Hotspot detection based on machine learning
View Patent ↗Aspects of the invention relate to machine-learning-based hotspot detection techniques. These hotspot detection techniques employ machine learning models constructed using two feature encoding schemes. When two-level machine learning methods are also employed, a total four machine learning models are constructed: scheme-one level-one, scheme-one level-two, scheme-two level-one and scheme-two level-two. The four models are applied to test patterns to derive scheme-one hotspot information and scheme-two hotspot information, which are then used to determine final hotspot information.
1. A method of machine learning-based hotspot detection, executed by at least one processor of a computer, comprising:
receiving data of layout patterns of a layout design;
determining, by using the at least one processor of a computer, scheme-one hotspot information by classifying the layout patterns based on one or more scheme-one machine learning models;
determining scheme-two hotspot information by classifying the layout patterns based on one or more scheme-two machine learning models, the one or more scheme-two machine learning models and the one or more scheme-one machine learning models both employing a same density-based encoding method and a same set of training patterns, but being constructed differently by placing, while encoding each training pattern in the set of training patterns, a context window in different positions around the each training pattern;
determining final hotspot information based on the scheme-one hotspot information and the scheme-two hotspot information; and
storing the final hotspot information in a tangible processor-accessible medium, or displaying the final hotspot information on a tangible medium.
2. The method recited in claim 1 , wherein the context windows used by the one or more scheme-two machine learning models and the one or more scheme-one machine learning models differ in their horizontal positions by a half of a horizontal dimension of a sliding window and in their vertical locations by a half of a vertical dimension of the sliding window.
3. The method recited in claim 1 , wherein the one or more scheme-one machine learning models and the one or more scheme-two machine learning models are support vector machine models.
4. The method recited in claim 1 , wherein each of the one or more scheme-one machine learning models and the one or more scheme-two machine learning models comprises a level-one model and a level-two model, wherein the level-one model is used to classify the layout patterns into potential non-hotspots and potential hotspots, the level-two model is used to identify false positives and hotspots in the potential hotspots.
5. The method recited in claim 1 , wherein the one or more scheme-one machine learning models and the one or more scheme-two machine learning models both employ a two-level machine learning method, wherein the two-level machine learning method employs a level-one model and a level-two model, the level-one model being used to classify the layout patterns into potential non-hotspots and potential hotspots, the level-two model being used to identify false positives and hotspots in the potential hotspots.
6. The method recited in claim 1 , wherein the data of layout patterns are derived from a global density database that is generated by using a global density pre-computation technique.
7. The method recited in claim 1 , wherein the determining final hotspot information comprises: labeling layout patterns that are labeled as hotspots in either the scheme-one hotspot information or the scheme-two hotspot information.
8. A hotspot detection system, comprising:
a scheme-one machine learning classification unit comprising one or more processors and being configured to generate scheme-one hotspot information by classifying layout patterns based one or more scheme-one machine learning models;
a scheme-two machine learning classification unit configured to generate scheme-two hotspot information by classifying the layout patterns based one or more scheme-two machine learning models, the one or more scheme-two machine learning models and the one or more scheme-one machine learning models both employing a same density-based encoding method and a same set of training patterns, but being constructed differently by placing, while encoding each training pattern in the set of training patterns, a context window in different positions around the each training pattern; and
a hotspot determination unit configured to determine final hotspot information based on the scheme-one hotspot information and the scheme-two hotspot information.
9. The hotspot detection system recited in claim 8 , wherein the context windows used by the one or more scheme-two machine learning models and the one or more scheme-one machine learning models differ in their horizontal positions by a half of a horizontal dimension of a sliding window and in their vertical locations by a half of a vertical dimension of the sliding window.
10. The hotspot detection system recited in claim 8 , wherein the scheme-one machine learning classification unit and the scheme-two machine learning classification unit both employ a two-level machine learning method, wherein the two-level machine learning method employs a level-one model and a level-two model, the level-one model being used to classify the layout patterns into potential non-hotspots and potential hotspots, the level-two model being used to identify false positives and hotspots in the potential hotspots.
11. A processor-readable device storing processor-executable instructions for causing one or more processors to perform a method of machine learning-based hotspot detection, the method comprising:
receiving data of layout patterns of a layout design;
determining scheme-one hotspot information by classifying the layout patterns based on one or more scheme-one machine learning models;
determining scheme-two hotspot information by classifying the layout patterns based on one or more scheme-two machine learning models, the one or more scheme-two machine learning models and the one or more scheme-one machine learning models both employing a same density-based encoding method and a same set of training patterns, but being constructed differently by placing, while encoding each training pattern in the set of training patterns, a context window in different positions around the each training pattern;
determining final hotspot information based on the scheme-one hotspot information and the scheme-two hotspot information; and
storing the final hotspot information in a tangible processor-accessible medium, or displaying the final hotspot information on a tangible medium.
12. The processor-readable device recited in claim 11 , wherein the context windows used by the one or more scheme-two machine learning models and the one or more scheme-one machine learning models differ in their horizontal positions by a half of a horizontal dimension of a sliding window and in their vertical locations by a half of a vertical dimension of the sliding window.
13. The processor-readable device recited in claim 11 , wherein each of the one or more scheme-one machine learning models and the one or more scheme-two machine learning models comprises a level-one model and a level-two model, wherein the level-one model is used to classify the layout patterns into potential non-hotspots and potential hotspots, the level-two model is used to identify false positives and hotspots in the potential hotspots.
14. The processor-readable device recited in claim 11 , wherein the one or more scheme-one machine learning models and the one or more scheme-two machine learning models both employ a two-level machine learning method, wherein the two-level machine learning method employs a level-one model and a level-two model, the level-one model being used to classify the layout patterns into potential non-hotspots and potential hotspots, the level-two model being used to identify false positives and hotspots in the potential hotspots.