Method and electronic device for processing semiconductor wafer image

IN594890BActive Publication Date: 2026-07-09SAMSUNG ELECTRONICS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
IN202141037350
Authority / Receiving Office
IN · IN
Patent Type
Patents
Current Assignee / Owner
Filing Date
2021-08-17
Publication Date
2026-07-09
Estimated Expiration
2041-08-17

AI Technical Summary

Technical Problem

Current AI inspection methods for semiconductor wafers face challenges in real-time defect detection due to imbalanced and heterogeneous labeled data, requiring manual intervention and being cumbersome with heterogeneous input data, leading to unreliable predictions.

Method used

An electronic device with a data-driven model that processes semiconductor wafer images using multiple adaptive operating modes, including standard, advanced, user-set, and adaptive modes, employing knowledge-guided filters and interpolation/restore modules to handle variable image sizes and qualities, and a multi-model repository for defect inspection, reducing the need for human intervention.

Benefits of technology

Enables efficient, reliable, and automated defect detection across various wafer and facility types, improving prediction accuracy and reducing manual intervention by preprocessing and selecting high-quality images, thus enhancing the reliability of defect prediction.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

ABSTRACT METHOD AND ELECTRONIC DEVICE FOR PROCESSING SEMICONDUCTOR WAFER IMAGE Embodiments herein provide a method for processing a semiconductor wafer image by an electronic device (100). The method includes receiving at least one semiconductor wafer image. Further, the method includes determining at least one operating mode to process the at least one semiconductor wafer image. Further, the method includes predicting at least one of defect of the at least one semiconductor wafer image and a quality of the at least one semiconductor wafer image based on the at least one determined operating mode. The proposed method can be used to suggest use of multiple models-based repository for requirement-based defect inspection and sustainable prediction performance. The proposed method eliminates the need for human intervention at multiple stages of the defect detection pipeline using a filter along with interpolation and restoration modules. FIG. 3
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[001] The present disclosure relates to the field of image processing and recommendation methods, and more particularly to methods and an electronic device for processing a semiconductor wafer image using a data driven model (e.g., Artificial intelligence (AI) model or machine learning (ML) model.BACKGROUND

[002] Currently, various visible and invisible defects occur during semiconductor manufacturing processes resulting in vast amounts of time and resource loss. Using artificial intelligence (AI) monitoring / inspection can prevent such losses and ensure excellent end-product quality / improved yield. Further, several promising defect detection models are available, but the key challenge in real-time deployment / application / prediction / defect identification still lies in addressing input data and appropriate model selection. Further, available labeled defect data is highly imbalanced and heterogeneous, often adversely affecting the learning of machine learning (ML) / Deep learning (DL) classifiers. Also, large chunks of defect data remain unlabeled because of inherent insufficiencies in image pre-processing and prediction modules. Further, real-time applications still require manual intervention at several stages of defect prediction.

[003] In other words, current AI inspection methods typically use wafer images / relevant data and most of the efforts for performance improvement in the AI inspection methods is model-centric. With such improvements, a testing performance may look promising, but in real-time application, when heterogeneous data arrives, selection of images, their pre-processing, validation and prediction become extremely cumbersome for the users. Unless the input data and model selection are appropriate, resulting predictions may become unreliable.

[004] It is desired to address the above-mentioned disadvantages or other short comings or at least provide a useful alternative.OBJECTS

[005] The principal object of the embodiments herein is to disclose methods and an electronic device for processing a semiconductor wafer image.

[006] Another object of the embodiments herein is to provide an unprecedented platform (e.g., AI model or machine learning (ML) model) for semiconductor wafer defection with multiple and adaptive modes of operation for generalizability across wafer, process and facility types in a quick, efficient and time effective manner.

[007] Another object of the embodiments herein is to suggest use of multiple models-based repository for requirement-based defect inspection and sustainable prediction performance.

[008] Another object of the embodiments herein is to eliminate the need for human intervention at multiple stages of the defect detection pipeline using a filter (e.g., knowledge-guided filter) along with interpolation and restoration modules.

[009] Another object of the embodiments herein is to allow for heterogeneous size and quality images in the prediction pipeline for wider application.SUMMARY

[0010] Accordingly, the embodiments herein provide a method for processing a semiconductor wafer image. The method includes receiving, by an electronic device, at least one semiconductor wafer image. Further, the method includes determining, by the electronic device, at least one operating mode to process the at least one semiconductor wafer image. Further, the method includes predicting, by the electronic device, at least one of defect of the at least one semiconductor wafer image and a quality of the at least one semiconductor wafer image based on the at least one determined operating mode.

[0011] In an embodiment, at least one of the defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image is predicted using a data driven model. In an embodiment, process the at least one semiconductor wafer image comprises at least one of a discard the at least one semiconductor wafer image that has unacceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the at least one semiconductor wafer image and accept the at least one semiconductor wafer image that has acceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the at least one semiconductor wafer image.

[0012] In an embodiment, the data driven model is trained by analysing and learning a plurality of mixed defect type data associated with a plurality of semiconductor wafer images over a period of time to configure a defect detection related prediction model. The defect detection related prediction model predicts at least one defect type data associated with the plurality of semiconductor wafer images.

[0013] In an embodiment, the at least one operating mode comprises at least one of a standard operating mode, an advanced operating mode, a user-set operating mode, and an adaptive operating mode.

[0014] In an embodiment, the standard operating mode predicts at least one of the defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image based on quality of the at least one semiconductor wafer image. The standard operating mode receives the at least one semiconductor wafer image with same size.

