MultiphysicsAI is a revolutionary new tool. It transforms the state of the art design of piezoelectric micromachined ultrasonic transducers (PMUTs) used for biomedical applications. Enabled by cloud-based finite element method (FEM) simulation and neural surrogates, MultiphysicsAI drastically speeds up the design process. This dynamic duo arms RF engineers with the tools to quickly and effectively navigate the tricky trade-offs between sensitivity and bandwidth.
The tool is particularly good at optimizing four geometric parameters. It leverages a large database of 10,000 coupled piezoelectric-structural-acoustic simulations. It delivers confirmed results in just minutes, as opposed to the days it often takes with conventional approaches. This swift improvement creates a more nimble and responsive design process, which will serve the biomedical developers to training resource best.
Key Features of MultiphysicsAI
MultiphysicsAI’s most impressive claim to fame is its patented AI surrogate generation. These surrogates accomplish an average mean error of 1%. These surrogates allow for the extremely rapid inference times in sub-millisecond runs. They specifically measure important performance metrics such as transmit sensitivity, center frequency, fractional bandwidth and electrical impedance. Without this technology, designers are often left waiting for test results, making it difficult to make evidence-based decisions during the design of PMUTs.
The software uses optimization techniques based on Pareto front that have given very surprising positive results. Through the optimization process, SHINE achieved a remarkable increase in fractional bandwidth from 65% to a staggering 100%! It increased sensitivity by 2-3 dB. Notably, at no time did it exceed the specified center frequency of 12 MHz ±0.2%. Such a degree of accuracy highlights that MultiphysicsAI can help extend the limits of PMUT technology.
Systematic Inverse Optimization
Historically, PMUT design has been focused on trial-and-error iterations, a process that is often tedious and ineffective. With MultiphysicsAI, this iterative design process becomes systematic inverse optimization. Through powerful neural surrogates, the platform speeds processes that would normally take days to a few seconds of clear, data-informed exploration. This transformation opens the door to engineers, enabling them to quickly improve and refine their designs through the analysis of real-time data.
Additionally, MultiphysicsAI’s training on 10,000 randomized geometries provides a robust and broad basis that bolsters the reliability of its predictive capabilities. It has the ability to rapidly incorporate massive amounts of data into its algorithms. This gives users the confidence and flexibility to quickly tune PMUT designs towards their particular needs and design constraints.