MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202641091291 A) filed by Dr. Bhikshagugulothu; Dr. G. Narasa Raju; Mr. Bathula Srinivas; Dr. S. Varalakshmi; Mr. S. Thimmarayan; Dr. B. Manikandan; Ch Naresh; Dr. N. Dhasarathan; Mr. Pranav Sagar Saladi; and Mr. Guttula Ramesh on July 27, 2026, for Deep Learning-Driven Autonomous Defect Inspection Framework For Semiconductor Manufacturing Systems.

Inventors include Dr. Bhikshagugulothu; Dr. G. Narasa Raju; Mr. Bathula Srinivas; Dr. S. Varalakshmi; Mr. S. Thimmarayan; Dr. B. Manikandan; Ch Naresh; Dr. N. Dhasarathan; Mr. Pranav Sagar Saladi; and Mr. Guttula Ramesh.

The application for the patent was published on July 31, 2026, under issue no. 31/2026.

Abstract: A deep learning-driven autonomous defect inspection framework is disclosed for semiconductor manufacturing systems. The framework comprises a multi-modal data acquisition module for receiving optical, scanning electron microscopy, and electrical parametric test data; a defect detection module employing a convolutional neural network and vision transformer ensemble to identify candidate defect regions; a defect classification module assigning taxonomy classifications and severity scores; and a self-supervised and active learning module that pretrains models on unlabelled wafer imagery and selects informative regions for expert annotation, reducing labelling burden. A multi-modal fusion module jointly reasons across imaging and electrical test data to reduce false positives and estimate yield impact, while a domain adaptation module enables transfer across fabrication tools and process nodes. An explainable artificial intelligence subsystem generates spatially localised attribution maps and natural language root-cause hypotheses, and a closed-loop process control interface transmits classification and root-cause information to upstream process control systems, enabling autonomous, continually self-improving, high-sensitivity defect inspection at production-compatible throughput.

Disclaimer: Curated by HT Syndication.