Expert System Optimizing Spiral Inductor Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Designing and optimizing spiral inductors for integrated circuits is a costly and inefficient process due to the need for repeated simulations and customization for each silicon process node, with existing methodologies lacking re-usability and accuracy across different manufacturing processes.
Innovation Solution
An expert system-based methodology for synthesizing and optimizing spiral inductors that uses a rule set to adapt to various process parameters, allowing for rapid design and optimization independent of manufacturing processes, with self-learning capabilities and sparse simulation for fast and accurate results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If geometric programming methodology is used with lumped-element inductor model, then optimization can be solved globally, but prior knowledge of mathematical expressions is required which is impractical for modern silicon processes
Solution Approach 1:
The patent creates a trained equivalent lumped-element inductor model that copies the electromagnetic behavior of the actual inductor structure. This trained model serves as a simplified representation that captures the complex relationships between geometric parameters and performance metrics without requiring explicit mathematical expressions, thus resolving the contradiction between accuracy and complexity.
Solution Approach 2:
The patent transforms the design approach by changing from requiring explicit mathematical expressions to using a trained model with adjustable parameters. The model is trained on simulation or measurement data, allowing it to adapt to different silicon processes through parameter adjustment rather than requiring complete reformulation of mathematical relationships.
2Productivity
If trained equivalent lumped-element inductor models are used, then synthesis and optimization can be performed, but the model must be separately trained for each manufacturing process leading to increased setup times
Solution Approach 1:
The patent develops a universal trained model framework that can be applied across multiple silicon processes. Once the model structure is established and trained on representative data, it can be reused for different process nodes with minimal additional training, enabling the same methodology to serve multiple manufacturing processes rather than requiring separate models for each.
Solution Approach 2:
The patent performs preliminary training of the lumped-element model using simulation or measurement data from full-wave or quasi-static EM solvers. This preliminary action creates a pre-trained model that captures the essential relationships, allowing rapid synthesis and optimization without requiring extensive re-training for each new design task or process variation.
3Productivity
If physical lumped-element inductor models are used with database lookup, then synthesis is fast, but the method is not re-usable between processes and requires extensive database population
Solution Approach 1:
The patent implements a dynamic model training approach where the lumped-element model parameters are adapted to different silicon processes based on available process data. Rather than using a static database that must be completely populated for each process, the model dynamically adjusts its parameters through training on process-specific measurements or simulations, enabling fast synthesis while maintaining process adaptability.
Data Source
AI summary
A system and method for designing an electrical component comprises a model extraction engine configured to generate a model based on a set of parameters, a simulator configured to simulate the generated model and measure performance, a rule-set usable to determine changes to the set of parameters, and an inference engine configured to change salience values of expert rules included in the rule set. The salience value determines when and if an expert rule is used to change the set of parameters. One or more microprocessors are configured to determine design characteristics of the electrical component by iteratively performing, until measured performance is within tolerance, the steps of generating a model based on an updated version of the set of parameters, simulating the generated model, measuring performance of the generated model, and updating the set of parameters using the rule-set if the measured performance is not within the predefined tolerance.