[0015] In an embodiment, the advanced operating mode receives the at least one semiconductor wafer image with variable sizes. The advanced operating mode uses at least one filter to determine an aspect ratio of the at least one semiconductor wafer image and a pixel range of the at least one semiconductor wafer image to predict at least one of the defect of the at least one semiconductor wafer image with the variable sizes and the quality of the at least one semiconductor wafer image with the variable sizes.

[0016] In an embodiment, the adaptive operating mode receives the at least one semiconductor wafer image with variable sizes. The advanced operating mode determines at least one of a variable size of the at least one semiconductor wafer image in a predefined range and region of interest (ROI) of the at least one semiconductor wafer image in a predefined range, identifies that the at least one semiconductor wafer image is not in the predefined range, performs a restoration operation on the at least one identified semiconductor wafer image and determines that the at least one restored semiconductor wafer image is in the predefined range. The adaptive operating mode predicts at least one of the defect of the at least one semiconductor wafer image with the variable sizes and the quality of the at least one semiconductor wafer image with the variable sizes based on at least one feature of the at least one semiconductor wafer image, wherein the at least one feature comprises a scene graph, a temperature, a noise, and a sensor data.

[0017] In an embodiment, the user-set operating mode receives the at least one semiconductor wafer image with variable sizes along with user suggestion. The advanced operating mode uses at least one filter to determine an aspect ratio of the at least one semiconductor wafer image and a pixel range of the at least one semiconductor wafer image to predict at least one of the defect of the at least one semiconductor wafer image with variable sizes and the quality of the at least one semiconductor wafer image with variable sizes based on the user suggestion.

[0018] In an embodiment, the at least one filter removes unwanted image of the at least one semiconductor wafer image and ensure that of the at least one semiconductor wafer image comprises desired region of interest. The at least one filter is determined based on at least one of size of the the at least one semiconductor wafer image, a quality of the the at least one semiconductor wafer image, an aspect ratio of the the at least one semiconductor wafer image and a region of interest of the at least one semiconductor wafer image.

[0019] In an embodiment, the at least one operating mode is configured in a multi-model repository. The multi-model repository includes at least one of a single defect detector, a multiple defect detector, a bounding box detector, a defect percentage detector, an ensemble detector, a hybrid defect detector, and a multi-modal input based detector. The single defect detector detects whether the at least one semiconductor wafer image contains defect or not. The multiple defect detector detects one or more defects on the at least one semiconductor wafer image with details of defect types. The bounding box detector detects precise location of defects on the at least one semiconductor wafer image. The defect percentage detector detects intensity of defect on at least one semiconductor wafer image. The ensemble detector uses one or more models to perform defect detection and uses some form of voting to arrive at reliable decision. The hybrid defect detector uses a mixed architecture for defect detection. The multi-modal input based detector uses at least one architecture that is suitable for accepting multiple inputs of different modalities.

[0020] In the adaptive operating mode, if inputs other than images are available, then a suitable model from the repository can be recommended for use.

[0021] In an embodiment, the electronic device receives at least one recommendation for predicting the at least one of defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image based on the at least one determined operating mode, a user history, an available input and defect information associated with the at least one semiconductor wafer image. In another embodiment, the electronic device receives at least one recommendation for the at least one operating mode.

[0022] Accordingly, the embodiments herein provide an electronic device for processing a semiconductor wafer image. The electronic device includes a semiconductor wafer image processing controller coupled with a processor and a memory. The semiconductor wafer image processing controller is configured to receive at least one semiconductor wafer image and determine at least one operating mode to process the at least one semiconductor wafer image. Further, the semiconductor wafer image processing controller is configured to predict at least one of defect of the at least one semiconductor wafer image and a quality of the at least one semiconductor wafer image based on the at least one determined operating mode.

[0023] These and other aspects of the example embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood, however, that the following descriptions, while indicating example embodiments and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the example embodiments herein without departing from the scope thereof, and the example embodiments herein include all such modifications.BRIEF DESCRIPTION OF FIGURES

[0024] Embodiments herein are illustrated in the accompanying drawings, throughout which like reference letters indicate corresponding parts in the various figures. The embodiments herein will be better understood from the following description with reference to the drawings, in which:

[0025] FIG. 1 shows various hardware components of an electronic device, according to embodiments as disclosed herein;

[0026] FIG. 2 shows various hardware components of a semiconductor wafer image processing controller included in the electronic device, according to embodiments as disclosed herein;

[0027] FIG. 3 is a flow chart illustrating a method for processing a semiconductor wafer image, according to embodiments as disclosed herein;

[0028] FIG. 4 is a flow chart illustrating a method for processing the semiconductor wafer image based on a standard operating mode, according to embodiments as disclosed herein;

[0029] FIG. 5 is a flow chart illustrating a method for processing a semiconductor wafer image based on an advanced operating mode, according to embodiments as disclosed herein;

[0030] FIG. 6 is a flow chart illustrating a method for processing a semiconductor wafer image based on a user-set operating mode, according to embodiments as disclosed herein;

[0031] FIG. 7 is a flow chart illustrating a method for processing a semiconductor wafer image based on an adaptive operating mode, according to embodiments as disclosed herein;

[0032] FIG. 8a and FIG. 8b are example illustrations in which an integrated image quality assurance steps for a removal of undesirable, poor qualityimages using a filter, according to embodiments as disclosed herein;

[0033] FIG. 9 is an example illustration in which restoration and preprocessing of acceptable quality images to meet standard requirements are depicted, according to embodiments as disclosed herein; and

[0034] FIG. 10 is an example illustration in which prediction using model trained for addressing inherent data imbalance - balanced deep learning based data augmentation to balance underrepresented defect classes are depicted, according to embodiments as disclosed herein.DETAILED DESCRIPTION

[0035] The example embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the nonlimiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The description herein is intended merely to facilitate an understanding of ways in which the example embodiments herein can be practiced and to further enable those of skill in the art to practice the example embodiments herein. Accordingly, this disclosure should not be construed as limiting the scope of the example embodiments herein.

