IC Cell Localization via String Matrix Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for object localization in SEM images of integrated circuits are inadequate due to scaling issues and the absence of points of interest, leading to difficulties in navigating and locating target cells for analysis, especially when comparing design files to SEM images.
Innovation Solution
The method involves encoding cell footprints into two-dimensional string matrices and using a string search algorithm to identify candidate regions, followed by sliding a mask window for fine-matching to locate target cells, which preserves the topology of the cells and reduces computational effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual inspection methods are used to locate target cells in SEM images, then inspection accuracy can be maintained, but inspection efficiency deteriorates due to the extremely difficult manual search among millions of condensed standard cells
Solution Approach 1:
The patent transforms the image data into a different parameter space by encoding cell footprints into two-dimensional string matrices. This parameter transformation allows the use of efficient string search algorithms instead of traditional image processing methods, dramatically improving inspection efficiency while maintaining the ability to accurately locate target cells among millions of standard cells
Solution Approach 2:
The patent replaces manual mechanical inspection with an automated computational system. By substituting the manual search process with computer-based string search algorithms operating on encoded footprint matrices, the system achieves both high efficiency and accuracy in locating target cells without human intervention
2Loss of information
If detailed SEM images are captured to provide rich information for hardware assurance analysis, then information quality improves, but image size reduces making it challenging to navigate and locate the captured area from the entire IC
Solution Approach 1:
The patent adds a new dimensional layer by creating encoded string matrix representations of cell footprints. This additional dimension allows simultaneous access to both detailed local information (through the encoded footprint patterns) and global navigation context (through the string search algorithm's ability to locate patterns within the larger matrix structure), resolving the navigation challenge
3Reliability
If existing matching approaches using vias, contacts, and corners as points of interest are applied, then feature matching can be performed, but reliability deteriorates due to process variations causing corners to be absent in SEM images and imaging errors accumulating through deprocessing
Solution Approach 1:
The patent extracts the essential matching information directly from the cell footprint patterns themselves, removing the dependency on external reference features like vias, contacts, and corners. By encoding the footprint geometry into string matrices and performing direct pattern matching, the system achieves reliable matching that is immune to process variations and deprocessing errors
Solution Approach 2:
The patent creates encoded copies of cell footprint patterns in the form of two-dimensional string matrices. These encoded representations serve as reliable templates for matching that are invariant to the physical imperfections present in raw SEM images, such as missing corners or variations caused by manufacturing processes
Data Source
AI summary
Embodiments of the present disclosure provide methods, apparatus, systems, and computer program products for using an image of an integrated circuit (IC) including a plurality of cells to locate one or more target cells within the IC. Accordingly, in various embodiments, a footprint for each cell of the plurality of cells is encoded to transform the image of the IC into a two-dimensional string matrix. A string search algorithm is then applied on each encoded dopant region found in the two-dimensional string matrix using an encoded target layout cell to identify one or more candidate regions of interest within the image. Finally, a mask window is slid over each candidate region of interest while performing matching using match criteria to identify any target cells in the one or more target cells that are located within the candidate region of interest.