductor manufacturing. However, the problems discussed and the solutions described are applicable to any manufacturing or design process involving heatmap generation and analysis to identify or infer defects in the manufacturing process and manufactured products.

[0025] Figure 1 A heat map of an example according to at least one embodiment of the present disclosure. Figure 1 The example shown is a heatmap 100 of a semiconductor wafer comprising multiple dies. In some embodiments, individual dies may contain memory devices, such as dynamic random access memory (DRAM). Pixels 104 of heatmap 100 may correspond to dies on the wafer. The color, intensity, and / or other properties of pixel 104 may be based at least in part on values ​​of data obtained from the corresponding die. Example data includes, but is not limited to, impedance, voltage, current, operating speed, temperature, and the number of faulty elements on the die. The values ​​of data points may be mapped to different colors and / or intensities shown in legend / colormap 102. In some embodiments, each value may be mapped to a different color and / or intensity (e.g., 1 = red, 2 = orange, 3 = yellow, etc.). In some embodiments, a range of values ​​may be mapped to different colors and / or intensities (e.g., 0 to 0.5 = red, 0.6 to 1 = orange, etc.).

[0026] In some embodiments, the range of color and / or intensity can be set such that the minimum value of the data is represented by the minimum value 106 of color and / or intensity, and the maximum value of the data is represented by the maximum value 108 of color and / or intensity. In other embodiments, one or more threshold (e.g., cutoff) values ​​can be assigned to the data. The color and / or intensity assigned to a pixel can be based on a comparison of the value with a threshold. For example, if the value of the data is in the range of 0 to 100, then a threshold can be assigned such that any value equal to or greater than 80 is represented by the maximum value 108 of color and / or intensity. Thus, all data points with values ​​equal to or greater than 80 are represented by the same color in the heatmap. In another instance, if the value of the data is in the range of 0 to 100, then a threshold can be assigned such that any value equal to or less than 20 is represented by the minimum value 106 of color and / or intensity. In some embodiments, both a minimum threshold and a maximum threshold can be applied when generating the heatmap. In some applications, thresholds can be assigned based on acceptable measured values. For example, when the data includes operating speed measurements, if it is unacceptable for the die to operate below 1500MHz, then the threshold can be set such that all pixels corresponding to the die are represented by a minimum value of 106 for color and / or intensity when the operating speed measurement is below 1500MHz.

[0027] In some embodiments, encoding data as a heatmap allows "non-image" data (e.g., repair data, current measurements) to be organized as an image. This can be useful in some applications, such as when data is associated with one or more physical locations. When non-image data associated with a physical location is provided as a heatmap, patterns (e.g., defects) and / or additional information can be derived from said data. In some cases, when data is provided in its original form (e.g., non-image), the patterns and / or additional information may not be obvious or may be undetectable.

[0028] Figure 2This is an example thermal image of a semiconductor wafer according to at least one embodiment of the present disclosure. Thermal image 200 is generated based on repair measurements obtained from individual dies on the semiconductor wafer. Individual pixels of thermal image 200 may correspond to individual dies on the semiconductor wafer. In example thermal image 200, darker pixels (e.g., pixel 206) correspond to dies with higher repair counts (e.g., word lines replaced with redundant word lines), and brighter pixels (e.g., pixel 208) correspond to dies with lower repair counts. Although references are made to brighter and darker pixels, in other embodiments, different colors may be associated with high repair counts (e.g., red) and low repair counts (e.g., blue). Thermal image 200 includes die lines 202 with high repair counts on the upper right side of the wafer, and die clusters 204 with high repair counts at the center of the wafer. These spatial patterns (e.g., lines 202 and clusters 204) may indicate defects in individual dies and in the wafer. For example, line 202 can indicate scratches on the wafer, and cluster 204 can indicate uneven doping on the wafer. In some cases, defects in the wafer can indicate problems in the process and / or equipment. Some defects, such as uneven doping, may not be readily apparent from optical images of the wafer and / or by reviewing raw repair count measurements of the die. Therefore, providing data as a heatmap 200 allows for the additional detection of manufacturing defects.

[0029] Although Figure 2 The examples shown illustrate that at least some defects can be detected by manually observing heatmaps, but properly detecting patterns (such as defects) from heatmaps can require time and skill. For manufacturing applications such as semiconductor manufacturing, where hundreds or thousands of products may be produced daily and tested multiple times during manufacturing, the number of heatmaps requiring inspection further increases. Therefore, it may be impossible for humans to inspect heatmaps.

[0030] According to embodiments of this disclosure, AI models can be trained to analyze heatmaps, thereby detecting and / or classifying patterns within the heatmaps. For example, a pattern may correspond to a defect in a semiconductor wafer. In some embodiments, the AI ​​model may be at least partially based on AI models commonly used for image analysis (e.g., facial recognition, object recognition, medical diagnostics). In some applications, AI models for image analysis can provide better performance in analyzing heatmaps compared to other non-image analysis AI models used to analyze raw data (e.g., non-heatmaps). These powerful AI image analysis models are not only able to detect and / or classify patterns faster than human observers, but the AI ​​models may also be more accurate and / or detect patterns that are unrecognizable to human observers. Furthermore, multiple measurements (e.g., repair counts and current) can be used to generate multidimensional heatmaps that can be analyzed by the AI ​​model, which may be difficult for humans to interpret.

