Deep Learning-Based High Throughput Inspection in 3D Nanofabrication and Defect Reversal in Nanopillar Arrays: Implications for Next Generation Transistors

Densely packed high-aspect-ratio (HAR) nanostructures are the core elements of future microelectronics components. Manufacturing these nanostructures for device applications requires multiple fabrication steps involving wet processes, followed by a drying step. During drying, these nanostructures experience strong capillary forces that induce their bending and cause them to permanently stick to their neighbors, a phenomenon often referred to as pattern collapse. The pattern collapse and the difficulty in reliably identifying damaged nanostructures pose a critical challenge for the fabrication of HAR devices. Here, we developed a machine learning-based approach to identify collapsed nanostructures from a large patterned array of vertical Si nanopillars with 99.84% accuracy. Furthermore, we show that the pattern collapse can be reversed by selectively etching the native surface SiO₂ layer of the nanopillars at their adhesions. Our approach for accurate and rapid identification of the collapsed nanostructures combined with the method to reverse this damage provides a versatile platform for developing high-yield fabrication processes for nanoscale semiconductor devices.