[0036] The embodiments herein achieve methods for processing a semiconductor wafer image. The method includes receiving, by an electronic device, at least one semiconductor wafer image. Further, the method includes determining, by the electronic device, at least one operating mode to process the at least one semiconductor wafer image. Further, the method includes predicting, by the electronic device, at least one of defect of the at least one semiconductor wafer image and a quality of the at least one semiconductor wafer image based on the at least one determined operating mode.

[0037] The proposed methods can be used to provide an unprecedented platform for semiconductor wafer defection with multiple and adaptive modes of operation for generalizability across wafer, process, and facility types. In the proposed methods, prediction models should be easily adaptable while using other input imaging techniques such across metrology techniques.

[0038] The proposed methods can be used to suggest use of multiple models based repository for requirement-based defect inspection and sustainable prediction performance. The proposed methods can be used to eliminate the need for human intervention at multiple stages of the defect detection pipeline using knowledge-guided filters along with interpolation and restoration modules. The proposed methods can be used to allow for heterogeneous size and quality images in the prediction pipeline for wider application. In the proposed method, theworkflow in part or whole can be used for image quality check and data augmentation purposes.

[0039] The proposed method can be used in part or whole to generate high-quality defect data from input images. This can help in augmenting datasets that have vast inherent imbalance for modeling and analysis. The proposed method can be used for highly reliable large-scale defect detection on unlabelled data. The proposed method can be used for performing image selection, find regions of interest, process various inputs and allocate models for multi-level defect detection and hence, largely eliminate the need for manual intervention.

[0040] Using different modes of operation, the user of the electronic device can receive recommendations or decide selection parameters for wafer inspection based on requirements. The proposed method can be deployed for realtime semiconductor fabrication process inspection. The proposed method can be adopted or expanded to across metrology techniques such as Critical Dimension Scanning Electron Microscope (CD-SEM), Critical Dimension (CD)-optical, Transmission electron microscopy (TEM) etc.

[0041] The method proposed method can be used for obtaining high quality wafer defect data using observation and knowledge-guided filters, available in multiple operational modes for AI-based semiconductor wafer inspection. The method proposed method can be used to obtain high quality wafer defect data from images using input-based recommendation. The method proposed method can be used to provide user-centric AI-inspection experience. The method proposed method uses aspect ratio as parameter for discarding undesirable data. The method proposed method uses data derived from a domain expert-designed filter for screening images, wherein the filter is based on a range observed in counts of defective and normal dies on wafer maps. The proposed method uses a plurality of inputs based on availability and deploys model from a repository. The proposed method determines workflows based on secondary data derived from input images, such that the primary data provided to the prediction model is more reliable. The method can generate one or more characteristics of the wafer using a single workflow.

[0042] The method can greatly reduce requirement of repeated image capturing, which is a very time consuming and computationally expensive process. The method can be used for processing the semiconductor wafer image with high accuracy, fast and cost-efficient manner. Instead of blindly passing all images to the model for prediction, the proposed method can be used to remove undesirable images, which is extremely important to ensure excellent model performance with minimal manual intervention.

[0043] The method performs the customizable semiconductor wafer defect analysis based on user requirements. In the proposed method, balanced training is achieved using the Adam optimizer, which shows better results than the prior arts. In the proposed method, a first knowledge-guided filter removes unwanted images from the prediction workflow and a second knowledge-guided filter assures the images have desired region of interest. The proposed method can be used for processing other type image apart from semiconductor image.

[0044] Referring now to the drawings, and more particularly to FIGS. 1 through 10, where similar reference characters denote corresponding features consistently throughout the figures, there are shown example embodiments.

[0045] FIG. 1 shows various hardware components of an electronic device (100), according to embodiments as disclosed herein. The electronic device (100) can be, for example, but not limited to a laptop, a desktop computer, a notebook, a smartphone, a foldable phone, a smart TV, a tablet, an immersive device, and an internet of things (IoT) device. In an embodiment, the electronic device (100) includes a processor (110), a communicator (120), a memory (130), a semiconductor wafer image processing controller (140), a data driven controller (150) and a filter (160). The processor (110) is coupled with the communicator (120), the memory (130), the semiconductor wafer image processing controller (140), the data driven controller (150) and the filter (160).

[0046] The semiconductor wafer image processing controller (140) is configured to receive one or more semiconductor wafer image(s). After receiving the one or more semiconductor wafer image(s), the semiconductor wafer image processing controller (140) is configured to determine one or more operating mode(s) to process the one or more semiconductor wafer image(s). In an embodiment, the one or more semiconductor wafer(s) image is processed by discarding the one or more semiconductor wafer image(s) that has unacceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the one or more semiconductor wafer image(s). In another embodiment, the one or more semiconductor wafer image(s) is processed by accepting the one or more semiconductor wafer image(s) that has acceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the one or more semiconductor wafer image(s).

[0047] The one or more operating mode(s) can be, for example, but not limited to a standard operating mode (explained in the FIG. 4), an advanced operating mode (explained in the FIG. 5), a user-set operating mode (explained in the FIG. 6), and an adaptive operating mode (explained in the FIG. 7).