[0031] Figure 3This diagram illustrates an overview of the data flow according to at least one embodiment of the present disclosure. As indicated in FIG300, data 302 may be provided to a thermal encoder 304. In some embodiments, data 302 may comprise a plurality of data points. In some embodiments, data points may be measurements (e.g., temperature, leakage current, voltage) obtained from one or more spatial locations (e.g., transistors in a circuit, dies on a wafer). In some embodiments, information about the spatial location of an individual measurement may be included in data 302. The spatial location information may be provided as Cartesian coordinates, polar coordinates, and / or any other suitable coordinate system. The spatial location information may be one-dimensional, two-dimensional, and / or three-dimensional. The thermal encoder 304 may generate a thermal map at least in part based on the measurements. The thermal map may comprise one or more pixels, wherein in some embodiments, individual pixels may correspond to individual spatial locations. One or more properties of the pixels (e.g., intensity, hue) may be at least in part based on the measurements to encode data 302 into a thermal map. For example, in some embodiments, the thermal encoder 304 may comprise a color map (e.g., color map 102). The heatmap encoder 304 can compare the values ​​of data points with values ​​contained in a color map and assign color and / or intensity to pixels based on the color and / or intensity associated with the values ​​in the color map. In some embodiments, the heatmap encoder 304 can assign the position of a pixel within a heatmap based at least in part on spatial location information associated with the data point.

[0032] A heatmap generated by heatmap encoder 304 can be provided to AI model 306. AI model 306 may include one or more models trained to make inferences from the heatmap, such as pattern recognition (e.g., detection) and / or classification in the heatmap. In some embodiments, AI model 306 may use semantic segmentation and / or instance segmentation to recognize and / or classify patterns. In other embodiments, other segmentation techniques may be used. In some embodiments, AI model may include one or more neural networks. In some embodiments, AI model may include one or more convolutional neural networks, such as region-based convolutional neural networks (R-CNN). Examples of suitable AI models include, but are not limited to, masked R-CNN, faster R-CNN with region proposal networks, DeepMask with fast R-CNN, fully convolutional instance segmentation (FCIS), or combinations thereof. AI model 306 can provide an output 308 containing inferences, such as the identified pattern (if present) and / or the classification of the pattern. In some embodiments, output 308 may indicate pixels in a pattern contained within the heatmap. Optionally, in some embodiments, output 308 may include a confidence level (e.g., a score) associated with the pattern and / or its classification.

[0033] The output 308 of AI model 306 can be used to provide one or more graphical overlays that can be displayed on a heatmap, the overlays providing inferred graphical indications. Figure 3 In the example shown, a graphical overlay including borders and text boxes is displayed as an overlay on heatmap 310. Heatmap 310 can be generated by encoding data obtained from a bare die on a semiconductor wafer by heatmap encoder 304. Borders 312, 314, 316, and 318 indicate the location of the pattern identified in heatmap 310. Text boxes 320, 322, 324, and 326 indicate the classification of the corresponding pattern and the confidence level of the classification. Figure 3 In the example shown, the pattern indicated by boundary 312 is classified as a defect (defect 1) with a 99% confidence level; the pattern indicated by boundary 314 is classified as another defect (defect 4) with a 75% confidence level; the pattern indicated by boundary 316 is classified as yet another defect (defect 3) with a 98% confidence level; and the pattern indicated by boundary 318 is classified as another defect (defect 2) with an 85% confidence level.

[0034] In other embodiments, output 308 may not be provided graphically. For example, output 308 may include text indicating which patterns (if present), corresponding categories, and / or corresponding confidence levels were identified. In some instances, output 308 may textually indicate which pixels of heatmap 310 are contained with each pattern. For example, a Cartesian coordinate system may be generated for heatmap 310, and the coordinates of pixels for individual patterns may be provided. In some instances, both textual and graphical overlay information may be included with output 308. Output 308 may be provided for display (e.g., a screen), storage in non-transitory computer-readable media, and / or storage in another AI model. Figure 3 (Not shown in the text) or used for further processing.

[0035] In some instances, output 308 may be provided to a thermal image decoder 328. The thermal image decoder 328 may convert the pixels of the thermal image 310 into physical locations and output location information 330. In some embodiments, the location information 330 may contain the physical location of the pattern of output 308. For example, the location information 330 may indicate which dies on a wafer are included in the pattern. Although in Figure 3 In one instance, location information 330 is displayed as text, but in other instances, location information 330 may contain graphic information and / or other information.

[0036] In some embodiments, the thermal encoder 304, AI model 306, and / or thermal decoder 328 may be implemented by a set of computer-readable instructions (e.g., software, software modules, software applications) executed by one or more processors. In some embodiments, the thermal encoder 304, AI model 306, and / or thermal decoder 328 may be implemented in hardware, such as application-specific integrated circuits (ASICs) and / or field-programmable gate arrays (FPGAs). In some embodiments, the thermal encoder 304, AI model 306, and / or thermal decoder 328 may be implemented as a combination of hardware and software.

[0037] As mentioned, AI model 306 can be trained to identify various patterns / defects in the heatmap. AI model 306 can also be further trained to classify various patterns in the heatmap. The type, classification, and number of patterns can vary depending on the application. Figure 3 In the example shown, AI model 306 is trained to identify and classify various defects in a semiconductor wafer. However, AI model 306 can also be trained to identify different patterns in a semiconductor wafer or patterns in different objects. For example, the AI ​​model can be trained to detect patterns in a heat map generated from data acquired from a circuit (e.g., a field-programmable gate array), and said patterns may include short circuits and / or bad connections.

[0038] Figure 4 Examples of thermal images of semiconductor wafers according to at least one embodiment of the present disclosure are shown. Thermal images 400, 402, and 404 illustrate different types of wafer defects that an AI model (e.g., AI model 306) can be trained to identify and / or classify. Thermal image 400 illustrates an example of a patch 412 defect on the wafer. Thermal image 402 was generated from data obtained from a wafer with a line 414 defect. Thermal image 404 illustrates an example of a wafer with a curve 416 defect. Figure 3 and 4 The defects shown and described are provided by way of example only. In other embodiments, the AI ​​model may be trained to identify more, fewer, and / or different defects in the chip based on heatmaps.

[0039] Figure 5 A flowchart is provided to provide an overview of a method according to at least one embodiment of the present disclosure. As shown in flowchart 500, data preparation 502, training 504, and / or analysis 506 may be performed. Data preparation 502 may be performed to train an AI model to analyze heatmaps, and to prepare data for analysis by the trained AI model. For example, numerical data may be encoded into heatmaps, such as heatmap 310, as indicated by box 510. In some embodiments, the heatmap may be generated from the data by an encoder (e.g., encoder 304).

