Semiconductor Process Policy Guidance Without Re-Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the semiconductor foundry industry, changing a semiconductor product design during simulation can be time-consuming and costly, and may inadvertently reveal manufacturing know-how, as existing methods require re-simulation and data sharing.
Innovation Solution
A method and electronic device that generate process policies for semiconductor manufacturing using TCAD and compact models, trained through machine learning, to reflect semiconductor characteristics, while assigning different authority levels to user information for enhanced security and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If re-simulation is performed based on changed design, then design flexibility is improved, but time consumption and cost increase enormously
Solution Approach 1:
The system performs preliminary actions by pre-generating multiple process policies based on different strategic references (e.g., performance optimization, power consumption minimization, area reduction) using TCAD simulation data and machine learning models. When design changes occur, the system can directly select or combine pre-generated policies without re-simulation, enabling rapid adaptation to design modifications while avoiding time-consuming re-simulation processes
Solution Approach 2:
The system creates simplified copies of complex TCAD simulation results through machine learning models that generate process policies representing different optimization strategies. These policy copies can be quickly generated and modified without performing full TCAD simulations, allowing design flexibility while significantly reducing computation time and resources
2Ease of operation
If simulation data is shared with client, then client requirements are met, but manufacturing know-how security is compromised
Solution Approach 1:
The system segments simulation data and process policies into different categories with varying levels of sensitivity. Process policies are presented as high-level optimization strategies (e.g., 'optimize for performance') rather than detailed manufacturing parameters. This segmentation allows clients to access necessary information for their requirements while protecting core manufacturing know-how from disclosure
Solution Approach 2:
The system acts as an intermediary between TCAD simulation results and client requests. It translates detailed simulation data into abstracted process policies that satisfy client performance requirements without revealing sensitive manufacturing processes. The machine learning models serve as intermediaries that process and filter information, providing clients with actionable insights while maintaining security boundaries
3Adaptability or versatility
If multiple process policies are generated for different strategic references, then design optimization options are improved, but system complexity increases
Solution Approach 1:
The system manages complexity by parameterizing process policies according to different strategic references (performance, power, area, etc.). Each policy is defined by a set of parameters derived from TCAD simulations under specific strategic conditions. This parameterization allows the system to generate multiple optimization options without proportionally increasing system complexity, as policies are created by varying strategic parameters rather than fundamental process structures
Data Source
AI summary
A method of guiding a semiconductor manufacturing process includes receiving semiconductor manufacturing process data corresponding to a target semiconductor product, generating first semiconductor characteristic data corresponding to the semiconductor manufacturing process data by using a technology computer-aided design (TCAD) model trained through machine learning based on training data including TCAD simulation data, generating second semiconductor characteristic data corresponding to the semiconductor manufacturing process data by using a compact model generated based on information of measurement of at least one semiconductor characteristic of a first semiconductor product, generating, based on the first semiconductor characteristic data and the second semiconductor characteristic data, a plurality of process policies respectively corresponding to a plurality of strategic references, by using a plurality of strategy models; and providing a final process policy corresponding to the target semiconductor product based on the plurality of process policies.