Embedded Passive Component Fabrication via Neural Network CAD Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for manufacturing embedded passive components in printed circuit boards face challenges with material and process variations, leading to costly iterative trimming processes and limited use in high-frequency applications due to increased variability and cost.
Innovation Solution
A refined manufacturing process that includes calibration and process control steps to adjust CAD geometry, utilizing a neural network model to adapt to material and etching variations, eliminating the need for costly laser trimming by modeling and accounting for variability across the circuit board surface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional manufacturing process with standard quality control limits is used, then production cost and time are reduced, but component precision falls outside specified control limits due to material and process variations
Solution Approach 1:
The patent applies preliminary action by performing calibration measurements and creating a neural network model before actual production. The system pre-characterizes material properties and process variations, storing this data in a database to guide subsequent manufacturing decisions, thereby avoiding the need for iterative trimming during production
Solution Approach 2:
The patent implements feedback by using calibration measurements from test structures to train a neural network model that predicts optimal geometry adjustments. This feedback loop allows the system to adapt to material and process variations, ensuring components meet specifications without requiring post-manufacturing trimming
2Manufacturing precision
If iterative laser trimming process is used to adjust component geometry, then component precision is improved, but production time and cost increase significantly
Solution Approach 1:
The system performs preliminary calibration measurements and neural network training before production, establishing a predictive model that eliminates the need for iterative trimming. The optimal geometry is calculated in advance based on calibration data, allowing direct manufacturing without time-consuming adjustments
Solution Approach 2:
The patent replaces the mechanical laser trimming process with a computational approach using neural networks and CAD geometry adjustments. Instead of physically trimming components to achieve precision, the system uses software-based geometry modification guided by calibration data, dramatically reducing production time
3Manufacturing precision
If large embedded components are etched in handling areas, then sheet resistance variations are compensated, but board area available for circuit implementation is reduced
Solution Approach 1:
The patent extracts the quality control function from the circuit board's functional areas by using calibration test structures in handling areas. These test structures are separate from the actual circuit components, allowing sheet resistance characterization without consuming valuable board area needed for implementation
Solution Approach 2:
The calibration test structures serve as intermediaries between the manufacturing process and the final circuit components. By measuring sheet resistance in these test structures and using the data to adjust CAD geometry, the system achieves precision control without requiring large test components in the circuit area
4Manufacturing precision
If global CAD geometry changes are applied to offset etching variations, then average component precision is improved, but local variations across the board are not addressed
Solution Approach 1:
The patent applies local quality by using a neural network model that can provide different geometry adjustments for different locations on the board. The calibration measurements taken at various positions enable the system to account for local variations in material properties and etching processes, rather than applying a single global correction
Solution Approach 2:
The system adds a spatial dimension to the correction approach by using calibration data from multiple locations across the board. The neural network model incorporates position information to generate location-specific geometry adjustments, transforming a one-dimensional global correction into a multi-dimensional local adaptation strategy
Data Source
AI summary
A fabricating process for a multi-layer printed circuit board containing embedded passive components is provided. The method includes a calibration step wherein a calibration measurement is taken of the geometry or at least one electrical parameter of an arrangement of calibration test points for a circuit forming process, such as masking, etching and/or lamination. A process control step is performed during the process, wherein a process control measurement is taken of at least one electrical parameter at one or more process control test points along one or more axes outside areas in which a circuit is to be formed. An analysis is performed of at least the calibration measurement and the process control measurement to calculate a CAD geometry change required to improve precision of embedded passive components to be printed on the multi-layer printed circuit board. The CAD geometry is modified in accordance with the calculated CAD geometry change, and multi-layer printed circuit boards containing embedded passive components are manufactured in accordance with the modified CAD geometry. The analyzing step may model variability and adapt to it.