[0040] A heatmap generated at box 510 can be provided for analysis 506 and / or training 504. When the heatmap is used for training, it can be annotated, as indicated by box 512. When the heatmap is provided as input, the annotations indicate the desired and / or correct output of the AI ​​model. Annotation can be performed manually and / or semi-automatically. For example, when annotating manually, the user can indicate the location of a pattern on the heatmap and / or the classification of the pattern in the heatmap. In an example, when annotating semi-automatically, image segmentation techniques (e.g., thresholding, watershed algorithms, gradient analysis) can be used on the heatmap to find patterns. The user can then add classifications to the patterns. In other embodiments, other annotation techniques can be used.

[0041] Annotated heatmaps (e.g., annotations and heatmaps) may be included in the training dataset. The training dataset may contain hundreds, thousands, or even thousands of annotated heatmaps. The size of the training dataset may be based at least in part on the complexity of the patterns and / or the number of different types of patterns to be identified and / or classified. The training dataset may be used to train an AI model, such as AI model 306, as indicated by box 514. The training dataset may be used to determine the architecture and / or other parameters of the AI ​​model. Parameters may include the number of layers, values ​​of matrix and / or vector weights. During training, acceptable parameters of the AI ​​model are determined at least in part based on the accuracy of the AI ​​model's predictions / inferences (e.g., identification and / or classification) using the parameters. In some embodiments, accuracy may be determined based on a comparison of the AI ​​model's output with the annotations produced at box 512. The AI ​​model may be considered trained when the output matches the annotation or is within the acceptable range of the annotation. The acceptable range of the output (e.g., accuracy within 1%, 5%, or 10% of the annotation) may be based on various factors, such as available computational resources, computation time, and / or potential damage due to undetected and / or misclassified patterns. For example, if the potential damage is low, then the acceptable range may be wider. When an AI model is deployed (e.g., implemented) for analysis of 506, AI model parameters that provide outputs with acceptable accuracy can be used with the AI ​​model.

[0042] In some embodiments, the accuracy of the prediction can be represented by a loss function. The value of the loss function may be high when the AI ​​model makes poor predictions (e.g., inaccurately identifying and / or classifying patterns in a heatmap) and low when the AI ​​model makes good predictions (e.g., more accurately identifying and / or classifying patterns in a heatmap). The AI ​​model can be considered trained when the loss function reaches its minimum. When the AI ​​model is deployed for analysis, the AI ​​model parameters that provide the minimum value of the loss function can be used with the AI ​​model.

[0043] In some embodiments, after training with the training dataset at box 514, the trained AI model can be evaluated, as indicated in box 516. In some embodiments, the trained AI model can analyze heatmaps not provided in an unannotated training dataset, referred to as the evaluation dataset. Although no annotations are provided to the AI ​​model, the evaluation dataset may contain annotations to determine the accuracy of the AI ​​model's output (e.g., how well the AI ​​model is trained). If the AI ​​model's output is within the acceptable error tolerance of the annotations in the evaluation dataset, then the AI ​​model can be deployed for analysis 506. If the AI ​​model's output exceeds the acceptable error tolerance, then the AI ​​model can be retrained with additional training datasets and / or trained again at box 514, and evaluated again at box 516.

[0044] Once trained, the AI ​​model can be deployed to analyze new, unannotated heatmaps to detect and / or classify patterns within them, as indicated in box 518. The AI ​​model can provide outputs that may include patterns similar to or identical to output 308, classifications, the location of patterns in the heatmap, and / or confidence levels for identification and / or classification. The output of the AI ​​model can be used in application 508.

[0045] The output of the AI ​​model can be provided to a decoder (e.g., decoder 328), as indicated in box 520. The decoder can map the patterns recognized by the AI ​​model in the heatmap to physical locations (e.g., the location of transistors in a circuit, the location of dies on a wafer) for obtaining data used to generate the heatmap. However, box 520 can be omitted when application 508 does not require physical locations.

[0046] The application 508 may provide the output of a trained AI model generated at box 518, a heatmap generated at box 510, and / or a location generated at box 520, the output, heatmap, and / or location collectively outputting data. The application 508 may be implemented by hardware, software (e.g., computer-executable instructions), human interaction, and / or a combination thereof. In some embodiments, the application 508 may be implemented at least in part by one or more AI models. The application 508 may include, but is not limited to, making decisions, detecting, tracking trends, and / or providing display information based at least in part on the output data.

[0047] In the example of generating a heatmap from a semiconductor wafer containing multiple dies, the output data can be used to determine whether to accept or reject individual dies and / or the entire wafer. For example, if the percentage of dies containing defects is equal to or greater than a threshold, the entire wafer can be rejected. When the percentage of dies containing defects is less than a threshold, only the individual dies containing defects may be discarded. Continuing with the semiconductor wafer example, the output data can be used to classify individual dies. For example, dies identified by the AI ​​model as containing defects can be classified as low quality, dies located near or adjacent to defects can be classified as medium quality, and dies located further away from defects can be classified as high quality. Similarly, for the semiconductor wafer example, output data related to defects (e.g., defect classification) can be analyzed alone or in combination with additional data (e.g., production line, lot number) to detect equipment failures. In some instances, output data can be collected over time to detect trends, such as the most common defect types, the most common defect locations, yield, etc. In some instances, output data can be provided for manual review. For example, a display can provide one or more graphic overlays to the heatmap, such as... Figure 3 The examples shown are for users to view.

