A trailblazing innovation in the fabrication of piezoelectric micromachined ultrasonic transducers (PMUTs) has arrived with the advent of MultiphysicsAI. This groundbreaking platform transforms how we practice design. By integrating cloud-based finite element method (FEM) simulation with neural surrogates, it eliminates trial-and-error iteration and substitutes it with an intelligent systematic process of inverse optimization. By building a better understanding, engineers can create positive impacts on the PMUT development efficiency and accuracy, especially for biomedical implementations.
What sets MultiphysicsAI apart from other machine-learning models is its unconventional training regimen, which used more than 10,000 different randomized geometries. This unique and expansive dataset allows the platform to create high quality AI surrogates with a mean error—the average deviation from real-world data—of only 1%. This technology reduces the time needed to iterate on a design exponentially. It further allows sub-millisecond inference on key performance indicators including transmit sensitivity, center frequency, fractional bandwidth, and electrical impedance.
Enhanced Design Efficiency
Perhaps the most astonishing advantage of multimodeMultiphysicsAI is the significant reduction of design time. What used to take days or even weeks can now be done in minutes! Using Pareto front optimization, designers are able to make significant advancements across their most important performance metrics. For example, fractional bandwidth may be extended from 65% to 100% and sensitivity improved by 2-3 dB. The platform achieves a very precise, stable center frequency of 12 MHz, within tight tolerance of ±0.2%. That consistency translates into high performance on a wide range of applications.
By automating the AlphaFold process with a systematic approach, engineers can explore more complex design trade-offs between sensitivity and bandwidth. Such capability is essential for many biomedical applications, where high precision measurements and reliable performance is of highest priority. The ability to leverage standard cloud infrastructure only continues to drive accessibility and collaboration, allowing teams to work on and support new games effortlessly from anywhere.
Practical AI-Accelerated Workflow
MultiphysicsAI offers a one-stop-shop workflow that embeds state-of-the-art AI innovations directly into the process of optimizing PMUTs. The collaborative nature of the platform allows for nearly immediate validation of performance improvements. This capability enables engineers to optimize decisions proactively in real time, leading to a faster time of project delivery and reduced project costs. This efficiency is especially valuable in sectors where time-to-market is of the essence.
The technology rapidly and continuously gets tested, validated performance improvements back downrange within minutes. This gives engineers the ability to quickly iterate on designs without the long turnaround times traditional methods often take. Consequently, organizations are able to react faster to market needs and technology innovations.
