Standard Cell Extraction from Raw IC Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The reverse engineering (RE) of semiconductor integrated circuits faces challenges in extracting standard cells from raw images due to high noise intensity, alignment and stitching issues, intra-cell similarity, and scale, especially when no prior knowledge of the IC design is available, leading to inaccurate results and high memory and time complexity.
Innovation Solution
A feature vector encoding mechanism and a set of model rules are used to analyze the contact layer image for cell boundary detection, allowing for the extraction of standard cells without prior knowledge of the IC design, utilizing common collector supply voltage (Vcc) lines and binarizing contact layer images to generate feature vectors and determine cell boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning algorithms are used to identify standard cells from raw SEM images, then cell identification can be performed, but accuracy deteriorates due to high noise intensity, alignment issues, and intra-cell similarity
Solution Approach 1:
The patent segments the standard cell identification process into multiple stages: raw image acquisition, preprocessing (noise reduction, alignment), feature extraction, and classification. By dividing the complex task into manageable segments with specialized processing for each, the system achieves higher accuracy in identifying cells despite noise and similarity challenges
Solution Approach 2:
The patent introduces an intermediary standard cell library containing pre-characterized cell patterns as a reference. This intermediary database mediates between the raw SEM images and the identification algorithm, providing known patterns for comparison and significantly improving identification accuracy by reducing the impact of noise and alignment variations
2Ease of manufacture
If a template standard cell library is required for extracting gates and fillers, then extraction can be performed, but the process becomes counterintuitive for reverse engineering where no prior knowledge should be needed
Solution Approach 1:
The patent performs preliminary actions by automatically generating a standard cell library from the device under test itself before the actual extraction process. This preliminary library generation step enables subsequent extraction operations to proceed with high processability while maintaining complete adaptability to different devices, as each device creates its own reference library
Solution Approach 2:
The system implements self-service by automatically creating its own standard cell library from the device being analyzed, without requiring external templates or prior knowledge. The device essentially serves itself by providing the reference patterns needed for its own extraction through automated analysis of its structure
3Measurement precision
If higher quality SEM scans are performed to reduce noise, then image quality improves, but imaging time increases to several days or months
Solution Approach 1:
The patent applies partial action by performing quality sufficient but not excessive scanning. Instead of maximizing scan quality to the extreme (which would take months), the system performs scans at a moderate quality level that provides sufficient signal-to-noise ratio for the extraction task, significantly reducing imaging time while maintaining adequate image quality for analysis
4Ease of manufacture
If cell boundary detection is performed using contact layer images, then standard cells can be extracted, but memory and time complexity become intractable at high integration levels
Solution Approach 1:
The patent extracts and focuses only on the essential features needed for cell boundary detection from the contact layer images, rather than processing the entire image data. By extracting only the relevant contact patterns and their spatial relationships, the system enables standard cell extraction while keeping memory and time complexity manageable even for highly integrated circuits
Data Source
AI summary
Methods and apparatus are provided for automatically extracting standard cells to form a standard cell library using raw multi-layer images of an IC. Accordingly, various embodiments involve: extracting the raw contact layer image from the raw multi-layer images; binarizing the raw contact layer image to generate a binarized contact layer image identifying a plurality of contact rows and a plurality of contact columns; determining a plurality of Vcc lines based on a subset of the plurality of contact rows having a periodic nature; extracting a plurality of binarized contact layer image strips from the binarized contact layer image; encoding each binarized contact layer image strip using feature vectors and column distance values; applying a model rule set to each encoded binarized contact layer image strip for detecting cell boundaries; extracting the standard cells based on the cell boundaries; and storing the extracted cells to form a standard cell candidate library.