Computational Intelligence and Wavelet Transform Based Metamodel for Efficient Generation of Not-Yet Simulated...

Area of Science:

• Electrical Engineering
• Computational Science

Background:

• Circuit-level simulations for complex electronic systems, particularly analog and mixed-signal ones, are time-intensive.
• Efficient analysis across diverse operating conditions is crucial for system design and verification.

Purpose of the Study:

• To develop a data-driven method for creating fast and accurate metamodels for electronic systems.
• To enable the generation of system waveforms for various parameter combinations, reducing simulation time.

Main Methods:

• Utilized a wavelet transform for characterizing system waveforms.
• Employed genetic algorithm optimization to select optimal wavelet transforms and relevant decomposition coefficients.
• Applied artificial neural networks to derive coefficients for new parameter combinations.

Main Results:

• Achieved high accuracy with a maximum mean squared error of 7.1x10^-5 for normalized waveforms.
• Demonstrated computational efficiency with metamodel build-up taking a maximum of 18 minutes.
• Enabled waveform generation in under 1 second for any parameter combination.

Conclusions:

• The developed metamodels are reliable, accurate, and efficient for analyzing electronic systems.
• These metamodels facilitate comprehensive design space exploration, including corner case analysis and sensitivity studies.
• The approach significantly reduces the computational effort required for extensive system analysis.