Smart Automation: Machine Learning Enabled Workflow for Logic and DRAM

Abstract

Current S/TEM automation requires highly trained microscopists to build specialized acquisition and analysis recipes for specific semiconductor structures. Recipes can be challenging to write and take a long time, therefore the industry continues to rely on manual operation to overcome TEM analysis demands. Time to develop robust automation is often too slow for the product change rate of industry development cycle. To overcome this barrier, Smart Automation was developed on a Thermo Scientific Metrios TM AX 200kV S/TEM as an application providing a generalized acquisition workflow that is ready to use without requiring specialist intervention. The application interacts with a Fully Convolutional Neural Network (FCN) through a user-friendly interface for navigation and auto alignments. The FCN is the key enabler for lowering the bar to automation set up with a time to robust automation within hours instead of days.

Automated (S)TEM metrology characterization of gate-all-around and 3D NAND devices

“…The alternating layers of silicon oxide and silicon nitride in 3D NAND memory cells is such a challenge. EDS metrology was previously shown [10][11][12][13][14] to greatly improve the contrast of these layers, leading to more robust metrology but at the cost of throughput and possible sample damage, relative to (S)TEM imaging. A higher-efficiency EDS detector, which can reduce the acquisition time and sample dose, is thus important for improved EDS metrology and materials characterization of low contrast and/or beam sensitive devices.…”