[0048] Figure 6 This is an illustration of an example application of output data according to at least one embodiment of the present disclosure. In application 600, the output of an AI model (e.g., AI model 306) may include the identification and classification of defects 604 on a heatmap 602. The heatmap 602 may correspond to individual pixels corresponding to individual dies of a semiconductor wafer. The output of the AI ​​model (e.g., output 308) may be combined with die location information of the wafer (e.g., location information 330) to determine which dies of the wafer are contained in defects 604. The output of the AI ​​model and the location information may be used to generate a pixel mask 606. The pixel mask 606 may include a set of pixels 608 corresponding to dies not contained in defects 604 and a set of pixels 610 corresponding to dies contained in defects 604. In some embodiments, the pixel mask 606 may be used to calculate wafer yield (e.g., acceptance / rejection ratio). In some embodiments, pixel mask 606 may be used in a programming device to sort the dies at the position corresponding to pixel 610 in a different manner than the dies at the position corresponding to pixel 608 when the dies are removed from the wafer (e.g., placed in different containers, placed in different manufacturing lines). Figure 6 The example shown is only one application of using the output of an AI model, and as referenced... Figure 5 The discussion suggests that other or additional applications can be executed based on the output of the AI ​​model.

[0049] like Figure 5As mentioned earlier, training datasets can contain many (e.g., hundreds or thousands) of annotated entries. Annotation is typically performed manually, and expert reviewers may require significant time. Often, AI models used for image analysis must be retrained to recognize and / or classify patterns in different contexts. For example, an AI model trained to recognize patterns in computed tomography images used for medical diagnosis cannot be reused to recognize patterns in optical images to guide autonomous vehicles. While the same underlying architecture (e.g., UNet, Mask R-CNN) can be used, the number of layers, various weights and coefficients, or other parameters of the AI ​​model may need to be modified for proper pattern recognition. Therefore, it may be necessary to prepare and provide new training datasets to retrain the AI ​​model.

[0050] In contrast, when encoding non-imaging / numerical data (e.g., current, temperature) into heatmaps, trained AI models can be "reused" across multiple datasets in some applications. For example, an AI model trained to identify and / or classify patterns in heatmaps generated from leakage current data from dies on a semiconductor wafer can accurately identify and / or classify patterns in heatmaps generated from impedance data from dies on a semiconductor wafer with minimal retraining. In some applications, the AI ​​model may require minimal retraining when identifying similar patterns and / or classifications. In some applications, the AI ​​model may require minimal retraining when using the same colormap (e.g., the range of pixel colors and intensities) to generate heatmaps across data types. Reducing or eliminating AI model retraining can reduce the time and / or cost associated with AI models and / or increase the applications that can use them.

[0051] While conventional imaging data can be multidimensional, these dimensions are typically limited to space (e.g., two-dimensional and three-dimensional) and time. Additionally, multi-channel imaging data is often limited to a finite range of tones (e.g., red, green, and blue channels). However, according to embodiments of this disclosure, AI models can be trained to analyze multidimensional and / or multi-channel datasets, where one or more dimensions and / or channels correspond to different data ranges, datasets, and / or data types. This can improve the predictive accuracy of the AI ​​model and / or allow for the identification and / or classification of a wider range of patterns and / or more complex patterns in some applications. Without being limited by any particular theory, in some cases, these potential improvements can be attributed at least in part to the greater prominence of different patterns across different dataset ranges or different data types. In some cases, these potential improvements can be attributed at least in part to the presence of at least some correlation between different data types (e.g., higher current may be associated with higher temperature). Therefore, the cumulative information from different data types can provide greater predictive information compared to data types analyzed individually in some applications.

[0052] Figure 7These are examples of multiple heatmaps generated from the same dataset according to at least one embodiment of this disclosure. Heatmaps 700, 702, and 704 may be generated from the same dataset, for example, a set of temperature readings for physical locations. However, individual heatmaps 700, 702, and 704 are generated from different data ranges. Heatmap 700 is generated from the entire range of the dataset (e.g., all values). Heatmaps 702 and 704 are generated from half of the dataset. The half used to generate heatmap 702 contains the highest values ​​of the dataset (e.g., excluding the half of the dataset with the lowest values). The half used to generate heatmap 704 contains intermediate values ​​(e.g., excluding one-quarter of the highest value and one-quarter of the lowest value).

[0053] exist Figure 7 In the examples shown, compared to heatmaps 700 and 704, heatmap 702 provides better detection of pattern 706 at its center. Conversely, compared to heatmaps 700 and 702, heatmap 704 provides better isolation of pattern 708 at its edges. Therefore, different patterns may be more easily detected in different heatmaps 700, 702, and 704. In some applications, analyzing all three heatmaps 700, 702, and 704 can provide more accurate pattern identification and / or classification than an AI model analyzing only one of them.

[0054] although Figure 7 The example shows three heatmaps based on three different ranges of a dataset. However, in other instances, two ranges can be used to generate two different heatmaps from the dataset. In other instances, more than three ranges can be used to generate more than three different heatmaps from the dataset.

[0055] Figure 8 This is an example of multiple heatmaps generated from different datasets of the same physical location according to at least one embodiment of the present disclosure. Heatmaps 800 and 802 may be generated from two different datasets. However, individual values ​​from the two different datasets may correspond to the same set of physical locations. For example, heatmap 800 may be generated based on repair data of a die on a semiconductor wafer, and heatmap 802 may be generated based on impedance data of a die on a semiconductor wafer.

[0056] exist Figure 8 In the examples shown, heatmap 800 provides better detection of ellipsoidal defects 804 compared to heatmap 802, while heatmap 802 provides better detection of scratch defects 806 compared to heatmap 800. Therefore, in some applications, analyzing both heatmaps 800 and 802 can provide more accurate identification and / or classification of patterns than an AI model analyzing only one of them.

[0057] although Figure 8 The example shows two heatmaps based on two different data types, but in other instances, more than two data types can be used to generate more than two heatmaps.

