In this context, MultiphysicsAI has proven to be a revolutionary tool for the design optimization of piezoelectric micromachined ultrasonic transducers (PMUTs) for biomedical applications. Together, cloud-based FEM simulation and neural surrogates promise to disrupt traditional design processes. This potent combination takes us from classic formative testing and trial-and-error approaches to empirical inverse optimization. This breakthrough gives experts the ability to make complex design trade-offs between sensitivity and bandwidth. They have great success stories—incredible leaps in historically low performance metrics.
This innovative method cuts down the time it takes to run a design iteration by leaps and bounds. Where traditional approaches would require days to produce a result, MultiphysicsAI delivers validated performance improvements in under minutes. Built on top of standard cloud infrastructure, it takes all the heavy-duty manual engineering work and turns it into an automated, data-driven process.
A Case Study in Optimization
Additionally, a recent case study showcased the efficiency of MultiphysicsAI by optimizing four geometric parameters over 10,000 coupled piezoelectric-structural-acoustic simulations. The results demonstrated the capacity of this tool to greatly improve PMUT design productivity. The data agnostic AI surrogates, trained only on random geometries, already made a remarkable 1% mean error. These models provide accurate projections for important key performance indicators (KPIs).
Other important KPIs include transmit sensitivity, center frequency, fractional bandwidth, and electrical impedance. With the help of MultiphysicsAI, researchers were able to perform Pareto front optimization that raised fractional bandwidth from 65% to an outstanding 100%. The sensitivity enhancements made for some good wins, up to 2-3 dB gain. They held a center frequency of 12 MHz with a very strict tolerance of ±0.2%.
Streamlined Engineering Overhead
By incorporating MultiphysicsAI, design cycles are more efficient and engineering overhead is significantly reduced. The platform reduces days worth of manual iteration down to just seconds of transparent exploration, all within the bounds of typical computational resources. It frees engineers to spend their expertise in the field interpreting results instead of spending efforts in getting lost in long, manual simulation workflows.
Plus, thanks to sub-millisecond inference across all critical performance indicators, engineers can efficiently evaluate design options and discover the best solution in record time. Being able to easily navigate intricate design trade-offs between sensitivity and bandwidth lets experts customize PMUT designs even more, making them more effective.