Predicting local layout effects in circuit design patterns
A method for predicting local layout effect in a circuit design pattern includes obtaining a plurality of circuit design patterns, generating layout images from the circuit design patterns, extracting feature vectors from the layout images by processing the layout images in a computer vision machine learning algorithm, comparing the feature vector extracted from a selected layout image to clusters of feature vectors extracted from the layout images, wherein the clusters of feature vectors include an in-range cluster and an outlier cluster, and labelling a circuit design pattern corresponding to the selected layout image, for which threshold voltage has not been experimentally measured, as being an in-range circuit design pattern or an outlier circuit design pattern, in response to the selected layout image respectively correlating with the in-range cluster or with the outlier cluster.
1. A method for rejecting a circuit design pattern that is predicted to have local layout effect, comprising:
obtaining a plurality of circuit design patterns;
generating a plurality of layout images each corresponding to one of the plurality of circuit design patterns;
training a computer vision machine learning algorithm in an unsupervised fashion to generate a training set of feature vectors for the plurality of layout images;
generating a cluster plot of the training set of feature vectors;
identifying an in-range cluster and an outlier cluster on the cluster plot;
identifying a circuit design pattern that corresponds to a layout image that produced a feature vector within the outlier cluster; and
rejecting the identified circuit design pattern.
2. The method of claim 1 further comprising:
fabricating samples of circuit design patterns that correspond to feature vectors present in each of the clusters; and
identifying the in-range cluster and the outlier cluster based on experimental measurements of threshold voltage for samples that are present in each of the clusters, wherein experimental measurements for samples in the outlier cluster indicate local layout effect.
3. The method of claim 2 further comprising using principal component analysis for dimensionality reduction of the training set of feature vectors.
4. The method of claim 1 further comprising:
clustering the feature vectors; and
identifying the in-range cluster and the outlier cluster based on the in-range cluster having more members than the outlier cluster.
5. The method of claim 1 wherein partitional clustering is used to cluster the feature vectors.
6. The method of claim 1 wherein the computer vision machine learning algorithm includes an autoencoder that extracts the feature vectors as intermediate codings from the layout images.
7. A method for rejecting circuit design patterns that are predicted to have local layout effect, comprising:
obtaining a plurality of circuit design patterns;
generating a plurality of layout images each corresponding to one of the plurality of circuit design patterns;
training a computer vision machine learning algorithm in an unsupervised fashion to generate a training set of feature vectors for the plurality of layout images;
generating a cluster plot of the training set of feature vectors;
identifying an in-range cluster and an outlier cluster on the cluster plot;
obtaining an inference set of circuit design patterns;
generating an inference set of layout images each corresponding to one of the inference set of circuit design patterns;
using the computer vision machine learning algorithm to generate an inference set of feature vectors for the inference set of layout images;
plotting on the cluster plot the inference set of feature vectors;
identifying outlier members of the inference set of layout images that produced feature vectors that overlie the outlier cluster on the cluster plot; and
rejecting circuit design patterns that correspond to the outlier members of the inference set of layout images.
8. The method of claim 7 wherein identifying the in-range cluster and the outlier cluster includes fabricating at least one sample of a circuit design pattern from each of the clusters and experimentally measuring threshold voltage of the at least one sample from each of the clusters, wherein experimental measurements for samples in the outlier cluster indicate local layout effect.
9. The method of claim 7 wherein identifying the in-range cluster and the outlier cluster includes counting a number of members in each of the clusters.
10. The method of claim 7 wherein the cluster plot is a 2-dimensional cluster plot produced by reducing dimensionality of the training set of feature vectors, and dimensionality of the inference set of feature vectors is reduced before plotting the inference set of feature vectors on the cluster plot.
11. The method of claim 7 wherein partitional clustering is used to identify the in-range and outlier clusters.
12. The method of claim 7 wherein the computer vision machine learning algorithm includes a convolutional neural network.
13. The method of claim 7 wherein the computer vision machine learning algorithm includes an autoencoder that extracts the feature vectors as intermediate codings from the layout images.
14. A method for rejecting a circuit design pattern that is predicted to have local layout effect, comprising:
obtaining a plurality of circuit design patterns;
generating layout images from the circuit design patterns;
extracting feature vectors from the layout images by processing the layout images in a computer vision machine learning algorithm;
comparing the feature vector extracted from a selected layout image to clusters of feature vectors extracted from the layout images, wherein the clusters of feature vectors include an in-range cluster and an outlier cluster; and
rejecting a circuit design pattern corresponding to the selected layout image, for which threshold voltage has not been experimentally measured, as being a circuit design pattern that is predicted to have local layout effect, in response to the feature vector of the selected layout image correlating with the outlier cluster.
15. The method of claim 14 further comprising:
fabricating samples of at least some of the circuit design patterns; and
identifying the in-range cluster and the outlier cluster based on experimental measurements of threshold voltage for samples that are present in each of the clusters, wherein experimental measurements for samples in the outlier cluster indicate local layout effect.
16. The method of claim 14 further comprising:
clustering the feature vectors; and
identifying the in-range cluster and the outlier cluster based on the in-range cluster having more members than the outlier cluster.
17. The method of claim 16 wherein partitional clustering is used to cluster the feature vectors.
18. The method of claim 14 wherein the computer vision machine learning algorithm includes a convolutional neural network.
19. The method of claim 18 wherein the computer vision machine learning algorithm is trained in a supervised fashion based on measured values of threshold voltage for a majority of the circuit design patterns.
20. The method of claim 14 wherein the computer vision machine learning algorithm includes an autoencoder that extracts the feature vectors as intermediate codings from the layout images.