[0058] AI models can analyze multi-channel and / or multi-dimensional data using one or more techniques. In some embodiments, channels and / or dimensions can be analyzed using individual AI models. The outputs of individual AI models can be provided individually and / or combined to provide a combined output. In some applications, analyzing dimensions and / or channels using different AI models can be faster. For example, in some applications, parallel processing may be easier when different AI models analyze different channels and / or dimensions. In some applications, analyzing dimensions and / or channels using different AI models can allow AI models to be trained to recognize and / or classify patterns specific to those channels and / or dimensions.

[0059] In some embodiments, channels and / or dimensions can be analyzed using an AI model. In some embodiments, inferences made by the AI ​​model may be influenced by multiple channels and / or dimensions. In some embodiments, information from multiple channels and / or dimensions may be accumulated by the AI ​​model to provide inferences. In some applications, analyzing all channels and / or dimensions using an AI model allows the AI ​​model to leverage relationships between data in different channels and / or dimensions to identify and / or classify patterns.

[0060] Figure 9 This is a block diagram of a model for analyzing data according to at least one embodiment of the present disclosure. Model 900 may receive multiple heatmaps 902, 904, 906 as input. Although Figure 9 Three heatmaps are displayed, but any number of heatmaps can be provided as input. In some embodiments, heatmaps 902, 904, and 906 can be generated from different ranges of the same dataset, for example, as shown in the reference. Figure 7 As described. In some embodiments, heatmaps 902, 904, and 906 may be generated from different data types, for example, as referenced. Figure 8 As described. In other embodiments, other relationships may exist between heatmaps 902, 904, and 906 (e.g., heatmaps 902, 904, and 906 are generated from data acquired at different times).

[0061] Model 900 may include multiple AI models 908, 910, and 912. In some embodiments, AI models 908, 910, and 912 may correspond to AI model 306. The number of AI models including model 900 may correspond to the number of heatmaps provided as input to model 900. Figure 9The example shown is three. AI models 908, 910, and 912 can receive heatmaps 902, 904, and 906, respectively. AI models 908, 910, and 912 can analyze the corresponding heatmaps 902, 904, and 906 and provide corresponding outputs that may include information such as any identified and / or classified patterns. In some embodiments, the outputs of AI models 908, 910, and 912 may correspond to output 308.

[0062] In some embodiments, the individual outputs of AI models 908, 910, and 912 may be the output of model 900. In other embodiments, two or more of the individual outputs may be combined by combiner 914 to provide a single combined output from model 900. In some embodiments, combiner 914 may perform one or more operations on the outputs of AI models 908, 910, and 912 to produce a combined output. Instance operations include, but are not limited to, sum, weighted sum, average, weighted average, and / or combinations thereof. Combiner 914 may be implemented in software (e.g., executable instructions) and / or hardware. In some embodiments, the combined output may be used in various applications, such as application 508.

[0063] In some embodiments, AI models 908, 910, and 912 may have the same architecture (e.g., all may be masked R-CNN models). In some embodiments, at least one of AI models 908, 910, and 912 may have a different architecture than the other AI models. In some embodiments, AI models 908, 910, and 912 may be trained separately. In other embodiments, a single AI model may be trained and replicated to provide AI models 908, 910, and 912.

[0064] In some applications, Model 900 allows independent analysis of heatmaps 902, 904, and 906. In other applications, Model 900 facilitates parallel processing of heatmaps 902, 904, and 906, which can reduce processing time.

[0065] Figure 10 This is a block diagram of a model for analyzing data according to at least one embodiment of the present disclosure. Model 1000 may receive multi-channel and / or multi-dimensional inputs. In some embodiments, heatmaps may be provided for individual channels and / or dimensions. Figure 10 In the examples shown, heatmaps 1002, 1004, and 1006 represent channels for multi-channel and / or multi-dimensional input. Although Figure 10 Three heatmaps are displayed, but any number of heatmaps can be provided. The number of heatmaps provided may correspond to the number of channels and / or dimensions. In some embodiments, heatmaps 1002, 1004, and 1006 may be generated from different ranges of the same dataset, for example, as shown in the reference... Figure 7As described. In some embodiments, heatmaps 1002, 1004, and 1006 may be generated from different data types, for example, as referenced Figure 8 As described. In other embodiments, other relationships may exist between heatmaps 1002, 1004, and 1006 (e.g., heatmaps 1002, 1004, and 1006 are generated from data acquired at different times).

[0066] Model 1000 may include an AI model, such as AI model 306. Model 1000 may analyze heatmaps 1002, 1004, and 1006 and provide an output that may include, for example, information about any identified and / or classified patterns. In some embodiments, the output of AI model 1000 may correspond to output 308. In some embodiments, model 1000 may allow the accumulation of information from heatmaps 1002, 1004, and 1006 for inference / prediction. In some applications, this may improve the accuracy of identification and / or classification. In some applications, this may allow model 1000 to identify and / or classify additional patterns.

[0067] Figure 11 This is a schematic illustration of a computing system arranged according to at least one embodiment of the present disclosure. The computing system 1100 may be used to implement one or more models, such as AI model 306, model 900, AI model 908, AI model 910, AI model 912, and / or model 1000. The computing system 1100 may be used to implement one or more components, such as encoder 304, decoder 328, and / or combiner 914, and / or to implement applications, such as reference frame 508 and / or... Figure 6 The application described. The computing system 1100 may include one or more processors 1102, one or more computer-readable media 1104, a memory controller 1110, a memory 1112, and one or more interfaces 1114. In some instances, the computing system 1100 may include a display 1116.

[0068] Computer-readable medium 1104 is accessible by processor 1102. Computer-readable medium 1104 is encoded with executable instructions 1108. Executable instructions 1108 may include executable instructions for generating one or more heatmaps from data (e.g., numerical data, non-imaging data). Executable instructions 1108 may include executable instructions for implementing one or more models (which may include one or more AI models) to identify and / or classify patterns in the heatmaps. Executable instructions 1108 are executable by processor 1102. In some instances, executable instructions 1108 may also include instructions for generating or processing training datasets and / or training one or more models. Alternatively or additionally, in some instances, one or more of the models, or a portion thereof, may be implemented in hardware including computer-readable medium 1104 and / or processor 1102, such as application-specific integrated circuits (ASICs) and / or field-programmable gate arrays (FPGAs).

