AVS1996 Session ASTuM: Advanced Quantitative Analysis
Tuesday, October 15, 1996 8:20 AM in Room 105B
Tuesday Morning
Time Period TuM Sessions  Abstract Timeline  Topic AS Sessions  Time Periods  Topics  AVS1996 Schedule
Start  Invited?  Item 

8:20 AM 
ASTuM1 Comparison of Two Models and Equations for Predicting Electron Inelastic Mean Free Paths
S. Tanuma (Japan Energy ARC Co. Ltd.); C. Powell, D. Penn (National Institute of Standards & Technology) Gries [1] has recently developed an atomistic model for inelastic electron scattering relevant to AES and XPS and has derived an equation (designated G1) for the estimation of inelastic mean free paths (IMFPs). We present an analysis of the Gries model and the G1 equation in terms of the similarities and differences of inelastic electron scattering by free atoms and by solids. We also compare the G1 equation with our TPP2M equation for IMFP estimation. The former equation was developed from fits to our published IMFPs over the 2002000 eV energy range, and is identical in its energy dependence to the Bethe equation for inelastic scattering cross sections and to a simplification of our TPP2M equation for the same energy range. Comparison of parameters indicates that the Gries fitting parameter k/sub 1/ should be approximately constant (for all materials) although there can be substantial (up to 50%) and unpredictable variations for some materials. In contrast, the TPP2M equation is believed to be applicable to all materials. Although we believe that the Gries model is inconsistent with current theories for the electronic structure of most materials, we find (from sumrule considerations) that it provides a convenient means for estimating IMFPs in compounds, although again there are unexplained deviations. Gries claims that the G1 equation can be extrapolated to energies lower than 200 eV on the basis of limited agreement with some experimental IMFPs over the 10100 eV range for Be and the alkali metals, and has questioned the reliability of our IMFPs for energies below 200 eV. We consider this comparison to be inadequate, and we recommend that the G1 equation not be used in the 50200 eV range.1. W. H. Gries, Surf. Interface Anal. 24, 38 (1996) 

8:40 AM 
ASTuM2 Quantitative Analysis of Overlapping Spectral Features
N. Russell, J. Ekerdt (University of Texas, Austin) Information from closelyoverlapping spectral features is often extracted through the use of nonlinear curvefitting procedures. Here, we examine the ramifications of a number of frequentlyemployed assumptions in the curvefitting procedure from a statistical viewpoint, with particular reference to the analysis of Xray photoelectron spectra. In particular, we show with the aid of simulated data that the signaltonoise ratio required to resolve overlapping features depends strongly on whether the peak widths or positions are assumed to be known. The joint confidence region indicates a strong interaction between the relative intensities of closelyoverlapping features; a substantial increase in signaltonoise over that required to identify the second feature is needed to estimate the intensities of the features with reasonably high precision. A phasediagram is developed that shows the combination of signaltonoise ratio, relative peak heights and instrument resolution needed to identify and precisely estimate the relative contributions of overlapping features (with 95% confidence) as a function of the separation between peaks. Experimental data for the Si2p XPS feature, obtained with unmonochromated MgK\sub \alpha\\ radiation, is analyzed for various combinations of pass energy and dwell time to verify these results by determining the experimental conditions under which the wellknown 2p\sub 1/2\ and 2p\sub 3/2\ features may be identified, and estimated within 10% relative precision at the 95% confidence level. 

