Machine-learning assisted scatterometry metrology on nanosheet transistors

This research presents an innovative approach in scatterometry metrology for accurately determining the critical dimensions of GAA nanosheet transistor structures, a key aspect in advancing nanoelectronics. We employed Finite-Difference Time-Domain (FDTD) simulation to construct a comprehensive library that correlates diverse nanostructure dimensions with their specific reflectance spectra across visible light wavelengths. This library serves as the foundation for our analytical model. To address the inverse problem of optics, a 4th order polynomial regression with L2 ridge regularization was applied to characterize transistor dimensions from light spectra. This approach allowed for the efficient decoding of the reflectance spectra to extract precise dimensional information of the nanosheet transistors. The results showcase high accuracy in measuring all critical dimensions, indicating the method's effectiveness and potential for broad application in nanotechnology fabrication. In addition, a simulated Gage R&R study was developed to demonstrate the preciseness of this model as a measuring tool. The methodology discussed in the paper is shown to be a promising part of high throughput manufacturing for future microelectronic designs.