Surrogate Modeling For High-frequency Design: Recent Advances

Contents
Preface
About the Editors
List of Contributors
Acknowledgments
1. Fundamentals of Data-Driven Surrogate Modeling
1.1 Data-Driven Surrogates: Overview
1.2 Design of Experiments
1.2.1 Factorial designs
1.2.2 Space-filling designs
1.2.3 Sequential sampling
1.3 Modeling Methods
1.3.1 Polynomial regression
1.3.2 Radial basis functions
1.3.3 Kriging
1.3.4 Polynomial chaos expansion
1.3.5 Support vector regression
1.3.6 Other methods
1.4 Surrogate Model Validation
Acknowledgments
References
2. Fundamentals of Physics-Based Surrogate Modeling
2.1 Physics-Based Surrogates: Overview
2.2 Low-Fidelity Models
2.2.1 Overview
2.2.2 Variable-resolution and variable-accuracy models
2.2.3 Variable-fidelity physics models
2.2.4 Selecting low-fidelity models
2.3 Physics-Based Surrogates — Basic Techniques
2.4 Response Correction Techniques
2.4.1 Quasi-global modeling using multi-point space mapping
2.4.2 Space mapping with a function approximation layer
2.4.3 Multi-point output space mapping
2.4.4 Generalized shape-preserving response prediction
2.4.5 Feature-based modeling
2.5 Physics-Based Surrogates for Design Optimization
Acknowledgments
References
3. Parametric Modeling of Microwave Components Using Combined Neural Network and Transfer Function
3.1 Introduction
3.2 Parametric Modeling Using Neuro-Transfer Function in Rational Format
3.2.1 Formulation of the neuro-TF model in rational format
3.2.2 Two-stage training process of the neuro-TF model
3.3 Parametric Modeling Using Neuro-Transfer Function in Pole/Zero Format
3.3.1 Formulation of the neuro-TF model in pole/zero format
3.3.2 Pole/zero-matching algorithm for addressing the issue of mismatch of poles and zeros
3.4 Parametric Modeling Using Neuro-TF Model in Pole/Residue Format
3.4.1 Formulation of the neuro-TF model in pole/residue format
3.4.2 Vector fitting for parameter extraction
3.4.3 Pole–residue tracking technique for order-changing
3.5 Sensitivity-Analysis-Based Neuro-TF Modeling Technique
3.5.1 Structure of the sensitivity-analysis-based neuro-TF model
3.5.2 Preliminary training process for the sensitivity-analysis-based neuro-TF model
3.5.3 Refinement training process for the sensitivity-analysis-based neuro-TF model
3.6 Neuro-TF Modeling Examples and RF/Microwave Applications
3.6.1 Neuro-TF modeling in pole/zero format of a three-pole H-plane filter
3.6.2 Neuro-TF modeling in pole/residue format of a microwave junction
3.6.3 Sensitivity-analysis-based neuro-TF modeling of a diplexer
3.7 Discussion
3.8 Conclusion
References
4. Surrogate Model-Assisted Global Optimization for Antenna Design
4.1 Introduction
4.2 Overview of the SADEA Algorithm Family
4.3 The PSADEA Method
4.3.1 GP surrogate modeling
4.3.2 The DE algorithm
4.3.3 Implementation of PSADEA
4.4 Case Study
4.4.1 Example one
4.4.2 Example two
4.5 Conclusions
References
5. Surrogate-Based Modeling and Design Optimization Techniques for Signal Integrity in High-Performance Computer Platforms
5.1 Introduction
5.2 Post-Silicon Validation
5.2.1 Post-silicon tuning
5.2.2 System margining
5.2.3 Jitter tolerance testing
5.3 PHY Surrogate Modeling
5.3.1 Design of experiments
5.3.1.1 Box Behnken
5.3.1.2 Orthogonal arrays
5.3.1.3 Sobol sequence
5.3.2 Polynomial-based surrogate modeling
5.3.3 Generalized regression neural networks
5.3.4 Supported vector machines
5.3.5 Kriging
5.3.6 Artificial neural networks
5.3.6.1 ANN topology
5.3.6.2 ANN modeling and training
5.3.7 PHY surrogate modeling results and comparisons
5.3.7.1 SATA Gen3 HSIO link PHY modeling
5.3.7.2 USB3.1 Gen1 HSIO link PHY modeling
5.4 Receiver Equalization Surrogate-Based Optimization
5.4.1 Objective function for system margining
5.4.2 Surrogate-based optimization for system margining
5.4.3 Objective function for system margining and jitter tolerance
5.4.4 Surrogate-based optimization for system margining and jitter tolerance
5.4.4.1 Test case 1: USB3
5.4.4.2 Test case 2: SATA3
5.4.4.3 Test case 3: PCIe
5.5 Space Mapping Optimization for PHY Tuning
5.5.1 Broyden-based input space mapping
5.5.2 Fine model
5.5.3 Coarse model
5.5.4 ASM optimization
5.5.5 Optimization results
5.6 Discussion and Conclusion
References
6. Performance-Driven Inverse/Forward Modeling of Antennas in Variable-Thickness Domains
6.1 Introduction
6.2 Antenna Modeling Using Nested Kriging
6.2.1 Design space objective space: First-level (inverse) surrogate
6.2.2 Surrogate model domain: Second-level surrogate
6.3 Variable-Thickness Domain
6.3.1 Domain thickness: Model accuracy vs. utility trade-offs
6.3.2 Variable-thickness domain: Definition and properties
6.3.3 Nested kriging with variable-thickness domain
6.4 Demonstration Case Studies
6.4.1 Case I: Dual-band microstrip dipole antenna
6.4.2 Case II: Broadband patch antenna
6.5 Summary and Discussion
