(92c) Real-Time Thin Film Characterization during Chemical Vapor Deposition Using Moving Horizon Estimation

Thin film deposition has a wide range of applications and is particularly important in semiconductor device fabrication. The current technical trend in the semiconductor industry toward miniaturization in device size and greater complexity in structure requires closer process monitoring, fault detection and eventually closed-loop control of deposition process. To achieve these goals a real-time sensor that can measure thin film properties in situ is needed. Optical sensors such as reflectometers and ellipsometers are ideal candidates because they are compatible with the harsh processing environment of CVD. However optical sensors do not measure film properties directly. The key questions are the observability of the process using optical sensors and how to optimally estimate film properties from optical measurements.

This work used an emissivity correcting pyrometer as an in-situ sensor which measures radiation and reflectance of thin film at two different wavelengths, for a total of four measurements. The goal is to estimate thin film thickness and microstructure features like grain size and surface roughness from these indirect measurements. The challenge with estimation of film microstruture lies in the process dynamic model for microstructure evolution at small scale, which in general is extremely high-dimensional, nonlinear, stochastic and multiscale. In this work we used a simple but adaptive process dynamic model and a detailed optical sensor model to study if a simple process dynamic model could aid in estimation. Based on these models the observability of the process was studied. Then a moving horizon estimator (MHE) was used to estimate thin film growth rate, film thickness and roughness in situ from the optical measurements during chemical vapor deposition. The moving horizon estimator was compared with an extended Kalman filter in terms of estimation accuracy and computation efficiency. The effect of physical state constraints on state estimation was also studied. Experimental deposition of yttria thin film on silicon substrate via chemical vapor deposition was carried out to demonstrate this real-time sensor. The deposited films at various stages were studied ex situ using an ellipsometer for film thickness and an atomic force microscopy for surface roughness.