Multimodal Deep Learning for Consistent Semiconductor Mask Designs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for predicting mask designs in semiconductor manufacturing produce inconsistent and ambiguous results due to varying training data, leading to instability and difficulty in controlling the semiconductor manufacturing process.
Innovation Solution
A deep learning model that learns a continuous multimodal distribution of mask features, using a U-net type architecture with conditional mapping, to select variants that result in valid wafer imaging, reducing ambiguity and enhancing defect prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning models are used to predict mask designs, then the manufacturing process can proceed with standard tools, but the prediction results are inconsistent and ambiguous
Solution Approach 1:
The patent transforms the static mask design prediction into a dynamic process by introducing a continuous latent space where the model can explore multiple possible mask designs. The dynamic relaxation algorithm allows the system to iteratively adjust and converge to optimal solutions, converting a single-point prediction into a dynamic optimization process that resolves ambiguity through continuous refinement.
Solution Approach 2:
The patent changes the parameter representation by introducing a continuous latent space with multiple dimensions instead of direct discrete mask design parameters. This continuous parameterization allows for smooth variations and multiple valid solutions, transforming the rigid binary prediction into a flexible continuous optimization problem that can capture the inherent variability in mask design.
2Manufacturing precision
If sophisticated fine-tuning steps are applied to overcome resolution limits, then smaller features can be printed, but the process complexity increases
Solution Approach 1:
The patent applies preliminary action by performing optical proximity correction and resolution enhancement directly in the mask design generation phase rather than as separate post-processing steps. The deep learning model learns to incorporate these corrections during the fundamental mask design creation, eliminating the need for subsequent fine-tuning operations and reducing overall process complexity.
Solution Approach 2:
The patent merges multiple previously separate lithography optimization steps into a single unified deep learning model. The model simultaneously performs mask design generation, optical proximity correction, and resolution enhancement in one integrated process, combining what were traditionally separate complex steps into a single coherent system.
3Adaptability or versatility
If multiple fabrication processes are used to form various features and layers, then device functionality is achieved, but the manufacturing time and complexity increase
Solution Approach 1:
The patent replaces traditional mechanical and chemical fabrication processes with a computational approach. Instead of physically trial-and-erroring through multiple fabrication processes, the deep learning model computationally predicts the optimal mask design that will achieve the desired device features, substituting physical experimentation with intelligent simulation and prediction.
Data Source
AI summary
A method of determining a mask design, the method includes generating a continuous multimodal representation of a probability distribution of a target design in at least a portion of a latent space. The latent space includes a distribution of feature variants that can be used to generate mask designs based on the target design. The method includes selecting a variant from the continuous multimodal representation in the latent space. The variant includes a latent space representation of one or more features to be used to determine the mask design. The method includes determining the mask design based on the target design and the variant.