[0048] In an embodiment, the standard operating mode receives the one or more semiconductor wafer image(s) with same size with respect to a model input requirements and set parameters. The standard operating mode predicts one or more of the defect(s) of the one or more semiconductor wafer image(s) and the quality of the one or more semiconductor wafer image(s) based on quality of the one or more semiconductor wafer image(s).

[0049] In an embodiment, the advanced operating mode receives the one or more semiconductor wafer image(s) with variable sizes. The interpolator changes or modifies the input size to the model requirement based on the selected input (e.g., operating modes, user input or the like). The advanced operating mode uses one or more filter(s) (160) to determine an aspect ratio of the one or more semiconductor wafer image(s) and a pixel range of the one or more semiconductor wafer image to predict one or more of the defect(s) of the one or more semiconductor wafer image(s) with the variable sizes and the quality of the one or more semiconductor wafer image(s) with the variable sizes.

[0050] In an embodiment, the adaptive operating mode receives the one or more semiconductor wafer image(s) with variable sizes. The interpolator changes or modifies the input size to the model requirement based on the selected input(e.g., operating modes, user input or the like). The advanced operating mode determines one or more of a variable size of the one or more semiconductor wafer image(s) in a predefined range and region of interest (ROI) of the one or more semiconductor wafer image(s) in a predefined range, identifies that the one or more semiconductor wafer image(s) is not in the predefined range, performs a restoration operation on the one or more identified semiconductor wafer image(s) and determines that the one or more restored semiconductor wafer image is in the predefined range. The adaptive operating mode predicts one or more of the defect of the one or more semiconductor wafer image(s) with the variable sizes and the quality of the one or more semiconductor wafer image(s) with the variable sizes based on one or more feature(s) of the one or more semiconductor wafer image(s). The one or more feature(s) can be, for example, but not limited to a scene graph, a temperature, a noise, and a sensor data.

[0051] In the adaptive operating mode, if inputs other than images are available, then a suitable model from the repository can be recommended for use.

[0052] In an embodiment, the advanced operating mode receives the one or more semiconductor wafer image(s) with variable sizes along with user suggestion. The advanced operating mode uses one or more filter(s) (160) to determine an aspect ratio of the one or more semiconductor wafer image(s) and a pixel range of the one or more semiconductor wafer image(s) to predict one or more of the defect of the one or more semiconductor wafer image(s) with variable sizes and the quality of the one or more semiconductor wafer image(s) with variable sizes based on the user suggestion.

[0053] The one or more filter(s) (160) removes unwanted image of the one or more semiconductor wafer image(s) and ensures that of the one or more semiconductor wafer image(s) comprises desired region of interest. The one or more filter(s) (160) is determined based on one or more of size of the the one or more semiconductor wafer image(s), a quality of the the one or more semiconductor wafer image(s), an aspect ratio of the the one or more semiconductor wafer image and a region of interest of the one or more semiconductor wafer image(s).

[0054] The one or more operating mode(s) is configured in a multi-model repository. The multi-model repository includes one or more of a single defect detector, a multiple defect detector, a bounding box detector, a defect percentage detector, an ensemble detector, a hybrid defect detector, and a multi-modal input based detector. The single defect detector detects whether the one or more semiconductor wafer image(s) contains a defect or not. The multiple defect detector detects one or more defects on the one or more semiconductor wafer image(s) with details of defect types (e.g., unknown defect, single defect, random defect, centre defect, donut defect, edge-ring defect, edge-loc defect or the like). The bounding box detector detects precise location of defects on the one or more semiconductor wafer image(s). The defect percentage detector detects intensity of defect on one or more semiconductor wafer image(s). The ensemble detector uses one or more models to perform defect detection and uses some form of voting to arrive at reliable decision. The hybrid defect detector uses a mixed architecture for defect detection, and wherein the multi-modal input based detector uses one or more inputs of different types that is suitable for accepting multiple inputs of different modalities.

[0055] Based on the one or more determined operating mode(s), the semiconductor wafer image processing controller (140) is configured to predict a defect of the one semiconductor wafer image and a quality of the one or more semiconductor wafer image(s). The defect of the one or more semiconductor wafer image(s) and the quality of the one or more semiconductor wafer image(s) is predicted using a data driven model. The data driven model is trained by analysing and learning a plurality of mixed defect type data associated with a plurality of semiconductor wafer images over a period of time to configure a defect detection related prediction model. The defect detection related prediction model predicts one or more defect type data(s) associated with the plurality of semiconductor wafer images.

[0056] Further, the semiconductor wafer image processing controller (140) is configured to receive one or more recommendation(s) for predicting the one or more of defect of the one or more semiconductor wafer image(s) and the quality of the one or more semiconductor wafer image(s) based on the one or more determined operating mode, a user history, an available input and defect information associated with the one or more semiconductor wafer image(s).In another embodiment, the electronic device (100) receives the recommendation for the one or more operating mode(s).

[0057] The semiconductor wafer image processing controller (140) is physically implemented by analog or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hardwired circuits, or the like, and may optionally be driven by firmware.

[0058] Further, the processor (110) is configured to execute instructions stored in the memory (130) and to perform various processes. Various applications (e.g., virtual assistance application, voice assistance application or the like) are stored in the memory (130). The communicator (120) is configured for communicating internally between internal hardware components and with external devices via one or more networks. The memory (130) also stores instructions to be executed by the processor (110). The memory (130) may include non-volatile storage elements. Examples of such non-volatile storage elements may include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable (EEPROM) memories. In addition, the memory (130) may, in some examples, be considered a non-transitory storage medium. The term "non-transitory" may indicate that the storage medium is not embodied in a carrier wave or a propagated signal. However, the term "nontransitory" should not be interpreted that the memory (130) is non-movable. In certain examples, a non-transitory storage medium may store data that can, over time, change (e.g., in Random Access Memory (RAM) or cache).

