TL;DR
As chip manufacturing scales down, creating precise "cuts" in metal wires for unidirectional designs becomes challenging. This paper introduces an optimization method using Directed Self-Assembly (DSA) and Integer Linear Programming (ILP) to efficiently and accurately place these cuts, significantly reducing manufacturing errors and wire extensions.
What Problem Does This Paper Address?
Modern integrated circuits are shrinking, pushing traditional lithography techniques to their limits. For unidirectional layouts, which offer process advantages at these small scales, precise placement of "cuts" in metal wires is crucial for circuit connectivity and manufacturability. Traditional methods struggle with dense, randomly distributed cuts, leading to issues like cuts extending into adjacent wires or extra lines being cut, impacting connectivity. Directed Self-Assembly (DSA) offers a promising solution by enabling the formation of small, controlled patterns. However, optimizing the placement of these DSA-guided cuts to minimize manufacturing conflicts and line-end extensions remains a significant challenge. This paper tackles the problem of optimizing DSA-based cut mask placement for unidirectional designs.
Methods
The authors propose a two-pronged approach:
- Integer Linear Programming (ILP) Formulation: They first formulate the DSA cut mask optimization problem as an ILP. This mathematical model aims to assign each required cut to a specific DSA guiding template while minimizing two key objectives: the number of unresolvable cuts (conflicts) and the total length of wire extensions needed to accommodate the cuts. The ILP formulation considers constraints related to line-end extension limits, template spacing, and the internal alignment requirements of different DSA template types.
- Speed-up Method (Layout Division): Recognizing that ILP can be computationally expensive for large-scale industrial designs, they introduce a speed-up technique. This method involves dividing the overall layout into smaller, manageable pieces. The ILP optimization is then applied independently to each piece. Finally, the solutions from these smaller pieces are merged, and any conflicts arising at the boundaries between pieces are resolved. This decomposition aims to reduce runtime significantly while maintaining solution quality.
Datasets
The authors evaluated their methods using six benchmark unidirectional metal layer layouts provided by the authors of a previous work [9]. These benchmarks vary in size and represent typical industrial designs.
Key Insights and Ideas
- DSA as a Solution for Dense Cuts: The paper highlights DSA's potential to overcome the limitations of traditional lithography for creating dense, precise cuts in scaled technologies.
- ILP for Optimal Assignment: Formulating the problem as an ILP allows for a systematic search for the optimal assignment of cuts to DSA templates, directly targeting the minimization of critical manufacturing metrics.
- The Trade-off Between Runtime and Optimality: The ILP approach guarantees optimality but can be slow. The proposed speed-up method demonstrates that a practical balance can be struck by dividing the problem, achieving near-optimal results with significantly reduced computation time.
- Generality of the Approach: The proposed algorithms are designed to be adaptive to various DSA guiding template shapes, making them valuable as DSA technology evolves.
Key Results
The proposed methods show significant improvements over a prior state-of-the-art approach [9]:
- Reduced Wire Extensions: The ILP formulation achieved an average of 77% fewer wire extensions compared to [9]. The speed-up method also achieved an average of 69% fewer wire extensions.
- Zero Conflicts: Both the ILP and the speed-up method successfully eliminated template conflicts for all test cases, whereas the prior method had an average of 30 remaining conflicts.
- Significant Speed-up: The speed-up method was approximately 2934x faster on average than the pure ILP approach for the tested benchmarks, while still delivering excellent results.
Limitations and Future Work
- Runtime for Very Large Problems: While the speed-up method significantly reduces runtime, the paper acknowledges that for extremely large-scale problems, the ILP formulation (even when divided) can still be computationally intensive.
- Merging Conflicts: The speed-up method requires a legalization step to resolve conflicts that arise at the boundaries of the divided layout pieces. The effectiveness and efficiency of this merging process are crucial.
- Future Work: The authors suggest that their algorithms can be extended to handle more complex DSA guiding patterns, which is a natural progression as DSA technology advances. Further research could focus on even more efficient decomposition and merging strategies for massive layouts.
Summary updated