Abstract: Methods and systems for accelerated training of a machine learning based model for semiconductor applications are provided. One method for training a machine learning based model includes acquiring information for non-nominal instances of specimen(s) on which a process is performed. The machine learning based model is configured for performing simulation(s) for the specimens. The machine learning based model is trained with only information for nominal instances of additional specimen(s). The method also includes re-training the machine learning based model with the information for the non-nominal instances of the specimen(s) thereby performing transfer learning of the information for the non-nominal instances of the specimen(s) to the machine learning based model.

Abstract: An explanation of a detection/classification algorithm made using a deep learning neural network clarifies the results that are formed and helps a user to identify the root cause of defect detection/classification model performance issues. A relevance map is determined based on a layer-wise relevance propagation algorithm. A mean intersection over union score between the relevance map and a ground truth is determined. A part of one of the semiconductor images that contributed to the classification using the deep learning model based on the relevance map and the mean intersection over union score is determined.

Abstract: Real-time job distribution software architectures for high bandwidth, hybrid processor computation systems for semiconductor inspection and metrology are disclosed. The imaging processing computer architecture can be scalable by changing the number of CPUs and GPUs to meet computing needs. The architecture is defined using a master node and one or more worker nodes to run image processing jobs in parallel for maximum throughput. The master node can receive input image data from a semiconductor wafer or reticle. Jobs based on the input image data are distributed to one of the worker nodes. Each worker node can include at least one CPU and at least one GPU. The image processing job can contain multiple tasks, and each of the tasks can be assigned to one of the CPU or GPU in the worker node using a worker job manager to process the image.

Abstract: Methods and systems for training a machine learning model using synthetic defect images are provided. One system includes one or more components executed by one or more computer subsystems. The one or more components include a graphical user interface (GUI) configured for displaying one or more images for a specimen and image editing tools to a user and for receiving input from the user that includes one or more alterations to at least one of the images using one or more of the image editing tools. The component(s) also include an image processing module configured for applying the alteration(s) to the at least one image thereby generating at least one modified image and storing the at least one modified image in a training set. The computer subsystem(s) are configured for training a machine learning model with the training set in which the at least one modified image is stored.

Abstract: Autoencoder-based, semi-supervised approaches are used for anomaly detection. Defects on semiconductor wafers can be discovered using these approaches. The model can include a variational autoencoder, such as a one that includes ladder networks. Defect-free or clean images can be used to train the model that is later used to discover defects or other anomalies.

Abstract: Methods and systems for detecting and classifying defects on a specimen are provided. One system includes one or more components executed by one or more computer subsystems. The one or more components include a neural network configured for detecting defects on a specimen and classifying the defects detected on the specimen. The neural network includes a first portion configured for determining features of images of the specimen generated by an imaging subsystem. The neural network also includes a second portion configured for detecting defects on the specimen based on the determined features of the images and classifying the defects detected on the specimen based on the determined features of the images.

Abstract: Defect classification includes acquiring one or more images of a specimen, receiving a manual classification of one or more training defects based on one or more attributes of the one or more training defects, generating an ensemble learning classifier based on the received manual classification and the attributes of the one or more training defects, generating a confidence threshold for each defect type of the one or more training defects based on a received classification purity requirement, acquiring one or more images including one or more test defects, classifying the one or more test defects with the generated ensemble learning classifier, calculating a confidence level for each of the one or more test defects with the generated ensemble learning classifier and reporting one or more test defects having a confidence level below the generated confidence threshold via the user interface device for manual classification.

Abstract: Methods and systems for training a machine learning model using synthetic defect images are provided. One system includes one or more components executed by one or more computer subsystems. The one or more components include a graphical user interface (GUI) configured for displaying one or more images for a specimen and image editing tools to a user and for receiving input from the user that includes one or more alterations to at least one of the images using one or more of the image editing tools. The component(s) also include an image processing module configured for applying the alteration(s) to the at least one image thereby generating at least one modified image and storing the at least one modified image in a training set. The computer subsystem(s) are configured for training a machine learning model with the training set in which the at least one modified image is stored.

Abstract: Autoencoder-based, semi-supervised approaches are used for anomaly detection. Defects on semiconductor wafers can be discovered using these approaches. The model can include a variational autoencoder, such as a one that includes ladder networks. Defect-free or clean images can be used to train the model that is later used to discover defects or other anomalies.

Abstract: Methods and systems for detecting defects in patterns formed on a specimen are provided. One system includes one or more components executed by one or more computer subsystems, and the component(s) include first and second learning based models. The first learning based model generates simulated contours for the patterns based on a design for the specimen, and the simulated contours are expected contours of a defect free version of the patterns in images of the specimen generated by an imaging subsystem. The second learning based model is configured for generating actual contours for the patterns in at least one acquired image of the patterns formed on the specimen. The computer subsystem(s) are configured for comparing the actual contours to the simulated contours and detecting defects in the patterns formed on the specimen based on results of the comparing.

Abstract: Methods and systems for detecting and classifying defects on a specimen are provided. One system includes one or more components executed by one or more computer subsystems. The one or more components include a neural network configured for detecting defects on a specimen and classifying the defects detected on the specimen. The neural network includes a first portion configured for determining features of images of the specimen generated by an imaging subsystem. The neural network also includes a second portion configured for detecting defects on the specimen based on the determined features of the images and classifying the defects detected on the specimen based on the determined features of the images.