[0059] Further, at least one of the pluralities of modules / controller may be implemented through the AI model using the data driven controller (150). The data driven controller (150) can be a ML model based controller and AI model based controller. A function associated with the AI model may be performed through the non-volatile memory, the volatile memory, and the processor (110). The processor (110) may include one or a plurality of processors. At this time, one or a plurality of processors may be a general purpose processor, such as a central processing unit (CPU), an application processor (AP), or the like, a graphics-only processing unit such as a graphics processing unit (GPU), a visual processing unit (VPU), and / or an AI-dedicated processor such as a neural processing unit (NPU).

[0060] The one or a plurality of processors control the processing of the input data in accordance with a predefined operating rule or AI model stored in the non-volatile memory and the volatile memory. The predefined operating rule or artificial intelligence model is provided through training or learning.

[0061] Here, being provided through learning means that a predefined operating rule or AI model of a desired characteristic is made by applying a learning algorithm to a plurality of learning data of the characteristic features associated with the image data. The learning may be performed in a device itself in which AI according to an embodiment is performed, and / or may be implemented through a separate server / system.

[0062] The AI model may comprise of a plurality of neural network layers. Each layer has a plurality of weight values, and performs a layer operation through calculation of a previous layer and an operation of a plurality of weights. Examples of neural networks include, but are not limited to, convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), restricted Boltzmann Machine (RBM), deep belief network (DBN), bidirectional recurrent deep neural network (BRDNN), generative adversarial networks (GAN), and deep Q-networks.

[0063] The learning algorithm is a method for training a predetermined target device (for example, a robot) using a plurality of learning data to cause, allow, or control the target device to make a determination or prediction. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0064] Although FIG. 1 shows various hardware components of the electronic device (100) but it is to be understood that other embodiments are not limited thereon. In other embodiments, the electronic device (100) may include less or more number of components. Further, the labels or names of the components are used only for illustrative purpose and does not limit the scope of the invention. One or more components can be combined together to perform same or substantially similar function in the electronic device (100).

[0065] FIG. 2 shows various hardware components of the semiconductor wafer image processing controller (140) included in the electronic device (100), according to embodiments as disclosed herein. In an embodiment, the semiconductor wafer image processing controller (140) includes an input image acquisition unit (140a), a size and quality validator (140b), a model input preprocessor (140c), a region of interest finder (140d), a wafer defect analyzer (140e), a defect type identifier (140f) and a restoration unit (140g).

[0066] The input image acquisition unit (140a) receives the input image and determines the size of the input image and quality of the input image using the size and quality validator (140b). The model input pre-processor (140c) performs the fast and accurate size conversion using the nearest neighbour interpolation technique. The image does not meet criteria removed from the predication flow.

[0067] The region of interest finder (140d) can perform operations such as calculating a number of normal and defective dies, which in turn are reflective of a circular region in wafer maps, where the defects are usually found. Using data analysis, the region of interest finder (140d) arrive at a threshold for discarding unsuitable images, e.g., those with incomplete circular regions The wafer defect analyser (140e) determines the image defect and indicate the defect type as centre defect, donut defect, edge-ring defect, edge-loc defect, a random defect, an unknown defect, and near-full defect using the defect type identifier (140f). The restoration unit (140g) provides the high-quality restored images for the further use using various techniques (e.g., de-noising techniques, fuzzy based filter techniques or the like). The high-quality restored images are analysed by the predication model.

[0068] In this platform, unlike the standard modes of operation where user has to process images based on model requirements, select ROI etc., the system takes care of processing heterogeneous images, checks if they are within the acceptable range and allows only those that directly meet / can be pre-processed and / or restored to meet model requirements. In an example embodiment, we show the design of two knowledge-based filters for labelling the WM-811K data. Depending on the data, our workflow can provide / recommend similar filters / deploy other state-of-the-art filters for removing poor quality images.

[0069] Although FIG. 2 shows various hardware components of the semiconductor wafer image processing controller (140) but it is to be understood that other embodiments are not limited thereon. In other embodiments, the semiconductor wafer image processing controller (140) may include less or more number of components. Further, the labels or names of the components are used only for illustrative purpose and does not limit the scope of the invention. One or more components can be combined together to perform same or substantially similar function in the semiconductor wafer image processing controller (140).

[0070] FIG. 3 is a flow chart (300) illustrating a method for processing the semiconductor wafer image, according to embodiments as disclosed herein. The operations (302-306) are performed by the semiconductor wafer image processing controller (140).

[0071] At 302, the method includes receiving the semiconductor wafer image. At 304, the method includes determining the operating mode to process the semiconductor wafer image. At 306, the method includes predicting the defect of the semiconductor wafer image and a quality of the semiconductor wafer image based on the determined operating mode.

[0072] The proposed method deploys knowledge-guided filters such as aspect ratio, sum of normal and defective dies on wafer maps etc., for selection of high-quality input images. The proposed method can be used to facilitate generation of filter based on different optical modes for profile-based matching and defect detection. The filters are configured based on the operating mode.. The proposed method removes images with noise if they do not fall in the acceptable range or else restores them.

