In-Line Scatterometry for Semiconductor Hotspot Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes, such as Chemical-Mechanical Polishing, can lead to defects like dishing and metal residuals, causing 'hotspots' that are difficult to detect until electrical testing, which is time-consuming, expensive, and often destructive.
Innovation Solution
A method using machine learning to predict manufacturing outcomes by acquiring and analyzing scatterometric signatures and manufacturing outcome data from training wafers, allowing for early detection of defects and potential hotspots during the manufacturing process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If electrical testing is performed to detect hotspots, then manufacturing defects can be identified, but the process is time-consuming and expensive
Solution Approach 1:
The patent applies preliminary action by performing scatterometric measurements during the manufacturing process to detect hotspots before electrical testing. Training wafers are measured after CMP steps to capture scatterometric signatures, and a prediction model is trained to identify hotspots in advance, eliminating the need for time-consuming post-manufacturing electrical testing.
Solution Approach 2:
The patent replaces the mechanical/electrical testing system with an optical measurement system. Instead of using electrical probes to detect defects after manufacturing, the system uses scatterometric measurements (optical techniques) to detect hotspots during manufacturing, significantly reducing testing time and cost.
2Loss of time
If in-line tests are performed during manufacturing to detect hotspots, then early detection is possible, but the tests are slow and expensive
Solution Approach 1:
The patent uses training wafers as copies or representatives of production wafers. Scatterometric measurements are performed on training wafers during manufacturing to build a prediction model, which is then applied to production wafers. This copying approach enables fast prediction without performing slow, expensive tests on every production wafer.
Solution Approach 2:
The system performs preliminary measurements on training wafers during manufacturing to capture scatterometric signatures and train the prediction model. Once trained, the model can rapidly predict hotspots in production wafers, enabling fast detection without repeating the slow measurement process on each production unit.
3Reliability
If scatterometric measurements are performed on training wafers during manufacturing, then a prediction model can be trained, but additional measurement steps are required
Solution Approach 1:
The patent makes the scatterometric measurement system multi-functional by using it for both training purposes and production monitoring. The same measurement tool and methodology are used to capture training data and to perform predictions on production wafers, eliminating the need for separate testing equipment and reducing overall system complexity.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables early detection and prevention of defects, reducing the need for costly and destructive testing, and allowing for real-time adjustments to the manufacturing process to prevent hotspot-related issues.
Implementation Method 1
a scatterometric measurement tool to acquire training scatterometric signatures from a plurality of training locations on the one or more training wafers
Data Source
AI summary
Controlling semiconductor device manufacture by acquiring training scatterometric signatures collected at training locations on training semiconductor wafers and corresponding to locations within a predefined design of a training semiconductor device, the training signatures collected after predefined processing steps during manufacture of the device on the training wafers, acquiring manufacturing outcome data associated with the training locations, training a prediction model using the training signatures and the manufacturing outcome data, and applying the prediction model to a candidate scatterometric signature to predict a manufacturing outcome, the candidate signature collected at a candidate location on a candidate semiconductor wafer, the candidate location corresponding to a location within the same predefined design of a candidate semiconductor device, the candidate signature collected after any of the processing steps during manufacture of the candidate device on the candidate wafer.