A recurrent neural network-enabled 3D dynamic focusing laser for high-fidelity microstructures toward ultrasensitive and linear pressure sensing

Abstract The performance of flexible pressure sensors critically depends on the morphology of their microstructures, which governs both sensitivity and linearity. However, owing to the reduced two-dimensional (2D) Z-axis accuracy caused by focal depth mismatch, the rapid and precise laser manufacturing of complex three-dimensional (3D) functional surface microstructures remains a great challenge, limiting sensor performance. Here, we propose a 3D dynamic focusing laser (3D-DFL) fabrication approach driven by a recurrent neural network (RNN), which can directly predict laser processing parameters and adaptively tune laser multi-parameter, overcoming the defocusing limitations of 2D laser and enabling high-precision, customized fabrication of complex 3D surface microstructures. The optimized sensor, featuring a sparse pyramid array, achieves a sensitivity of 104 kPa ⁻¹ over a linear range up to 400 kPa ( R ² = 0.996), with a linear sensing factor (LSF) over 100 times higher than square pyramid structure design. Leveraging its ultra-high sensitivity, this sensor can be applied to robots for damage-free adaptive grasping of slippery jelly. Furthermore, we develop a 4 × 4 sensor array and a crosstalk-free acquisition system (measurement error <0.3%), demonstrating its utility in human–machine interface applications. This laser-based fabrication approach for pressure sensors can be extended to other sensor types or materials to achieve outstanding performance, with considerable potential in robotic intelligent grasping, wearable systems, and human–machine interaction devices.