[0073] FIG. 4 is a flow chart (400) illustrating a method for processing the semiconductor wafer image based on the standard operating mode, according to embodiments as disclosed herein.

[0074] Referring to the FIG. 4, the operations (402-406) are performed by the semiconductor wafer image processing controller (140). At 402, the method includes receiving the input image with preset size. At 404, the method includes configuring the wafer defect detection (WDD) preset model to process the input image with the preset size. At 406, the method includes detecting the wafer defect using the WDD preset model.

[0075] The input images of preset size determined by the expert can be accepted for prediction by the model, e.g., 52x52.The output and its format is also preset by the expert, e.g., single defect classification. Only one model is available and loaded. It is expected that all images are of high quality as no pre-processing or filtering is involved in the pipeline. This does not involve user-personalization. The system can work on it's own for long hours. This mode does not facilitate prediction on variable input size. The prediction depends on input image quality.

[0076] FIG. 5 is a flow chart (500) illustrating a method for processing a semiconductor wafer image based on the advanced operating mode, according to embodiments as disclosed herein.

[0077] Referring to the FIG. 5, the operations (502-510) are performed by the semiconductor wafer image processing controller (140).

[0078] At 502, the method includes receiving the input image acceptor with the variable size. At 504, the method includes using the interpolator used to determine the best approximation of the pixel's intensity of the input image. At 506, the method includes performing the quality check of the input image. At 508, the method includes configuring the WDD preset model to process the input image. At 510, the method includes detecting the wafer defect using the WDD preset model.

[0079] The images of variable sizes are acceptable, e.g., 52x52, 26x26. The output and its format are also preset by the expert, e.g., single defect classification. Only one model is available and loaded. For ensuring high quality images, two filters are applied to check for the aspect ratio and pixel range. This does not involve user-personalization. The system can work on it's own for long hours. This mode facilitates prediction on variable input size. The prediction is more reliable since the electronic device (100) performs quality check images input to the model.

[0080] FIG. 6 is a flow chart (600) illustrating a method for processing a semiconductor wafer image based on a user-set operating mode, according to embodiments as disclosed herein. The operations (302-306) are performed by the semiconductor wafer image processing controller (140).

[0081] Referring to the FIG. 6, At 602, the method includes receiving the input image with user-set size. At 604, the method includes using the interpolator changing or modifying the input size to the model requirement based on the selected input (e.g., operating modes, user input or the like). At 606, the method includes performing the quality check of the input image. At 608, the method includes configuring the WDD user set model to process the input image. At 610, the method includes detecting the wafer defect using the WDD user set model.

[0082] The images of variable sizes are acceptable, e.g., 52x52, 26x26. User sets what image sizes are expected in the pipeline. The output and its format are also set by the user, e.g., single defect classification with location details. User can choose which model to use from the repository, e.g., Single shot detector (SSD) model. For ensuring high quality images, the user may or may not apply two filters to check for the aspect ratio and pixel range, depending upon time and computational constraints. This mode offers user-personalization. The system may work on its own for long hours, user may reserve any images for decision-making as per requirement. This mode facilitates prediction on variable input size. The prediction is more reliable since we perform quality check images input to the model. It is user-centric.

[0083] FIG. 7 is a flow chart (700) illustrating a method for processing a semiconductor wafer image based on an adaptive operating mode, according to embodiments as disclosed herein. The operations (302-306) are performed by the semiconductor wafer image processing controller (140).

[0084] Referring to the FIG. 7, At 702, the method includes receiving the input image with the user set size. At 704, the method includes using the interpolator changing or modifying the input size to the model requirement based on the selected input (e.g., operating modes, user input or the like). At 706, the method includes performing the quality check of the input image. At 708, the method includes configuring the WDD adaptive model to process the input image. At 710, the method includes detecting the wafer defect using the WDD adaptive model.

[0085] The images of variable sizes are acceptable, e.g., 52x52, 26x26, 42x44, etc. The system automatically determines the size and checks if it is in the acceptable range. The system determines if input size requirements match one or more of the models, if not the image is interpolated. The electronic device (100) performs quality check to see if ROI is present and if at noise etc. is present, it is within the acceptable range. Poor quality images within the acceptable range are sent in for restoration. If restored images satisfy the quality filters, it is processed further. The system can check if any additional inputs or features are available, e.g., scene graph, sensor data etc. Based on the inputs, the best model is loaded. This mode facilitates prediction on variable input size. The prediction is more reliable since the electronic device (100) performs quality check images input to the model. It is adaptive to various inputs available.

[0086] FIG. 8a and FIG. 8b are example illustrations (800a and 800b) in which an integrated image quality assurance steps for a removal of undesirable, poor quality images using the filter (160), according to embodiments as disclosed herein.

[0087] The FIG. 8a depicts the images suitable for prediction. The FIG. 8b depicts the images which may meet size requirements but are not suitable for prediction. The knowledge-guided filter determines the input dimensions have to fall within an acceptable ratio of width / height, expert determined based on data analysis / domain knowledge / observation. For the knowledge-guided filter, the number of defect and normal dies should be within a standard or expert determined based on data analysis / domain knowledge / observation for input images. The knowledge-guided filter identifies the circular region of interest of the image.

[0088] FIG. 9 is an example illustration (900) in which restoration and preprocessing of acceptable quality images to meet standard requirements are depicted, according to embodiments as disclosed herein.

[0089] The images with the minor issues which pass the other quality assurance filter are passed through the restoration unit (140g). The restoration unit (140g) provides the high quality restored images for the further use using various techniques (e.g., de-noising techniques, fuzzy based filter techniques or the like). The high quality restored images are analysed by the prediction model.