Abstract: Methods and systems for detecting defects on a specimen are provided. One system includes a generative model. The generative model includes a non-linear network configured for mapping blocks of pixels of an input feature map volume into labels. The labels are indicative of one or more defect-related characteristics of the blocks. The system inputs a single test image into the generative model, which determines features of blocks of pixels in the single test image and determines labels for the blocks based on the mapping. The system detects defects on the specimen based on the determined labels.

Abstract: Real-time job distribution software architectures for high bandwidth, hybrid processor computation systems for semiconductor inspection and metrology are disclosed. The imaging processing computer architecture can be scalable by changing the number of CPUs and GPUs to meet computing needs. The architecture is defined using a master node and one or more worker nodes to run image processing jobs in parallel for maximum throughput. The master node can receive input image data from a semiconductor wafer or reticle. Jobs based on the input image data are distributed to one of the worker nodes. Each worker node can include at least one CPU and at least one GPU. The image processing job can contain multiple tasks, and each of the tasks can be assigned to one of the CPU or GPU in the worker node using a worker job manager to process the image.

Abstract: Methods and systems for detecting defects in patterns formed on a specimen are provided. One system includes one or more components executed by one or more computer subsystems, and the component(s) include first and second learning based models. The first learning based model generates simulated contours for the patterns based on a design for the specimen, and the simulated contours are expected contours of a defect free version of the patterns in images of the specimen generated by an imaging subsystem. The second learning based model is configured for generating actual contours for the patterns in at least one acquired image of the patterns formed on the specimen. The computer subsystem(s) are configured for comparing the actual contours to the simulated contours and detecting defects in the patterns formed on the specimen based on results of the comparing.

Abstract: Defect classification includes acquiring one or more images of a specimen, receiving a manual classification of one or more training defects based on one or more attributes of the one or more training defects, generating an ensemble learning classifier based on the received manual classification and the attributes of the one or more training defects, generating a confidence threshold for each defect type of the one or more training defects based on a received classification purity requirement, acquiring one or more images including one or more test defects, classifying the one or more test defects with the generated ensemble learning classifier, calculating a confidence level for each of the one or more test defects with the generated ensemble learning classifier and reporting one or more test defects having a confidence level below the generated confidence threshold via the user interface device for manual classification.

Abstract: Defect classification includes acquiring one or more images of a specimen including multiple defects, grouping the defects into groups of defect types based on the attributes of the defects, receiving a signal from a user interface device indicative of a first manual classification of a selected number of defects from the groups, generating a classifier based on the first manual classification and the attributes of the defects, classifying, with the classifier, one or more defects not manually classified by the manual classification, identifying the defects classified by the classifier having the lowest confidence level, receiving a signal from the user interface device indicative of an additional manual classification of the defects having the lowest confidence level, determining whether the additional manual classification identifies one or more additional defect types not identified in the first manual classification, and iterating the procedure until no new defect types are found.

Abstract: Defect classification includes acquiring one or more images of a specimen, receiving a manual classification of one or more training defects based on one or more attributes of the one or more training defects, generating an ensemble learning classifier based on the received manual classification and the attributes of the one or more training defects, generating a confidence threshold for each defect type of the one or more training defects based on a received classification purity requirement, acquiring one or more images including one or more test defects, classifying the one or more test defects with the generated ensemble learning classifier, calculating a confidence level for each of the one or more test defects with the generated ensemble learning classifier and reporting one or more test defects having a confidence level below the generated confidence threshold via the user interface device for manual classification.

Abstract: Methods and systems for generating inspection results for a specimen with an adaptive nuisance filter are provided. One method includes selecting a portion of events detected during inspection of a specimen having values for at least one feature of the events that are closer to at least one value of at least one parameter of the nuisance filter than the values for at least one feature of another portion of the events. The method also includes acquiring output of an output acquisition subsystem for the sample of events, classifying the events in the sample based on the acquired output, and determining if one or more parameters of the nuisance filter should be modified based on results of the classifying. The nuisance filter or the modified nuisance filter can then be applied to results of the inspection of the specimen to generate final inspection results for the specimen.

Abstract: Methods and systems for accelerated training of a machine learning based model for semiconductor applications are provided. One method for training a machine learning based model includes acquiring information for non-nominal instances of specimen(s) on which a process is performed. The machine learning based model is configured for performing simulation(s) for the specimens. The machine learning based model is trained with only information for nominal instances of additional specimen(s). The method also includes re-training the machine learning based model with the information for the non-nominal instances of the specimen(s) thereby performing transfer learning of the information for the non-nominal instances of the specimen(s) to the machine learning based model.

Abstract: Methods and systems for detecting defects on a specimen are provided. One system includes a generative model. The generative model includes a non-linear network configured for mapping blocks of pixels of an input feature map volume into labels. The labels are indicative of one or more defect-related characteristics of the blocks. The system inputs a single test image into the generative model, which determines features of blocks of pixels in the single test image and determines labels for the blocks based on the mapping. The system detects defects on the specimen based on the determined labels.