[0069] Computer-readable media 1104 may store data 1106. In some instances, data 1106 may contain one or more training datasets, such as training dataset 1118. Training dataset 1118 may contain one or more annotated heatmaps. In some instances, training dataset 1118 may be received from another computing system (e.g., a cloud computing system, a testing system). In other instances, training dataset 1118 may be generated by computing system 1100. In some instances, the training dataset may be used to train one or more AI models. In some instances, data 1106 may contain data used in the AI ​​model (e.g., weights, connections between nodes). In some instances, data 1106 may contain other data, such as new data 1120. New data 1120 may contain one or more heatmaps not included in training dataset 1118. In some instances, new data 1120 may be analyzed by a trained AI model to provide output, which may include the recognition and / or classification of patterns in the heatmaps. In some instances, data 1106 may include output generated by one or more applications implemented by computing system 1100 based on the output of one or more models (e.g., pixel masks, graphic overlays). Computer-readable media 1104 may be implemented using any media containing non-transitory computer-readable media. Examples include memory, random access memory (RAM), read-only memory (ROM), volatile or non-volatile memory, hard disk drives, solid-state drives, or other storage devices. Although Figure 11 A single media may be displayed, but multiple media may be used to implement computer-readable media 1104.

[0070] In some instances, processor 1102 may be implemented using one or more central processing units (CPUs), graphics processing units (GPUs), ASICs, FPGAs, or other processor circuitry systems. In some instances, processor 1102 may execute some or all of the instructions 1108. In some instances, processor 1102 may communicate with memory 1112 via memory controller 1110. In some instances, memory 1112 may be volatile memory, such as dynamic random access memory (DRAM). In some instances, memory 1112 may provide information to and / or receive information from processor 1102 and / or computer-readable medium 1104 via memory controller 1110. Although a single memory 1112 and a single memory controller 1110 are shown, any number may be used. In some instances, memory controller 1110 may be integrated with processor 1102.

[0071] In some instances, interface 1114 may provide a communication interface to another device (e.g., a test system, test probe), a user, and / or a network (e.g., a LAN, WAN, the Internet). Interface 1114 may be implemented using wired and / or wireless interfaces (e.g., Wi-Fi, Bluetooth, HDMI, USB, etc.). In some instances, interface 1114 may include user interface components that can receive input from a user. Examples of user interface components include a keyboard, mouse, touchpad, touchscreen, and microphone. In some instances, interface 1114 may transmit information, including user input, data 1106, training dataset 1118, and / or new data 1120, between an external device and one or more components of computing system 1100 (e.g., processor 1102 and computer-readable media 1104).

[0072] In some instances, the computing system 1100 may communicate with a display 1116 as a separate component (e.g., using wired and / or wireless connections), or the display 1116 may be integrated with the computing system. In some instances, the display 1116 may display data 1106, such as output (e.g., output 308) generated by one or more models implemented by the computing system 1100. Any number or variety of displays may be present, including one or more LED, LCD, plasma, or other display devices.

[0073] Figure 12 This is a flowchart of a method according to at least one embodiment of the present disclosure. In some embodiments, method 1200 may be performed wholly or partially by AI model 306, model 900, AI model 908, AI model 910, AI model 912, model 1000 and / or computing system 1100.

[0074] At box 1202, the action "encoding numerical data into a heatmap" can be performed. In some embodiments, the numerical data may comprise multiple data points corresponding to multiple spatial locations, and the heatmap may comprise multiple pixels, wherein the multiple pixels correspond to the multiple data points. In some embodiments, the properties of the pixels among the multiple pixels may be based at least in part on the values ​​of corresponding data points among the multiple data points. In some embodiments, the numerical data into the heatmap includes at least in part a colormap, with colors assigned to the multiple pixels based on the values ​​of corresponding data points among the multiple data points. For example, as referenced... Figure 1 As described. In some embodiments, the color map assigns a color to the entire range of values ​​for multiple data points. In some embodiments, the color map assigns the same color to the values ​​of multiple data points that are equal to or below a threshold. In some embodiments, the color map assigns the same color to the values ​​of multiple data points that are equal to or above a threshold.

[0075] At box 1204, it is possible to “provide output from an artificial intelligence (AI) model based at least in part on a heatmap, wherein the output includes an indication of whether a pattern exists in the heatmap.”

[0076] In some embodiments, at block 1206, the action may be "decoding the output to provide position information of the pattern, wherein the position information includes a spatial position among a plurality of spatial positions." For example, as referenced Figure 3 and 5 As described. In some embodiments, when block 1206 is executed, at block 1208, the "generation of a pixel mask based at least in part on location information" can be performed. In some embodiments, it can be as follows Figure 6 The pixel mask is generated as described in the document.

[0077] In some embodiments, the plurality of spatial locations correspond to a plurality of dies on a semiconductor wafer, and method 1200 may further include “assigning a first grade to dies included in a pattern and assigning a second grade to dies outside the pattern” at block 1210. In some embodiments, the first grade and the second grade are different. In some embodiments, more than two different grades may be assigned. In some embodiments, additional measurements and / or other data may also be used to assign grades to dies. Although shown after blocks 1206 and 1208, in some embodiments, block 1210 may be performed before, simultaneously with, and / or instead of blocks 1206 and / or 1208.

[0078] Figure 13This is a flowchart of a method according to at least one embodiment of the present disclosure. In some embodiments, method 1300 may be performed wholly or partially by AI model 306, model 900, AI model 908, AI model 910, AI model 912, model 1000 and / or computing system 1100.