[0090] FIG. 10 is an example illustration in which prediction using model trained for addressing inherent data imbalance - balanced deep learning based data augmentation to balance underrepresented defect classes are depicted, according to embodiments as disclosed herein.

[0091] At 1002, all mixed defect type data trained using balanced number of examples for obtaining highly accurate predictions. At 1004, additional data obtained for defect detection related prediction model development / analysis. At 1006, the deep learning model analyses the all mixed defect type data trained using balanced number of examples for obtaining highly accurate predictions and Additional data obtained for defect detection related prediction model development / analysis. At 1008, the method includes processing the data with no label using the deep learning model. At 1010, the deep learning model provides the data with label.

[0092] In the proposed method, mixed data comprising of single and multiple defects are used of various dimensions right from the beginning. Such mixed but acceptable range of dimensions introduces variations in the model learning during the training and validation phases

[0093] The various actions, acts, blocks, steps, or the like in the flow charts (300 to 700) may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some of the actions, acts, blocks, steps, or the like may be omitted, added, modified, skipped, or the like without departing from the scope of the invention.

[0094] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and / or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the scope of the embodiments as described herein.

Claims

1. A method for processing a semiconductor wafer image, comprising: receiving, by an electronic device (100), at least one semiconductor wafer image; determining, by the electronic device (100), at least one operating mode to process the at least one semiconductor wafer image; and predicting, by the electronic device (100), at least one of defect of the at least one semiconductor wafer image and a quality of the at least one semiconductor wafer image based on the at least one determined operating mode.

2. The method as claimed in claim 1, wherein at least one of the defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image is predicted using a data driven model, wherein process the at least one semiconductor wafer image comprises at least one of a discard the at least one semiconductor wafer image that has unacceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the at least one semiconductor wafer image and accept the at least one semiconductor wafer image that has acceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the at least one semiconductor wafer image.

3. The method as claimed in claim 2, wherein the data driven model is trained by analysing and learning a plurality of mixed defect type data associated with a plurality of semiconductor wafer images over a period of time to configure a defect detection related prediction model, wherein the defect detection related prediction model predicts at least one defect type data associated with the plurality of semiconductor wafer images.

4. The method as claimed in claim 1, wherein the at least one operating mode comprises at least one of a standard operating mode, an advanced operating mode, a user-set operating mode, and an adaptive operating mode.

5. The method as claimed in claim 4, wherein the standard operating mode predicts at least one of the defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image based on quality of the at least one semiconductor wafer image, wherein the standard operating mode receives the at least one semiconductor wafer image with same size.

6. The method as claimed in claim 4, wherein the advanced operating mode receives the at least one semiconductor wafer image with variable sizes, wherein the advanced operating mode uses at least one filter (160) to determine an aspect ratio of the at least one semiconductor wafer image and a pixel range of the at least one semiconductor wafer image to predict at least one of the defect of the at least one semiconductor wafer image with the variable sizes and the quality of the at least one semiconductor wafer image with the variable sizes.

7. The method as claimed in claim 4, wherein the adaptive operating mode receives the at least one semiconductor wafer image with variable sizes, wherein the advanced operating mode determines at least one of a variable size of the at least one semiconductor wafer image in a predefined range and region of interest (ROI) of the at least one semiconductor wafer image in a predefined range, identifies that the at least one semiconductor wafer image is not in the predefined range, performs a restoration operation on the at least one identified semiconductor wafer image and determines that the at least one restored semiconductor wafer image is in the predefined range, wherein the adaptive operating mode predicts at least one of the defect of the at least one semiconductor wafer image with the variable sizes and the quality of the at least one semiconductor wafer image with the variable sizes based on at least one feature of the at least one semiconductor wafer image, wherein the at least one feature comprises a scene graph, a temperature, a noise, and a sensor data.

8. The method as claimed in claim 4, wherein the user-set operating mode receives the at least one semiconductor wafer image with variable sizes along with user suggestion, wherein the advanced operating mode uses at least one filter (160) to determine an aspect ratio of the at least one semiconductor wafer image and a pixel range of the at least one semiconductor wafer image to predict at least one of the defect of the at least one semiconductor wafer image with variable sizes and the quality of the at least one semiconductor wafer image with variable sizes based on the user suggestion.

9. The method as claimed in claim 6 or claim 8, wherein the at least one filter (160) removes unwanted image of the at least one semiconductor wafer image and ensure that of the at least one semiconductor wafer image comprises desired region of interest, wherein the at least one filter (160) is determined based on at least one of size of the the at least one semiconductor wafer image, a quality of the the at least one semiconductor wafer image, an aspect ratio of the the at least one semiconductor wafer image and a region of interest of the at least one semiconductor wafer image.

10. The method as claimed in claim 1, wherein the at least one operating mode is configured in a multi-model repository, wherein the multi-model repository comprises at least one of a single defect detector, a multiple defect detector, a bounding box detector, a defect percentage detector, an ensemble detector, a hybrid defect detector, and a multi-modal input based detector, wherein the single defect detector detects whether the at least one semiconductor wafer image contains defect or not, wherein the multiple defect detector detects one or more defects on the at least one semiconductor wafer image with details of defect types, wherein the bounding box detector detects precise location of defects on the at least one semiconductor wafer image, wherein the defect percentage detector detects intensity of defect on at least one semiconductor wafer image, wherein the ensemble detector uses one or more models to perform defect detection and uses some form of voting to arrive at reliable decision, wherein the hybrid defect detector uses a mixed architecture for defect detection, and wherein the multi-modal input based detector uses at least one architecture that is suitable for accepting multiple inputs of different modalities.

