In this article, we will discuss Genetic Algorithm and how it can be used for Film Bulk Acoustic Resonator (FBAR) design optimization.
FBAR filters for 5G
An FBAR is a Microelectromechanical (MEMs) based device used as RF filters in 5G applications. FBAR structures typically consist of thin piezoelectric film sandwiched between two electrodes mounted over an air cavity on a silicon substrate.
FBARs are suited more to 5G applications than other typical BAW filters. Such as SAWs as they have better out of band rejection. It is well understood that resonators with non-parallel electrode sides support weaker lateral resonances, and both quadrilateral and pentagonal designs are often explored.
Using Simulation for FBAR Design
The structure of a pentagonal resonator has a large effect on filter performance, so the shape and size are optimized for the performance requirements. However, pentagons are five sided shapes so there are a vast number of possible designs. Since FBARs are intricate devices and very costly to fabricate, simulation is a useful tool to save money on fabrication costs and reduce time to market.
Optimization using the Genetic Algorithm
Genetic Algorithm is a method used to optimize constrained or unconstrained problems and is modeled after biological evolution. GA mimics evolutionary processes to evolve a design towards an optimum. There are four main stages:
- Evaluation – analyse fitness score of the individuals in the population
- Selection – select the best individuals to reproduce
- Mutation – reproduction method where the new individual takes attributes from one parent
- Crossover – reproduction method where the new individual takes attributes from both parents
This algorithm is particularly useful for multivariate optimization problems such as designing a pentagonal FBAR.
How to perform this GA Optimization with OnScale?
OnScale can be controlled through its Python or MATLAB API, called the OnScale CLI. OnScale Command Line Interface is a way to submit, download and post process FE simulations through the command line.
This allows users to use OnScale’s HPC Platform to run 1000’s of parallel simulations from their preferred program (Python, MATLAB, etc…) and making batch processing of these simulations a trivial task. When combining OnScale’s ability to evaluate many designs in parallel with GA it unlocks the ability to rapidly solve complex design problems.
FBAR Design Optimization Case Study
Matlab’s global optimization toolbox is ideal for this problem. For a case study, OnScale created a series of pentagonal electrodes using 7 input variables (5 electrode side lengths and the width and length of the substrate).
An initial population of 70 was created by randomly generating the input variables.
OnScale simulated each of the 70 designs and Matlab’s global optimization toolbox performed GA, iterating through populations until the best designs were found. The problem ran for 52 generations until a solution was found, meaning 3,640 designs were investigated. The results made intuitive sense. The best designs had edges angled from the substrate edges and each other (reflection resources). The worst designs had edges aligned with the substrate and parallel to one another.
Figure 6: Best & worst designs from GA optimization
The optimized designs showed a significant reduction in ripple when compared to a square device.
How can you try it?
Genetic Algorithm paired with the OnScale allows engineers to accelerate the design process for complex high-performance devices such as FBARs.
If you need some help setting up or performing your own optimization studies, please contact us!
 R. Ruby, “A decade of FBAR success and what is needed for another successful decade,” 2011 Symposium on Piezoelectricity, Acoustic Waves and Device Applications (SPAWDA), Shenzhen, 2011, pp. 365-369.