AVS 57th International Symposium & Exhibition
    Vacuum Technology Tuesday Sessions
       Session VT-TuP

Paper VT-TuP9
A Computationally Simple, Wafer-to-Feature-Level Model of Etch Rate Variation in Deep Reactive Ion Etching

Tuesday, October 19, 2010, 6:00 pm, Room Southwest Exhibit Hall

Session: Vacuum Technology Poster Session and Student Posters
Presenter: J.O. Diaz, Massachusetts Institute of Technology
Authors: J.O. Diaz, Massachusetts Institute of Technology
H.K. Taylor, Massachusetts Institute of Technology
R.J. Shul, Sandia National Laboratories
R.L. Jarecki, Sandia National Laboratories
T.M. Bauer, Sandia National Laboratories
D.S. Boning, Massachusetts Institute of Technology
D.L. Hetherington, Sandia National Laboratories
Correspondent: Click to Email
Modeling etch rate variation in Deep Reactive Ion Etching (DRIE) helps to identify possible defects in MEMS and IC devices arising from sub-optimal etching depths and times. Besides tool-specific properties, such as the chamber design, another cause for the observed non-uniformity effects is the particular wafer pattern employed. At the wafer scale, previous studies have shown that wafers with a large percentage of open (exposed Si) area, or pattern density, exhibit a radial center-low etch-rate distribution, while those with low pattern density achieve radial center-high etch rates. At the die scale, it is widely known that etch rate decreases as local pattern density increases. Furthermore, at the feature scale, the microloading effect describes how adjacent features tend to compete for radical species, thus decreasing overall etch rates within individual features.

We present a model to capture these pattern-dependencies by tracking the spatial and temporal distribution of the ion and radical species within the DRIE chamber. The model implementation uses a time-stepped algorithm with three levels – corresponding to the three different length scales – and a coarse-grain approach where multiple features in a given region are characterized by a particular shape, size and density. The local radical species concentration distribution above the wafer is determined at each time step using current feature geometries to compute their Knudsen transport coefficient which is linked to the radical transport mechanisms within other areas in the chamber. At the end of each time step, etch rate estimates based on this radical concentration distribution and current feature geometries are used to update feature depth information for the next time interval. At the wafer scale, our modeling results achieve a success comparable to that of previously-developed wafer-level models with an etch rate RMS error percentage between 2.1% and 8.2%. The results also show that feature-level etch evolution substantially impacts the wafer-level fluorine concentration and thereby modifies the wafer and die etch rate uniformity. We expect a similar model could be incorporated into CAD software tools to evaluate masks and correct potential design issues before they are made. Our results also shed light on possible tool and process modifications to allow users the capability of altering across-wafer etch rate variability. Sandia National Laboratories is a multi program laboratory operated by Sandia Corporation, a Lockheed Martin Company for the United States Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000.