Characterization of Atomic Layer Deposition using X-Ray Reflectometry

Abstract

Abstract.This work addresses current limitations of X-ray reflectometry (XRR) for modeling thin films and provides a basis for their improvement. Better accuracy in the characterization of novel thin film structures requires better model selection techniques and better knowledge of the theoretical limitations of current XRR analysis techniques. We use hafnium dioxide (HfO 2 ) nanoscale (≈ 1 nm) thin films deposited by atomic layer deposition (ALD) to study the limitations of current techniques. These structures are of strategic importance as CMOS gate and barrier materials. We show that XRR modeling-for our measured data range and counting statistics-will fail for thicknesses less than 1 nm. We also show that a two-layer model (HfO 2 /SiO x Hf y /Si substrate) is more plausible than a one-layer model (HfO 2 /Si substrate) for the measured data.

X-Ray Reflectometry Determination of Structural Information from Atomic Layer Deposition Nanometer-scale Hafnium Oxide Thin Films

“…In Table I, we compare the MCMC XRR results with genetic algorithm (GA) XRR and XTEM results from previous analysis [9]. For model 1 and model 2, we see excellent agreement between the GA and MCMC data refinement.…”

“…This paper demonstrates the effectiveness of a Markov Chain Monte Carlo (MCMC) sampling approach for XRR refinement. The film was deposited on a Si wafer for 65 cycles with a per cycle deposition rate of 0.088 ± 0.006 nm [9]. This work uses XRR with MCMC refinement to examine the thickness of an atomic layer deposition (ALD) Hf x O y thin film with nominal thickness of 6 nm [8].…”

Determining sample alignment in x-ray reflectometry using thickness and density from GaAs/AlAs multilayer certified reference materials

“…A second study was performed with a wider range of allowed roughness parameters which uses table 1 and extends allowed roughness values to 0.1-1.0 nm for layers 2-8 (and 0.1-2.0 nm for layer 1) to address issues discovered in the refinement of surface layers for highly misaligned XRR data. We performed GA refinement with both NIST-developed software [11,12] and a commercial package for comparison (Bede REFS 4.5). Results from both software packages and both parameter ranges are shown.…”

“…Data taken in 2004 by researchers at NMIJ on a similarly deposited structure, is the focus of this study. The National Institute of Standards and Technology (NIST) has been developing Bayesian approaches (see Sivia [9]) to estimate uncertainty in modeling XRR parameters [10,11].…”