Multi-Model Metrology for CMP Layer Thickness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional metrology methods face challenges in accurately measuring strongly correlated thicknesses of materials like oxide and nitride during semiconductor processing, particularly near the end-point of chemical mechanical polishing (CMP) or etching processes.
Innovation Solution
The method involves using multiple models to simulate the sample before and after processing, where the top layer is either partially removed or considered as an independent variable, allowing for precise determination of desired parameters by constraining the underlying layers' parameters to be the same across models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional modeling techniques are used to measure layer thicknesses, then the measurement process is simple, but the measurement precision deteriorates for strongly correlated thicknesses near the end-point
Solution Approach 1:
The patent segments the measurement process into multiple distinct modeling approaches: a first model that varies both top and underlying layer thicknesses, and a second model that fixes the top layer thickness while varying only the underlying layer thickness. This segmentation allows the system to decouple the strongly correlated parameters and measure them independently, thereby improving measurement precision for strongly correlated thicknesses without overwhelming complexity
Solution Approach 2:
The patent introduces a new dimension to the measurement space by adding a second model that constrains the top layer thickness. This dimensional addition transforms the problem from a single correlated parameter space into a multi-dimensional space where the underlying layer thickness can be measured independently, effectively breaking the strong correlation that plagues conventional single-model approaches
2Measurement precision
If multiple models are used to measure strongly correlated thicknesses, then measurement precision improves, but the complexity of the measurement process increases
Solution Approach 1:
The patent uses the first model as an intermediary step to determine the top layer thickness before applying the second model to measure the underlying layer thickness. This intermediary approach allows the system to break down the complex measurement of strongly correlated parameters into sequential, more manageable steps, improving ease of operation while maintaining high measurement precision
3Adaptability or versatility
If the top layer thickness is varied in the model, then the model flexibility increases, but the ability to accurately measure the underlying layer thickness decreases due to strong correlation
Solution Approach 1:
The patent dynamically adjusts the modeling approach by using two different models with different degrees of freedom. The first model allows both top and underlying layer thicknesses to vary for initial characterization, while the second model dynamically constrains the top layer thickness to accurately measure the underlying layer. This dynamic adaptation resolves the contradiction between model flexibility and measurement precision
Data Source
AI summary
A sample that is processed to remove a top layer, e.g., using chemical mechanical polishing or etching, is accurately measured using multiple models of the sample. The multiple models may be constrained based on a pre-processing measurement of the sample. By way of example, the multiple models of the sample may be linked in pairs, where one pair includes a model simulating the pre-processed sample and another model simulating the post-processed sample with a portion of the top layer remaining, i.e., under-processing. Another pair of linked models includes a model simulating the pre-processed sample and a model simulating the post-processing sample with the top layer removed, i.e., the correct amount of processing or over-processing. The underlying layers in the linked model pairs are constrained to have the same parameters. The modeling process may use a non-linear regression or libraries.