TL;DR
This paper introduces nDSE (nanoscale displacement sensing and estimation), a novel framework for achieving nanoscale overlay alignment in fabrication processes. By leveraging advanced image correlation techniques, nDSE enables precise alignment using arbitrary patterns, offering a cost-effective and simpler alternative to current industry standards.
What Problem Does This Paper Address?
Achieving precise alignment of multiple components or layers is critical in micro- and nanofabrication. Traditional overlay alignment methods often rely on specialized, high-precision alignment marks and can be limited by the wavelength of light or optical resolution. This becomes a significant bottleneck as fabrication scales down to the nanoscale, where even minute misalignments can lead to device failure. The authors identify a gap in current metrology and alignment techniques, particularly for nanoscale applications, which are often expensive, complex, and require specific, difficult-to-fabricate targets. They aim to provide a more versatile, cost-effective, and simpler solution.
Methods
The core of the paper lies in the development of the nanoscale displacement sensing and estimation (nDSE) framework. This framework is built upon two fundamental theories:
- Phase Delay Detection (PDD): This approach treats displacement as a phase difference in the frequency domain. It's related to Phase Correlation Method (PCM) but focuses on extracting displacement directly from the phase response, rather than relying solely on the inverse Fourier Transform to find a correlation peak. PDD aims to mask out noise and distortions to improve accuracy.
- Derivatives-based Maximum Likelihood Estimation (DML): This theory provides an error propagation model for nDSE, aiming to link the performance of the observation system and the target's information content to the theoretical maximum achievable accuracy.
The paper highlights a specific algorithm derived from these theories:
- Nearest Neighbor Navigation (N-Cubed) algorithm: This is a correlation-based algorithm that models an arbitrary correlation surface using a 2nd-order Taylor expansion. It provides a closed-form analytical solution for sub-pixel displacement estimation without iteration, making it robust and efficient. The authors claim it can achieve displacement accuracy down to 1/200th of a pixel.
The paper also introduces two alignment frameworks based on nDSE:
- DDMA (Direct Displacement-Measurement-based Alignment): This method directly uses pseudo-displacement estimation between nominally identical features on different substrates or layers. It's simpler but can be less accurate due to variations between features.
- IDMA (Indirect Displacement-Measurement-based Alignment): This method acquires alignment images of all substrates simultaneously and uses nDSE to track individual movements. By subtracting these trajectories, relative layer-to-layer positioning is determined. This is considered the more accurate, albeit potentially more complex, approach.
Datasets
Not discussed in the paper. The experiments described use custom-designed patterned silicon wafer fragments and nanoimprint molds.
Key Insights and Ideas
- Arbitrary Patterns as Alignment Targets: A major breakthrough is the demonstration that nDSE can work with virtually any pattern, not just specialized alignment marks. This significantly reduces the cost and complexity of target fabrication.
- Bridging Displacement and Alignment: The paper clearly defines displacement and alignment and proposes methods to bridge these concepts, particularly through the idea of a "reference" frame. This allows for the transformation of individual object movements into relative positioning measurements.
- Pseudo-Displacement for Practicality: The introduction of "pseudo-displacement" allows for alignment between features that are similar but not identical, making the approach more practical for real-world fabrication processes where perfect feature replication is challenging.
- Cost and Simplicity Advantages: The nDSE framework, particularly when combined with conventional optical microscopes, offers a significantly more cost-effective and simpler system design compared to existing methods.
- Versatility of nDSE: The underlying DSE theory and algorithms are presented as being applicable to various imaging systems (optical, SEM, etc.) and dimensionalities, making the framework highly versatile.
Key Results
- 1 nm precision: The authors report achieving a measurement standard deviation (one sigma) error of approximately 1 nm when tracking 5 nm steps with their nDSE algorithm.
- 5 nm step discernment: They demonstrated the ability to easily discern 5 nm steps using only an optical microscope and nDSE.
- DDMA accuracy: In experiments, DDMA achieved 81 nm accuracy over a 23,300 nm offset and 21 nm accuracy over an 8500 nm offset.
- Potential for sub-10 nm alignment: The paper expresses confidence in achieving overlay alignment better than 10 nm with their developed frameworks.
Limitations and Future Work
- Pseudo-DSE Accuracy: The accuracy of DDMA (based on pseudo-displacement) is acknowledged to be dependent on how closely the features resemble each other and the control of systematic errors. The paper categorizes pseudo-DSE into three levels of accuracy, with the lowest (Pseudo-DSE 3) achieving accuracies in the tens of nanometers.
- IDMA Practicality: While IDMA is presented as the most accurate method, it's noted that it might be "a bit impractical" without features providing approximated absolute alignment, potentially requiring human operator intervention.
- Hybrid Methods: The authors suggest that practical solutions will likely involve hybrid methods combining elements of both DDMA and IDMA, or using traditional geometric alignment marks for an initial best-guess before relying on IDMA.
- Further Experiments: The paper mentions planned experiments using nanoimprint molds to explore IDMA concepts and achieve overlay alignment better than 10 nm, with further details to be provided upon completion. They are also investigating applications in printed electronics and focused ion beam milling.
Summary updated