Multilayer OPC Model for Edge Placement Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical proximity correction (OPC) techniques in semiconductor photolithography are limited by their inability to account for multilayer patterning effects, leading to significant edge placement errors and increased development time and resource consumption due to reliance on rule-based adjustments.
Innovation Solution
A multilayer OPC model that incorporates layout information from neighboring reference layers into a machine learning algorithm to predict and correct contour shifts, reducing residual errors and improving edge placement accuracy through a semi-physical model enhancement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional single-layer OPC techniques are used, then the OPC process is simple and fast, but edge placement errors increase and manufacturing precision deteriorates due to inability to account for multilayer patterning effects
Solution Approach 1:
The patent transitions from single-layer OPC to multilayer OPC by adding the dimension of multiple patterning layers. The system now considers interactions between current layer and reference layers simultaneously, capturing cross-layer optical effects that were previously ignored. This dimensional expansion enables accurate prediction of edge placement errors caused by multilayer patterning effects.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the physical lithography process and the OPC correction. This ML intermediary learns the complex mapping between multilayer pattern configurations and resulting edge placement errors, serving as a bridge that translates physical multilayer interactions into actionable correction data without requiring explicit analytical modeling of all interaction mechanisms.
2Productivity
If rule-based adjustments are used for multilayer effects, then the OPC process remains simple, but development time increases and productivity decreases due to iterative manual corrections
Solution Approach 1:
The patent performs preliminary action by pre-training the machine learning model on comprehensive multilayer pattern data before actual OPC execution. This pre-training phase captures all possible multilayer interaction scenarios in advance, so that during production OPC, the system can directly query pre-computed correction values without iterative manual adjustments. This shifts the time investment to an upfront training phase rather than repeated correction cycles.
Solution Approach 2:
The patent creates a virtual copy of the multilayer patterning process through the machine learning model. Instead of physically iterating through multiple correction passes, the system uses the trained ML model to predict and correct edge placement errors in a single pass by copying the learned relationships from training data to production scenarios, eliminating repetitive manual correction cycles.
3Volume of moving object
If features are designed at or near manufacturing tolerances to achieve compact devices, then device size decreases, but infidelities in manufactured masks occur leading to reduced reliability
Solution Approach 1:
The patent applies preliminary anti-action by using the multilayer OPC model to predict and counteract potential manufacturing infidelities before mask fabrication. The system calculates correction amounts that pre-compensate for expected deviations due to tight tolerances and multilayer interactions, so that the final manufactured features match design specifications despite being designed near manufacturing limits.
Data Source
AI summary
A method for optical proximity correction (OPC) comprises creating a semi-physical model of a mask for a current layer in an IC design layout using physical parameters of a lithography process used to create the mask, the semi-physical model specifying contours of the plurality of features of the mask. It is determined from design information whether the current layer is deformed by the one or more reference layers that overlap the current layer near the contours. Responsive to determining that the current layer is deformed by the one or more reference layers, the semi-physical model and the design information of the one or more reference layers are input into a trained machine learning algorithm to generate a contour shift prediction for the current layer, the contour shift prediction estimating a residual error of the semi-physical model. The contour shift prediction is then used for multilayer OPC correction of the current layer.