Development of Plasma Information Based Virtual Metrology (PI-VM) for Drifting Deposition Process

Abstract

In semiconductor-fabrication industry, virtual metrology (VM) is one of promising technology to achieve advanced process control (APC) for plasma-assisted process because it can provide metrology data for every wafer. VM is defined as the technology of prediction of metrology variables using process state (equipment and sensor) and wafer state variables. However, prediction reliability of VM is very sensitive to the quality of data and the selection of data mining technique, causing the limitation of its APC applications. Also, the required prediction reliability of VM is getting higher as the critical dimension of semiconductor devices sharply shrinks. In line with this trend, development of phenomenological-based VM which includes information about the plasma-assisted process reactor is important in the sophisticated control of semiconductor and display device manufacturing. Note that reactor is defined as process vessels that are used in semiconductor manufacturing industries for carrying out chemical reactions with assistance of plasma.<br/> The wafer state variables of plasma-assisted process, especially plasma-enhanced chemical vapor deposition (PECVD), are governed by the influx of reactive species on wafer, such as radicals, metastables, and ions. These influxes are mainly produced by electron-impact collision reactions with collision partners and destructed by reactions at reactor-wall. Thus, the wafer state variables are complex function of electron-energy distribution function (EEDF) which governs reaction rate constant (m3s-1) in plasma-volume, and reactor-wall condition which governs reaction rate at reactor-wall, such as recombination probability. <br/> The plasma-assisted process often drifts with the operation time caused from the phenomena related to the gradual deposition of byproducts or fragments of feedstock gas on reactor-wall. This drift of reactor-wall condition changes the flux of reactive species through the desorption of residues or deactivation of reactive species at reactor-wall. It also changes the gas composition in plasma and EEDF through additional elastic and inelastic collision processes. The changed EEDF induces the variation of influx of reactive species. Therefore, the reliable VM should include the variables that represent EEDF and reactor-wall condition. This concern is important specifically for the long-term application of VM.<br/> In this dissertation, two plasma-information (PI) variables are defined as the state variables of plasma-assisted processes affecting the flux of reactive species: Monitoring variable for reactor-wall condition (named PIWall) which represents the film buildup on reactor-wall, and another variable for plasma-volume (named PIVolume) which corresponds to electron density and electron temperature (or EEDF). These PI variables are determined by using plasma spectroscopy which analyzes the optical emission spectra based on the excitation equilibria in plasma. PIWall is determined by analyzing the light transmittance at the contaminated window based on line-intensity ratios from the same upper energy level. In the development of PIWall, it is assumed that the film deposition on quartz window is roughly following that of reactor-wall surface. PIVolume is determined by analyzing line-intensity ratios from different upper energy levels based on collisional-radiative model (CRM). <br/> The evaluation of influence of each variable on PI-VM for single-layer nitride PECVD shows that the electron density affects the formation of N containing species through governing the collision reaction with low threshold energy. It also shows that the electron temperature affects the formation of Si containing species through electron-impact collisions with high threshold energy. Meanwhile, PIWall affects the formation of N containing species, especially near the reactor-wall.<br/> PI-VM is applied to predict layer-to-layer nitride film thickness in nitride/oxide multi-layer PECVD used for 3D NAND fabrication in the mass production line. The evaluation of influence of each variable on PI-VM demonstrates that PIWall is the highest contributing variable in whole layers and PIVolume contributes in the latter region of layers. In the trend of increasing the number of nitride/oxide layers to increase device density, PI-VM offers provide the key phenomena of plasma to be managed. Therefore, PI-VM is expected to provide the fundamentals to develop advanced process control (APC) and fault detection and classification (FDC) for plasma-assisted processes. Also, it provides the way to apply the plasma spectroscopy in the basic plasma diagnostics technology to plasma-assisted process.<br/>