9:00 AM 
ASTuM3 Analysis of Carbide and Nitride Films by Auger Electron Spectroscopy
P. Schm\um o\lz, H. St\um o\ri, W. Werner (Technische Universit\um a\t Wien, Austria) Auger Electron Spectroscopy (AES) is a relatively old and common method for surface chemical analysis, widely utilized in metallurgy and semiconductor technology. Nevertheless theoretical understanding of some processes involved is still incomplete. Recent developements in the area of electron transport allow for improved quantitative interpretation of Auger spectra. Although the methods presented can be applied to spectra of a large variety of samples, the most important improvement is to be expected for systems like refractory metalÊ nonÊmetal compounds. A mayor difficulty in interpreting Auger spectra derives from their complex peak form, which is due to the scattering history of electrons leaving a sample. Theories of electron scattering allow for simple correction methods, provided that the precise scattering properties are known. Unfortunately these vary with chemical composition, rendering a clean separation of various peaks unfeasible. This problem calls for approximate methods, which allow to deconvolute Auger spectra while preserving their essential features, especially peak areas. A method is presented that shows several advantages over other methods like Peak to Peak Heights and spectrum deconvolution methods in common use. Another weakness of AES is the necessity of an external standard, calibrated by absolute methods. An approach is demonstrated which allows for internal calibration of binary systems. This is achieved by recording several spectra which are well distributed over the whole stochiometric range and which are calibrated to yield absolute intensities. Validity of these methods are demonstrated on binary nitride and carbide systems. 

9:20 AM  Invited 
ASTuM4 101 Uses for Eigenvectors
D. Watson (Physical Electronics, Inc.) Or so it may seem. The availability and lower cost of scientific computing has resulted in a steady increase in numerical experimentation aimed at bringing the field of surface analysis out of the data reduction dark ages. This has happened largely within the past five years. At first glance it might appear that there is an unlimited number of techniques available for the extraction of information from spectral data. However, most of the successful methods are quite closely related.This talk will focus on the multivariate data analysis methods that are often referred to under the broader heading of "chemometrics". In particular, methods that have shown the greatest benefit in our laboratories, Linear Least Squares (LLS), Target Factor Analysis (TFA), and Partial Least Squares (PLS) will be discussed in terms of what can and cannot be expected from them. Variations on these techniques will also be mentioned. 
10:00 AM 
ASTuM6 Advanced Data Analysis Techniques in XPS Quality Control Applications
T. Carney, G. Jones, K. Robinson, J. Wolstenholme (VG Scientific, United Kingdom) Modern XPS instruments with improved monochromatic Xray sources and high performance lens technology offer very high sensitivity enabling rapid analysis of samples, resulting in a high throughput of samples. Hence, XPS is now routinely used for quantitative analysis of surfaces in quality control (QC) environments. The typical XPS analysis area is now less than 1mm and hence a large number of analysis points may be selected, even from relatively small specimens. This enables acquisition of data from many small specimens, or acquisition of repeat data from the same specimen. Such repeat data may be used in the calculation of errors. When combined with an automated sample manipulator, a large number of analyses can be performed automatically overnight, leading to a large amount of data to be processed. In routine quality control applications, the most important requirements are to determine that the material composition is constant and where the material does deviate, to identify the other elements which are present.This paper will show how mathematical techniques such as Target Factor Analysis and NonLinear Least Squares Fitting (NLLSF) can be used to simplify the process of data analysis from multiple similar samples. A "standard" spectrum, acquired from a sample which is known to be satisfactory, is defined as a reference. This reference spectrum, which may either be a survey spectrum or from a high energy resolution region, is fitted to the spectra from the analysed samples. The NLLSF method allows the energy of the reference spectrum to be adjusted to obtain the best fit to each of the sample spectra. A poor fit is indicated by an increased misfit parameter and such spectra can be quickly identified. The residual spectrum can then be used to identify which peaks have changed. The implementation and examples of the use of this method will be presented. 

10:20 AM 
ASTuM7 Depth Profiling in XPS and AES by Means of Partial Intensity Analysis
W. Werner (TU Vienna, Austria) Partial Intensity Analysis of electron spectra is a rigorous quantitative spectrum evaluation technique. It is based on the separation of the contribution of electrons that have suffered a given number of inelastic collisions. These contributions, the socalled partial intensities, contain all information regarding the specimen's depth profile assessable to quantitative electron spectroscopy. Thus an entire spectrum (peak and background) can be subjected to a partial intensity analysis yielding not only the specimen's depth profile in the surface near region, but also the true intrinsic spectrum. A particular advantage of the method is that it is easy to account for a very involved signal generation process. This implies that phenomena like elastic scattering, overlapping peaks, the exact experimental geometry, surface roughness, depth and energy dependent scattering characteristics etc. can be taken into account in routine quantitative analysis. The method is illustrated by application to samples consisting of thin Au layers on Pt and Ni substrates. 