[0079] At block 1302, the action "encoding a first portion of the numerical dataset into a first heatmap" can be performed. In some embodiments, the numerical dataset contains a plurality of data points corresponding to a plurality of spatial locations, and the first heatmap contains a first plurality of pixels corresponding to the plurality of data points of the first portion. At block 1304, the action "encoding a second portion of the numerical dataset into a second heatmap" can be performed. In some embodiments, the second heatmap may contain a second plurality of pixels corresponding to the plurality of data points of the second portion. In some embodiments, the first and second portions of the dataset may correspond to data points with different value ranges, as referenced. Figure 7 As described. In some embodiments, the first and second portions of the dataset may correspond to different data types (e.g., voltage, impedance, temperature), as referenced. Figure 8 As described. In some embodiments, the first portion and the second portion may correspond to the same set of spatial locations.

[0080] At box 1306, it is possible to “provide output from the model based at least in part on the first heatmap and the second heatmap, wherein the output includes an indication of the presence of a pattern in at least one of the first heatmap or the second heatmap.”

[0081] In some embodiments, the model includes a first artificial intelligence (AI) model and a second AI model. In these embodiments, method 1300 may further include "analyzing a first heatmap using the first AI model to generate a first output" at block 1308 and "analyzing a second heatmap using the second AI model to generate a second output" at block 1310. In some embodiments, the output includes a first output and a second output. In some embodiments, the first output and the second output may be combined to provide an output.

[0082] In some embodiments, method 1300 may further include training a first AI model with a first training dataset and training a second AI model with a second training dataset. In some embodiments, the second training dataset is different from the first training dataset. In other embodiments, method 1300 may further include training the AI ​​model with the training dataset and replicating the AI ​​model to provide the first AI model and the second AI model.

[0083] Instead of executing boxes 1308 and 1310, the model may contain an artificial intelligence model, and the "Analyze the first and second heatmaps with the AI ​​model to provide output" function may be executed at box 1312.

[0084] The systems, methods, and apparatuses disclosed herein allow "non-image" data (e.g., repaired data, current measurements) to be organized into images, such as heatmaps. Generating heatmaps allows for the analysis of data using one or more AI models designed for analyzing image data. Heatmaps and / or AI models can allow for the extraction of additional information from data in some applications. For example, patterns and / or additional information may not be readily apparent or may be undetectable when the data is in its original form. In some applications, the use of heatmaps allows for the "reuse" of AI models across multiple data types, which can reduce the amount of training required for the AI ​​models. In some applications, the use of heatmaps allows for the analysis of multiple data ranges and / or data types as different channels and / or dimensions of the data.

[0085] While the examples described herein generally refer to semiconductor devices, including semiconductor wafers and / or circuits including dies, the disclosed systems, methods, and apparatus are not limited to these applications. In fact, the techniques disclosed herein can be applied to any numerical data in which individual data points are associated with spatial locations, whether in one, two, or three dimensions.

[0086] This document sets forth certain details to provide a full understanding of examples of various embodiments of the present disclosure. However, it should be understood that the examples described herein can be practiced without these specific details. Furthermore, the specific examples of the present disclosure described herein should not be construed as limiting the scope of the disclosure to those specific examples. In other instances, well-known circuits, control signals, timing protocols, artificial intelligence models, and software operations have not been shown in detail to avoid unnecessarily obscuring the embodiments of the present disclosure. Additionally, terms such as “coupled” mean that two components can be electrically coupled directly or indirectly. Indirect coupling may imply that two components are coupled through one or more intermediate components.

Claims

1. A system comprising: At least one processor; as well as The processor can access at least one non-transitory medium encoded with instructions that, when executed, cause the system to: Encoding a numerical dataset into a heatmap, wherein the numerical dataset includes data obtained from multiple dies on a semiconductor wafer, wherein individual data points in the numerical dataset correspond to corresponding spatial locations, and individual pixels in the heatmap correspond to the individual data points; as well as An artificial intelligence (AI) model is implemented, configured to provide an output including an indication regarding the presence of a pattern in the heatmap, wherein the pattern includes defects in the semiconductor wafer; the output further includes an indication of a subset of pixels in the heatmap that are contained within the pattern; and the instruction, when executed, further causes the system to decode the output to provide the spatial location of the pattern. When the instructions are executed, the system further causes the system to generate a pixel mask based on the spatial location of the pattern, wherein the pixel mask includes a first set of pixels corresponding to a die not included in the pattern and a second set of pixels corresponding to a die included in the pattern, and wherein the pixel mask is used to program a sorting device to separate the die corresponding to the first set of pixels and the die corresponding to the second set of pixels when the die is removed from the semiconductor wafer.

2. The system of claim 1, wherein the AI ​​model comprises a region-based convolutional neural network.

3. The system of claim 1, further comprising a display, wherein the instructions, when executed, further cause the system to generate display information for at least one of the heatmap or the output and provide the display information to the display.

4. The system of claim 3, wherein the display is configured to provide the output as a graphic overlay on the heatmap.

5. The system of claim 1, wherein the output further includes the classification of the pattern.

6. The system of claim 5, wherein the output further includes the confidence level of the classification.

7. The system of claim 1, wherein the output further includes the confidence level of the pattern.

8. The system of claim 1, wherein the spatial location corresponds to an individual die of the plurality of dies on the semiconductor wafer.

9. A method comprising: Numerical data is encoded into a heatmap, wherein the numerical data includes multiple data points corresponding to multiple spatial locations, wherein the data points are obtained from multiple dies on a semiconductor wafer, and wherein the heatmap includes multiple pixels, wherein the multiple pixels correspond to the multiple data points; The output is provided from an artificial intelligence (AI) model based at least in part on the heat map, wherein the output includes an indication, wherein the indication relates to the presence of a pattern in the heat map, and wherein the pattern includes a defect in the semiconductor wafer; The output is decoded to provide position information of the pattern, wherein the position information includes a spatial position among the plurality of spatial positions; as well as A pixel mask is generated at least in part based on the location information, wherein the pixel mask includes a first set of pixels corresponding to a die not included in the pattern and a second set of pixels corresponding to a die included in the pattern, wherein the pixel mask is used to program a sorting device to separate the die corresponding to the first set of pixels and the die corresponding to the second set of pixels when the die is removed from the semiconductor wafer.

