Spiderweb nanomechanicalresonator with a novel torsionalsoft clampingmotionfound by Bayesian optimization

Nanomechanical resonators are key enablers of next-generation technologies, from ultra-sensitive detectors of fundamental forces to quantum-limited commercial sensors and quantum networks operating at room temperature. Yet, the rational design of nanomechanical resonators is far from trivial. Apart from basic principles derived from one-dimensional analytical models and the broad use of silicon nitride as a highly tensile base material, human intuition remains the driving force behind the design process. Here, inspired by nature and guided by machine learning, a spiderweb-like resonator concept is presented that exhibits a novel vibration mode that reduces radiation losses without using phononic shields. This vibration mode was discovered by the data-driven exploration and was found to be essential for obtaining an unprecedented quality factor (1.8 billion) in a compact design (3 mm characteristic length) at low frequencies (around 130 kHz). This work demonstrates that machine learning and Bayesian optimization can play a key role in uncovering practical new directions in nanotechnology.