Major developments in artificial intelligence (AI) combined with advances in piezoelectric micromachined ultrasonic transducer (PMUT) development have opened the door to designing PMUTs specifically for biomedical applications. Taking the front seat in this arena has been MultiphysicsAI — pioneering the combination of cloud-based finite element method (FEM) simulation integrated with neural surrogates. This integration changes the paradigm from an ad hoc trial-and-error iterative integration to a systematic inverse optimization-driven integration.
MultiphysicsAI enables experts to explore advanced design trade-offs. It’s figuring out how to toe the line between sensitivity and bandwidth that makes this technique so great. By harnessing the power of AI, the tool streamlines the optimization process, enabling significant performance enhancements in mere minutes rather than days.
A Groundbreaking Approach to Optimization
In a recent application, MultiphysicsAI was able to optimize four geometric parameters over 10,000 coupled piezoelectric-structural-acoustic simulations. This pioneering engineering approach results in demonstrated performance benefits with reduced engineering burden. The outdated manual iteration process due to tedious review like RFIs and transmittals can be costly and time-consuming. Today, it’s becoming a rapid, data-rich market research project employing robust computational platforms.
The full system is then trained on these 10,000 random geometries. As a result, it creates AI surrogates that emulate these KPIs with startling precision. These surrogates are able to attain astonishing high accuracies. They demonstrate a mean error of just 1% for transmit sensitivity, center frequency, fractional bandwidth and electrical impedance. The inference time for these AI models is at sub-millisecond levels, making the design process even more efficient.
Enhancements Through Pareto Front Optimization
Perhaps the most impressive feature of MultiphysicsAI is that it uses Pareto front optimization. This approach results in positive improvements in trade—realizing two or more sometimes conflicting performance metrics at once. In tangible terms, it has been demonstrated to boost the nominal fractional bandwidth from 65% up to an astounding 100%. We obtained very significant sensitivity improvements of 2-3 dB. Simultaneously, we kept a center frequency of 12 MHz with a tolerance of ±0.2%.
With Pareto front optimization, engineers can easily find the ideal balance between competing design parameters. This unique feature greatly enhances the PMUTs firepower. Most importantly, it provides transparency about how additional design choices play off one another.
The Future of PMUT Design
The ramifications of MultiphysicsAI go well beyond just the performance improvements seen in the short term. This technology significantly reduces the time required to produce design iterations. Beyond that, it improves the quality of these simulations, enabling a broader range of creative and impactful biomedical solutions. A new generation of specialists are pioneering the use, potential and capabilities of new AI-driven design tools. This exploration is set to transform the ultrasonics technology space.