TL;DR
This paper introduces a novel "object-based" approach to designing phase-shift masks (PSMs) for semiconductor manufacturing. By treating mask features as discrete objects with controllable properties (position, size, phase) rather than individual pixels, the authors achieve significantly simpler mask designs that are easier to fabricate while maintaining or improving performance compared to previous pixel-based methods.
What Problem Does This Paper Address?
In semiconductor manufacturing, photolithography is used to pattern intricate circuits onto silicon wafers. Phase-shift masks (PSMs) are advanced masks that manipulate the phase of light passing through them to improve resolution and depth of focus, enabling the creation of smaller features. However, designing effective PSMs is complex. Previous systematic design methods, while powerful, often resulted in masks with very high complexity – many small, scattered features that were difficult and expensive to manufacture reliably. This paper addresses the need for PSM designs that are both high-performing and practical for high-volume manufacturing by reducing mask complexity.
Methods
The core innovation is an "object-based" mask design approach. Instead of optimizing a grid of individual pixels, the mask is represented as a collection of geometric "objects" (primarily rectangles). The optimization process then focuses on determining the optimal:
- Location: Where each object is placed on the mask.
- Size: The dimensions of each object.
- Phase: The phase shift imparted by the object (typically 0° or 180°).
Crucially, the authors introduce a constraint on the complexity of the mask, defined by the maximum number of objects and a minimum feature size. This constraint directly leads to simpler, larger features.
To efficiently search the design space under these constraints, they propose a "bacteria algorithm." This algorithm is inspired by a bacterium searching for food: it iteratively moves in directions that improve a performance metric (e.g., desired intensity pattern on the wafer) and avoids directions that worsen it. This is a local search algorithm designed to find a local minimum quickly, with modifications to improve the quality of the found minimum and avoid "infeasible directions" (situations where objects get stuck due to constraints).
Datasets
Not discussed in the paper. The paper focuses on the design methodology and simulation/experimental results for specific lithographic targets (e.g., 0.4 µm contact holes, bright lines, chromeless patterns) rather than using pre-defined datasets for training.
Key Insights and Ideas
- Complexity Control is Key: By explicitly limiting the number of features and their minimum size, the authors demonstrate that mask complexity can be directly controlled, leading to more manufacturable designs.
- Object-Based Design Simplifies Fabrication: Larger, discrete features are inherently easier to etch and inspect than numerous small, scattered pixels.
- Bacteria Algorithm for Efficient Search: The bacteria algorithm provides a faster alternative to Simulated Annealing for finding good PSM designs, especially when dealing with large solution spaces.
- Performance Trade-offs: While simpler, the object-based designs can achieve comparable or even superior performance (e.g., extended depth of focus, enhanced contrast) to more complex pixel-based designs.
- Symmetry is Not Always Optimal: Forcing symmetry in the optimization process can sometimes hinder finding the best solution, as demonstrated by the comparison of asymmetrical and quasi-symmetrical masks.
Key Results
The paper presents several simulation and experimental results showcasing the effectiveness of their approach:
- Contact Hole Designs: Optimized masks for 0.4 µm and 0.6 µm contact holes showed significantly extended depth of focus (up to 200% compared to biased binary masks) and enhanced contrast (around 35% over biased binary masks).
- Bright Line/Space Designs: An optimized mask for a 0.4 µm bright line achieved the correct line width and an extended depth of focus without sacrificing contrast, outperforming conventional binary masks.
- Chromeless Masks: An optimized chromeless mask for 0.4 µm lines and spaces successfully biased the 0° and 180° phases to achieve equal lines and spaces, a task difficult for initial guesses.
- Comparison with Prior Work: The object-based designs were shown to be comparable in performance to pixel-based optimized masks but with larger, more easily fabricated features.
- Experimental Verification: SEM images of contact resist post images demonstrated the improved performance (larger light intensity and depth of focus) of the new phase shift mask designs compared to conventional binary masks.
Limitations and Future Work
- Local Minima: The bacteria algorithm, while efficient, is prone to finding local minima rather than the absolute global optimum. The authors acknowledge this and suggest combining it with other algorithms or developing more sophisticated bacteria algorithms.
- Data Complexity: While the masks are simpler to fabricate, the object-based representation might contain more data than a conventional mask design, though results suggest this doesn't significantly increase write time.
- Future Research: The authors suggest further research into developing more sophisticated "bacteria" algorithms that can converge to near-global minima and exploring more complex mask patterns. They also highlight the potential for studying more complex object shapes beyond rectangles.
Summary updated