Electron Rutherford Backscattering, a versatile tool for the study of thin films

Electron Rutherford Backscattering, a versatile tool for the study of thin films Maarten Vos Research School of Physics and Engineering Australian National University Canberra Australia Acknowledgements: Many collaborators over the last years In particular: M. Went, R. Elliman P.L. Grande, G. Marmitt, R. Moreh, A. Winkelmann

Ubiquitous question in nano/surface science What is my sample composition 'near' the surface Two widely used tools: XPS: an electron based X-ray spectroscopy -very surface sensitive, requires UHV -often requires surface preparation - fully quantitative analysis requires the full understanding of many aspects (cross sections, line shapes, mean free path, analyser transmission) One of the main topics of ECASIA RBS: an ion based scattering experiment - measures deeper in the sample, but can also be used in a surface sensitive mode - interpretation simple (Rutherford cross section, stopping power) - routinely done fully quantitative (better than 10% accuracy) - in general does not require UHV - larger equipment ERBS is RBS with electrons. It is a scattering experiment (like RBS) and has depth sensitivity based on the electron inelastic mean free path (just like XPS). Here I want to demonstrate that it can be used to do quantitative surface analysis. The aim of this talk is to give you some insight in (im)possibilities of this technique

Definitions: -Elastic scattering: sum of kinetic energies of particles involved is constant - Inelastic scattering: kinetic energy of particles decreases due to creation of electronic excitations

Definitions: -Elastic scattering: sum of kinetic energies of particles involved is constant - Inelastic scattering: kinetic energy of particles decreases due to creation of electronic excitations before collision: e- Stationary atom k0 q = k1 - k0 k1 after collision: -q Atom (Mass M) acquires q2/2m kinetic energy, projectile kinetic energy is reduced by this amount

For ion scattering the energy loss of the projectile (mass M1) scattering from an atom with mass M2 is usually expressed as the kinematic factor k: The kinematic factor is considerable smaller than 1 (e.g. 0.5 for He scattering from Si over large angles) as M1 has of the same order of magnitude as M2. For electrons scattering from atoms the same formula applies. However now M1 is about 3 orders of magnitude smaller than M2 and k is very close to 1. (remember the electron to proton mass ratio is 1:1836) k= 0.998 for scattering from H (θ=135o) k = 0.999990 for scattering from Au Good energy resolution is essential for ERBS

Example: 40 kev electrons scattering over 135 from SiO2 on which about 1 Å of Au was evaporated.

Example: 40 kev electrons scattering over 135 from SiO2 on which about 1 Á of Au was evaporated. -Clean Gaussian peaks! A fitters delight. The area ratio of the Au, Si and O peaks can be determined with great precision Separation of peaks within a few % of calculated ones (except when sample charges)

Why different width? Atoms are vibrating, even at 0K (zero point motion) Hence energy transfer is Doppler broadened. before collision: e- moving Atom, momentum p Kinetic energy p2/2m k0 q = k1 - k0 k1 after collision: -q p-q Atom (Mass M): change in kinetic energy: (p - q)2/2m p2/2m=q2/2m-p q/m

Hemispherical analyser 0.3 ev resolution slit lens We measure a series of angles simultaneously This spectrometer was not designed for ERBS. It could be simplified in a big way.

History of 'ERBS' - First observation of shifts and broadening of the elastic peak in electron scattering H. Boersch et al -From 2000 onwards somewhat more activity in this field. Varga, Tokesi, in Debrecen (Surf. Interface Anal. 2001; 31: 1019) Vos, Canberra (Phys. Rev. A 2002; 65: 012703 ) -Detection of hydrogen: Yubero Sevilla Cooper, Hitchcock (McMaster Canada) gas phase Avanasev, Moscow and others

H2O O Example of measurement H (and D) by ERBS for the case of ice H H signal is weak, as cross section scales as Z2 O D2O at 3 kev resembles H2O at 1.5 kev, but D peak is sharper H O O D H D2O -2 Energy Loss (ev) 0 2 4 6 8

How well can one determine the sample composition? 4 samples of well-established composition, all containing oxygen, some low Z elements and high Z elements: Li2CO3 CaCO3 TiO2 HfO2 When possible evaporate Au on the surface, to check spectrometer performance and zero point of the energy scale

Experiment reproduces nominal sample composition Within 10%, better for the main peaks. Monte Carlo simulations suggest that for most cases multiple scattering effects are small and do not affect the interpretation (except when cross section of elements differ hugely)

For isotropic systems: peak width: (σ ) = 4 Ekin Erec 3 with Ekin the mean kinetic energy of the atom and Erec the mean recoil energy (q2/2m)

For isotropic systems: peak width: (σ ) = 4 Ekin Erec 3 with Ekin the mean kinetic energy of the atom and Erec the mean recoil energy (q2/2m) A comparison of measured and calculated mean kinetic energies for some compounds. (J. Chem. Phys. 143, 104203 (2015)) Kinetic energy marked by * corrected for multiple scattering based on Monte Carlo simulations. The width is more sensitive to multiple scattering effects than the composition.

Example: Study of surface modification by sputtering TiO2 film sputtered with Xe ions Now 3 peaks visible: Xe, Ti and O Measurement in 2 geometries: Xe Ti O

Example: Study of surface modification by sputtering TiO2 film sputtered with Xe ions Now 3 peaks visible: Xe, Ti and O Measurement in 2 geometries: 75 400 Xe Bulk sensitive surface sensitive Surface sensitive spectrum is not consistent with bulk stoichiometry. Needs at least two layers to fit the spectrum -one with Xe and O deficient -one pure TiO2 Thickness of surface layer,obtained from fit, is 14 nm, somewhat larger than implantation range of 3 kev Xe Ti O

Additional information from electronic excitations With the same spectrometer we can take energy loss data over a larger energy loss range, Using a lower incoming energy (5 kev, rather than 40 kev) the O peak is not resolved from the Ti peak and one can study the band gap region. After sputtering the sample is O deficient. Not all Ti atoms are in the 4+ state and their remaining d-electrons can be easily excited causing an additional hump near 2 ev.