10. The method of claim 9, wherein the characteristics of the pixels among the plurality of pixels are based at least in part on the values ​​of corresponding data points among the plurality of data points.

11. The method of claim 9, wherein encoding the numerical data into the heatmap comprises at least partially based on a colormap, assigning colors to the plurality of pixels based on the values ​​of corresponding values ​​among the plurality of data points.

12. The method of claim 11, wherein the colormap assigns colors to the entire range of values ​​for the plurality of data points.

13. The method of claim 11, wherein the colormap assigns the same color to the values ​​of the plurality of data points based on a comparison with a threshold.

14. The method of claim 13, wherein the colormap assigns the same color to the value of the plurality of data points when the value is equal to or higher than the threshold or when the value is equal to or lower than the threshold.

15. The method of claim 9, wherein the plurality of spatial locations correspond to the plurality of dies on the semiconductor wafer, and the method further comprises assigning a first grade to dies included in the pattern among the plurality of dies, and assigning a second grade to dies outside the pattern among the plurality of dies, wherein the first grade is different from the second grade.

16. A system comprising: At least one processor; as well as The processor can access at least one non-transitory medium encoded with instructions that, when executed, cause the system to: Encode a first portion of a numerical dataset into a first heatmap, wherein the numerical dataset includes multiple data points corresponding to multiple spatial locations, and wherein the first heatmap includes a first plurality of pixels, wherein the first plurality of pixels correspond to the multiple data points of the first portion; The second portion of the numerical dataset is encoded into a second heatmap, wherein the second heatmap includes a second plurality of pixels, wherein the second plurality of pixels correspond to a plurality of data points in the second portion, wherein the numerical dataset includes data obtained from a plurality of dies on a semiconductor wafer, and wherein the first portion includes a first data type and the second portion includes a second data type; and An implementation model is configured to provide an output including an indication relating to the presence of a pattern in at least one of a first heatmap or a second heatmap, wherein the pattern includes a defect in the semiconductor wafer, and wherein the model includes an artificial intelligence (AI) model configured to analyze the first heatmap corresponding to a first portion having a first data type and the second heatmap corresponding to a second portion having a second data type to provide the output, wherein the pattern includes a defect in the semiconductor wafer, wherein the output further includes an indication of a subset of pixels in the heatmap that are contained in the pattern, and wherein the instructions, when executed, further cause the system to decode the output to provide the spatial location of the pattern. When the instructions are executed, the system further causes the system to generate a pixel mask based on the spatial location of the pattern, wherein the pixel mask includes a first set of pixels corresponding to a die not included in the pattern and a second set of pixels corresponding to a die included in the pattern, and wherein the pixel mask is used to program a sorting device to separate the die corresponding to the first set of pixels and the die corresponding to the second set of pixels when the die is removed from the semiconductor wafer.

17. The system of claim 16, wherein the first portion includes a portion of a first range of values ​​corresponding to the plurality of data points, and the second portion includes a portion of a second range of values ​​corresponding to the plurality of data points, wherein the first range is different from the second range.

18. The system of claim 16, wherein the plurality of spatial locations of the plurality of data points in the first portion are the same as the plurality of spatial locations of the plurality of data points in the second portion.

19. The system of claim 16, wherein the model further comprises a first artificial intelligence (AI) model and a second AI model, wherein the first AI model analyzes the first heatmap and provides a first output, and the second AI model analyzes the second heatmap and provides a second output.

20. The system of claim 19, wherein the output includes the first output and the second output.

21. The system of claim 19, wherein the first output and the second output are combined to provide the output.

22. The system of claim 19, wherein the first AI model and the second AI model comprise at least one of different architectures or different parameters.

23. A method comprising: Encode a first portion of a numerical dataset into a first heatmap, wherein the numerical dataset includes multiple data points corresponding to multiple spatial locations, and wherein the first heatmap includes a first plurality of pixels corresponding to the multiple data points of the first portion; The second part of the numerical dataset is encoded into a second heatmap, wherein the second heatmap includes a second plurality of pixels corresponding to a plurality of data points of the second part, wherein the numerical dataset includes data obtained from a plurality of dies on a semiconductor wafer, and wherein the first part includes a first data type and the second part includes a second data type; The method provides output from a model based at least in part on the first heatmap and the second heatmap, wherein the output includes an indication, wherein the indication relates to the presence of a pattern in at least one of the first heatmap or the second heatmap, and wherein the pattern includes a defect in the semiconductor wafer, and wherein the model includes an artificial intelligence (AI) model, and the method further includes analyzing the first heatmap and the second heatmap with the AI ​​model to provide the output; The output is decoded to provide position information of the pattern, wherein the position information includes a spatial position among the plurality of spatial positions; as well as A pixel mask is generated at least in part based on the location information, wherein the pixel mask includes a first set of pixels corresponding to a die not included in the pattern and a second set of pixels corresponding to a die included in the pattern, wherein the pixel mask is used to program a sorting device to separate the die corresponding to the first set of pixels and the die corresponding to the second set of pixels when the die is removed from the semiconductor wafer.

24. The method of claim 23, wherein the model comprises a first artificial intelligence (AI) model and a second AI model, wherein the method further comprises: The first AI model is used to analyze the first heatmap to generate a first output; as well as The second AI model is used to analyze the second heatmap to generate a second output, wherein the output includes the first output and the second output.

25. The method of claim 24, further comprising combining the first output and the second output to provide the output.

26. The method of claim 24, further comprising: The first AI model is trained using the first training dataset; as well as The second AI model is trained using a second training dataset, wherein the second training dataset is different from the first training dataset.

27. The method of claim 24, further comprising: Training AI models using training datasets; and The AI ​​model is copied to provide the first AI model and the second AI model.

Images