Acknowledgments
References
7. Sampling Methods for Surrogate Modeling and Optimization
7.1 Introduction
7.2 Conventional Sampling Methods
7.2.1 Full factorial sampling method
7.2.2 Monte Carlo sampling method
7.2.3 Latin hypercube sampling method
7.2.4 Space-infill sampling method
7.3 A Hybrid Sampling Method for Surrogate Modeling and Optimization
7.3.1 Local sampling
7.3.2 Global sampling
7.3.3 Microwave verification example
7.3.3.1 Modeling performance
7.3.3.2 Optimization performance
7.4 Adaptive Sampling Region Updating for Surrogate-Assisted Optimization
7.4.1 Adaptive sampling region updating strategy
7.4.2 Surrogate refinement using fine model
7.4.3 Antenna verification examples
7.5 Conclusions
Acknowledgments
References
8. Statistical Design Centering of Microwave Systems via Space Mapping Technology and Modified Trust Region Algorithm
8.1 Introduction
8.1.1 Design centering problem
8.1.2 Microwave design centering
8.2 New Statistical Design Centering Technique for Microwave Systems
8.2.1 Modified trust region algorithm
8.2.1.1 Steps for the modified TR algorithm
8.2.2 Generalized space mapping technique
8.2.3 The statistical design centering algorithm
8.3 Practical Examples
8.3.1 Bandstop microstrip filter with open stubs
8.3.2 Ultra-wideband multiple-input–multiple-output antenna
8.4 Conclusion
Acknowledgments
References
9. Expedited Yield-Driven Design of High-Frequency Structures by Kriging Surrogates in Confined Domains
9.1 Introduction
9.2 Yield Optimization Problem and Benchmark Algorithms
9.2.1 Yield optimization problem
9.2.2 Surrogate-based yield optimization — benchmark Algorithm 1: One-shot optimization
9.2.3 Surrogate-based yield optimization — benchmark Algorithm 2: Sequential approximate optimization
9.3 Surrogate-Based Yield Optimization with Domain Confinement
9.3.1 Yield optimization of multi-band antennas
9.3.2 Yield optimization of microwave couplers
9.4 Demonstration Case Studies
9.4.1 Case I: Ring-slot antenna
9.4.2 Case II: Dual-band uniplanar dipole antenna
9.4.3 Case III: Triple-band uniplanar dipole antenna
9.4.4 Case IV: Compact microstrip rat-race coupler
9.5 Summary and Discussion
Acknowledgments
References
10. Solving the Inverse Problem Through Optimization — Applications to Analog/RF IC Design
10.1 Introduction
10.2 Overview of Proposed Design Flow
10.2.1 Design space analysis
10.2.2 Surrogate modeling method
10.2.3 Adaptive sample strategy
10.3 Bayesian Optimization Framework
10.3.1 Overview of Bayesian optimization
10.3.1.1 Acquisition function
10.4 Candidate Point Search
10.5 Design Analysis: Weight Setting
10.5.1 Weight setting and optimization
10.6 Example of Optimization and Design Reuse with Bayesian Optimization
10.7 Surrogate Model Extension in Physical Design: Multi-Fidelity Optimization for Electromagnetic Simulation Acceleration
10.7.1 Space mapping
10.7.2 Overview of EM simulation acceleration
10.7.2.1 Overview of multi-fidelity surrogate-based optimization with candidate search
10.7.3 Multi-fidelity surrogate-based optimization with candidate search flow
10.7.3.1 Design exploration
10.7.3.2 Statistical surrogate model
10.7.3.3 Adaptive sampling with dropout
10.7.3.4 Adaptive samples filtering
10.7.3.5 Low-fidelity dataset update
10.7.3.6 Sample generation for model rebuild
10.8 Experimental Results
10.8.1 Inductor design
10.8.1.1 Optimization result comparison for inductor
10.8.1.2 IP redesign example — VCO
10.9 Conclusions
Acknowledgments
References
11. An Automated and Adaptive Calibration of Passive Tuners Using an Advanced Modeling Technique
11.1 Introduction
11.2 Behavior of a Passive Mechanical Tuner
11.3 The Proposed Algorithm
11.3.1 Adaptive sampling technique
11.3.2 Final model generation
11.3.3 Validation and error estimation
11.4 Numerical Results Based on Measurements
11.5 Conclusion
References
12. Surrogate Modeling of High-Frequency Electronic Circuits
12.1 Introduction
12.2 Surrogate Modeling as a Circuit Optimization Tool
12.3 Variability Analysis of an LNA Using Surrogate Modeling
12.3.1 Case study I: Surrogate modeling of LNA performance parameters based on bondwire inductances
12.3.1.1 Surrogate modeling of S21 and S22
12.3.1.2 Surrogate modeling of NF and S11
12.3.2 Case study II: Surrogate model of the LNA gain variability
12.4 Concluding Remarks
References
13. Sensitivity Analysis and Optimal Design with PC-co-kriging
13.1 Introduction
13.2 Methods
13.2.1 Surrogate modeling and analysis workflow
13.2.2 Sampling plan
13.2.3 Constructing the surrogate model
13.2.4 Validation
13.2.5 Surrogate-based sensitivity analysis
13.2.6 Surrogate-based optimal design
13.3 Application Examples
13.3.1 Surrogate modeling of the borehole function
13.3.2 Model-based sensitivity analysis of ultrasonic testing
13.3.3 Optimal design of transonic airfoil shapes
13.4 Conclusion
Acknowledgments
References
Index