ERBS has isotope selectivity, a rare capability in electron spectroscopy Here the signal of TiO2 layers grown using 18O or 16O Using special designed samples ERBS was used to study diffusion of O atoms in TiO2 This material is of technological interest as these materials are candidates for RERAM devices.

Samples used for diffusion study Si3N4 layer prevents O exchange During annealing stage and is removed before the measurement. As grown Annealing Remove Si3N4 cap Measure by ERBS

ERBS spectra of annealed sandwich structure Amount of Diffusion occurred as a function of annealing temperature

ERBS in single crystals Crystal lattice has a big influence! TiO2 (rutile) The variation in intensity ratio is a consequence of presence of Kikuchi lines that affect Ti and O in different ways. By comparing with calculations one can test in this way the dynamical theory of diffraction. Ti O However, there is no magic here, just looking at the crystal planes will tell you a lot.

[001] Ti [112] [113] 1) (12 (001) [110] 1) (01 [100] (011) [011] [111] (1 21 ) [101] [112] (110) (101) (0 31 ) [111] (110) (01 0) 0) (10 [113] O [110] [210] 1) ( 1 10 (1 1) (001) [010] [110] Calculated influence of the outgoing direction on the measured intensity Some planes very similar e.g. (001) plane (type a) Some planes are bright for Ti but dark for O e.g. (011) plane (type b) Some planes visible for O but not for Ti e.g. (111) planes (type c) Some planes strong for Ti but weak for O e.g. (121) plane (type d)

Our analyser uses slit lenses and the detector tells us where the electron went through the slit Thus we can measure the intensity along a line. By rotating the crystal we can choose where we measure. Experiment (solid line) and theory (dynamical theory diffraction, Winkelmann) (dashed) show good agreement

[001] Ti [112] [113] 1) (12 (001) [110] [100] 1) 1 0 ( (110) [101] [112] D type Red: Ti atoms Blue O atoms [111] [110] [210] (011) [011] (1 21 ) (101) (0 31 ) [111] (110) (01 0) 0) (10 [113] O 1) ( 1 10 (1 1) (001) [010] [110] Easiest to understand using time reversal, Consider an electron coming from the analyser Impinging on the crystal Due to diffraction standing waves form when Bragg conditions is fulfilled (basically along the crystal planes) Maximum intensity if atoms near maxima of standing waves.

Metal interface formation Combining the measurement of atomic and electronic structure in the same experiment -Mo and Pt foils were used -Al was deposited on these foils.

Al concentration at surface can be monitored around 2 ev energy loss (Al elastic peak). At larger energy loss we see the `plasmon' spectra as in (Reflection) electron energy loss spectroscopy (REELS). The change in the evolution is very different for Mo and Pt. For deposition on Mo the Al plasmon at 15 ev is visible straight away, but for deposition on Pt no characteristic Al plasmon appears initially. Interpretation: for Mo an Al film is formed straight away after evaporation, but for Pt the Al reacts with the Pt and forms an alloy.

For much thicker evaporated layers pure Al is formed as well on Pt. Note that the plasmon appears now split. Are there two different plasmons?

For much thicker evaporated layers pure Al is formed on Pt. Note that the plasmon appears now split. Are there two different plasmons? No, remember all detected electrons have scattered elastically as well. Thus detected electrons have plasmon + recoil loss The electrons that scattered from Pt and created a plasmon have a smaller total energy loss than those scattered from Al.

Quick note on the analysing REELS measurements Normally REELS measurements are analysed by either first extracting an effective loss function from a single measurement (TougaardChorkendorff procedure) or alternatively by extracting a bulk and surface loss function from two spectra (Werner) As an alternative approach we can also fit directly a set of REELS spectra based on a model dielectric function and the corresponding bulk loss function (DIIMFP) and surface loss function (DSEP) Procedure uses partial intensities as input (obtained from Monte Carlo simulations), but is surprisingly insensitive to the precise values used. Here demonstrated for copper

Copper Calculate partial intensities from Monte Carlo Calculate normalised DIIMFP and DSEP at each energy assuming a model dielectric function (set of extended Drude oscillators) Spectra constructed from sum of n-fold convolution of the DIIMFP weighted by partial intensity Surface excitation parameter is left free, to be justified afterwards but appears to be reasonable The same dielectric function used for all measurement, and optimised by simultaneous fitting of all 6 spectra.

Copper } REELS } Optical The dielectric function obtained in this way is very consistent with the optical data down to 2.5 ev

Summary of ERBS -The good Simple peak shapes, well understood cross sections Good for quantitative analysis (but be careful with single-crystals) Multiple scattering can often be ignored, at least in first approximation Simple interpretation Larger probing depth compared to XPS, surface preparation often not required Information on electronic structure as well as elemental composition Sensitive to crystal structure New physics to be explored Laboratory based technique, requires less expensive equipment than RBS, When made commercially, should be comparable in cost to XPS

If you have a sample for which ERBS could give help addressing any outstanding questions, come and see me.

If you have a sample for which ERBS could give help addressing any outstanding questions, come and see me. Thanks for your attention!