10:40 AM 
ASTuM8 An ARXPS Investigation of Nitrided Gate Oxides
B. Tielsch, J. Fulghum (Kent State University); W. Harris, L. Le Tarte (Digital Equipment Corporation) A variety of new processing methods are currently being investigated for the modification of SiO2 films in ultrathin gate dielectrics. In particular, nitrogen incorporation near the SiO2/Si interface has been shown to improve transistor characteristics in MOS devices. For this reason, the N distribution and chemical bonding in these materials is of interest. However, the low N concentrations and shallow distributions make the use of destructive depth profiling methods difficult. Etchback methods can provide significant information, but are timeconsuming and thus not practical on a routine basis. Due to the thickness of current technologically important gate dielectrics however, the possibility of nondestructive profiling via angleresolved XPS becomes feasible. We have investigated the application of the ARXPS technique to gate dielectrics grown using several different processing methods. Differences in the resultant N distributions are shown to qualitatively affect the ARXPS data. Results from several different methods of ARXPS data analysis for the determination of N distribution and peak concentrations will be discussed and compared to Auger depth profiles. This study shows that ARXPS analysis of nitrided gate oxides is useful in relative comparisons of processing conditions for thin nitrided gate oxides. 

11:00 AM 
ASTuM9 DepthProfiling by AngleResolved XPS
P. Cumpson (National Physical Laboratory, United Kingdom) AngleResolved Xray Photoelectron Spectroscopy, ARXPS, is a nondestructive technique which can be used to probe chemical concentration as a function of depth from the surface to about 5nm. ARXPS is particularly valuable if chemical state information is important, since this is very rapidly destroyed by alternative depthprofiling methods involving sputtering.Many methods of analysis of ARXPS data have been proposed over the last 20 years, from simple methods of overlayer thickness measurement (used to check hard disc surfaces) to complex depthprofile calculations based on Maximum Entropy methods. Using a sampling argument successfully applied elsewhere, these methods are placed in a unified framework\super 1\, providing definite conclusions showing which is best for which type of specimen, and the optimum experimental configuration for ARXPS.Archetypal algorithms which implement each possible approach have been written into a popular computer spreadsheet, allowing the analyst to easily compare quantitation options. Results for a range of analytical specimens will be illustrated, showing the advantages and implicit assumptions of each approach.\super 1\ P J Cumpson, J. Elec. Spectrosc. 73 (1995) 2552. 

11:20 AM 
ASTuM10 Pooling Analysis of Scanning Probe Microscopy Images
P. Dooley, S. Bernasek (Princeton University) As the applications and popularity of scanning probe microscopy have increased, the volume of qualitative data (surface images) introduced into the scientific literature has skyrocketed. Although informative, these images can easily be misleading; there is no accepted standard method for the posttreatment of this data nor even an absolute set of definitions for the common data treatment techniques. This latter problem seems addressable; however, the range of image types currently being produced would invarably render any single (or even multiple) standard data treatment technique impotent. Acknowledging this problem, the need for quantitative image treatments is apparent. This imperative has been addressed by several popular techniques including fractal, fourier, and correlative analyses. We submit a new approach, pooling analysis, that we feel should have general applicability, ease of interpretation, and a fairly faithful representation of the surface topology while reducing the data set significantly. By analysis of the histograms of both the volumes and surface areas of surface cavities that are capable of holding a fluid, we are able to characterize many surface morphologies and make quantitative comparisons between them. Both the original surface image and its inverse are subjected to the analysis in order to extract as much information as possible. We present several model and real surface images that we have analyzed by pooling analysis as well as the more common quantitative techniques. We show that pooling analysis often gives a very accurate assessment of the surface while remaining grounded in a very physical interpretation. 