Skip to content
Close
Talk to an expert
Japanese
Talk to an expert
Japanese

PMUT sensor generative design

Generative design of PMUT sensors via MultiphysicsAI & surrogate modeling

 

Electrostatics   •   Solid mechanics   •   Acoustic waves  •   Acoustic-Structure Interaction

The challenge

Designing PMUTs involves a difficult trade-off: improving Sensitivity usually degrades Bandwidth.

Finding the optimal balance (the Pareto Front) is an "Inverse Design" problem that traditionally requires weeks of manual trial-and-error. Engineers often settle for "good enough" because they cannot simulate enough variations to find the "best possible."

Approach with Quanscient Allsolve

Quanscient employed a "MultiphysicsAI" workflow:

  1. Generated a dataset of 10,000 parallel simulations in the cloud (varying 4 geometric parameters).
  2. Trained a Deep Neural Network (Surrogate Model) on this data.
  3. Used the AI to predict the performance of millions of design combinations in sub-milliseconds to identify the optimal geometry.

Key results

  • Instant optimization: The AI Surrogate Model predicts complex physics outcomes in <1 millisecond with ~1% error, allowing the team to calculate the optimal Pareto Front in seconds.
  • Performance breakthrough: Discovered a novel design that increased Fractional Bandwidth from 65% to ~100% while maintaining the target 12 MHz frequency.
  • Workflow: Transformed the engineering process from "Guess and Check" to systematic, data-driven Inverse Design.
Design space with Pareto front and validation simulations constrained to 12 MHz (left). Initial design vs final design (right, blue initial, green final).