11. The method as claimed in claim 1, wherein the electronic device (100) receives at least one recommendation for predicting the at least one of defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image based on the at least one determined operating mode, an user history, an available input and defect information associated with the at least one semiconductor wafer image, wherein the electronic device (100) receives at least one recommendation for the at least one operating mode.

12. An electronic device (100) for processing a semiconductor wafer image, comprising: a processor (110), a memory (130); and a semiconductor wafer image processing controller (140), coupled with the processor (110) and the memory (130), configured to: receive at least one semiconductor wafer image; determine at least one operating mode to process the at least one semiconductor wafer image; and predict at least one of defect of the at least one semiconductor wafer image and a quality of the at least one semiconductor wafer image based on the at least one determined operating mode.

13. The electronic device (100) as claimed in claim 12, wherein at least one of the defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image is predicted using a data driven model, wherein process the at least one semiconductor wafer image comprises at least one of a discard the at least one semiconductor wafer image that has unacceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the at least one semiconductor wafer image and accept the at least one semiconductor wafer image that has acceptable image aspect ratio, an image size, an image quality and lack of region-of-interest associated with the at least one semiconductor wafer image.

14. The electronic device (100) as claimed in claim 13, wherein the data driven model is trained by analysing and learning a plurality of mixed defect type data associated with a plurality of semiconductor wafer images over a period of time to configure a defect detection related prediction model, wherein the defect detection related prediction model predicts at least one defect type data associated with the plurality of semiconductor wafer images.

15. The electronic device (100) as claimed in claim 12, wherein the at least one operating mode comprises at least one of a standard operating mode, an advanced operating mode, a user-set operating mode, and an adaptive operating mode.

16. The electronic device (100) as claimed in claim 15, wherein the standard operating mode predicts at least one of the defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image based on quality of the at least one semiconductor wafer image, wherein the standard operating mode receives the at least one semiconductor wafer image with same size.

17. The electronic device (100) as claimed in claim 15, wherein the advanced operating mode receives the at least one semiconductor wafer image with variable sizes, wherein the advanced operating mode uses at least one filter (160) to determine an aspect ratio of the at least one semiconductor wafer image and a pixel range of the at least one semiconductor wafer image to predict at least one of the defect of the at least one semiconductor wafer image with the variable sizes and the quality of the at least one semiconductor wafer image with the variable sizes.

18. The electronic device (100) as claimed in claim 15, wherein the adaptive operating mode receives the at least one semiconductor wafer image with variable sizes, wherein the advanced operating mode determines at least one of a variable size of the at least one semiconductor wafer image in a predefined range and region of interest (ROI) of the at least one semiconductor wafer image in a predefined range, identifies that the at least one semiconductor wafer image is not in the predefined range, performs a restoration operation on the at least one identified semiconductor wafer image and determines that the at least one restored semiconductor wafer image is in the predefined range, wherein the adaptive operating mode predicts at least one of the defect of the at least one semiconductor wafer image with the variable sizes and the quality of the at least one semiconductor wafer image with the variable sizes based on at least one feature of the at least one semiconductor wafer image, wherein the at least one feature comprises a scene graph, a temperature, a noise, and a sensor data.

19. The electronic device (100) as claimed in claim 15, wherein the user-set operating mode receives the at least one semiconductor wafer image with variable sizes along with user suggestion, wherein the advanced operating mode uses at least one filter (160) to determine an aspect ratio of the at least one semiconductor wafer image and a pixel range of the at least one semiconductor wafer image to predict at least one of the defect of the at least one semiconductor wafer image with variable sizes and the quality of the at least one semiconductor wafer image with variable sizes based on the user suggestion.

20. The electronic device (100) as claimed in claim 17 or claim 19, wherein the at least one filter (160) removes unwanted image of the at least one semiconductor wafer image and ensure that of the at least one semiconductor wafer image comprises desired region of interest, , wherein the at least one filter (160) is determined based on at least one of size of the the at least one semiconductor wafer image, a quality of the the at least one semiconductor wafer image, an aspect ratio of the the at least one semiconductor wafer image and a region of interest of the at least one semiconductor wafer image.

21. The electronic device (100) as claimed in claim 12, wherein the at least one operating mode is configured in a multi-model repository, wherein the multi-model repository comprises at least one of a single defect detector, a multiple defect detector, a bounding box detector, a defect percentage detector, an ensemble detector, a hybrid defect detector, and a multi-modal input based detector, wherein the single defect detector detects whether the at least one semiconductor wafer image contains defect or not, wherein the multiple defect detector detects one or more defects on the at least one semiconductor wafer image with details of defect types, wherein the bounding box detector detects precise location of defects on the at least one semiconductor wafer image, wherein the defect percentage detector detects intensity of defect on at least one semiconductor wafer image, wherein the ensemble detector uses one or more models to perform defect detection and uses some form of voting to arrive at reliable decision, wherein the hybrid defect detector uses a mixed architecture for defect detection, and wherein the multi-modal input based detector uses at least one architecture that is suitable for accepting multiple inputs of different modalities.

22. The electronic device (100) as claimed in claim 12, wherein the electronic device (100) receives at least one recommendation for predicting the at least one of defect of the at least one semiconductor wafer image and the quality of the at least one semiconductor wafer image based on the at least one determined operating mode, an user history, an available input and defect information associated with the at least one semiconductor wafer image, wherein the electronic device (100) receives at least one recommendation for the at least one operating mode.