Depth profiling of light-Z elements with elastic resonances: Oxygen profiling with the 3.045 MeV 16O(α, α)16O resonance

Nuclear Instruments and Methods in Physics Research B66 (1992) 283-291 North-Holland

NuclearInstnanents &Methods jn physicsResesrch Section B

Depth profiling of light-Z elements with elastic resonances : Oxygen profiling with the 3 .045 MeV '60(a, (X) 16 0 resonance W. De Coster, B. Brijs, J . Goetnans and W. Vandervorst 1MEC vzw., Kapeldreef 75, B-3001 Leuven, Belgium

Depth profiling of light elements in high-Z substrates can be done with RBS using elastic resonances . With the resonant signal in the backscattered spectrum several parameters can be associated such as height, position, yield and FWHM. The dependence of the parameters on incident beam energy will be described. The influence of the energy spread and of the detector resolution, which limits the use and accuracy, is discussed . Computer calculations are given to establish the validity of the model and to illustrate the restrictions . A methodology which uses the different parameters is proposed. Experimental results are shown for the depth profiling of oxygen in 20 nm, 89 nm and 500 nm thick oxides with the elastic 3 .045 MeV tbO(a, (X) 16 0 resonance.

1 . Introduction Detection of light elements (Z < 10) is very important in material analysis . Among the available analysis technique Rutherford backscattering spectroscopy (RBS) is interesting because it provides a quantitative depth profile without inducing major damage to the material [1]. However, due to low scattering cross sections for light elements, conventional RBS (2 MeV 4 He) is hampered by a low sensitivity for these elements. This is especially the case for the detection of light elements in high-Z substrates since then the low light-element signal is superimposed on a high "background" signal from the substrate . While keeping the quantitative nature, depth profiling of light elements can be done by RBS with higher sensitivity by employing nuclear resonances in the elastic scattering cross sections. Although the background of the substrate is still present, the signal to background ratio can be highly enhanced. Compared to nuclear reaction analysis (NRA), where the background can be totally discarded in some cases, resonant RBS has the advantage that the same equipment as for conventional RBS is sufficient . Oxygen is among the most important light elements in semiconductor industry. It is frequently present in processing, sometimes as an unwanted contaminant, other times as a necessary element to achieve the desired material properties. Depth profiling of oxygen by resonant RBS can be performed by employing the "0(a, 0`0 resonance at 3 .045 MeV [2]. This resonance has recently been applied for the analysis of the oxygen contents in high-Z materials [3,4]. Compared to

high energy scattering at 8.8 MeV where non-Rutherford scattering gives also an enhanced sensitivity for 16 0 [5], the 3 .045 MeV resonance has a better depth resolution and at these moderate energies, which can still be reached with small accelerators, no resonances will occur in substrates with Z > 12 (magnesium) [6]. Two methods can be applied for quantitative depth profiling when employing resonances . One method is to derive the depth distribution from the measured spectra by spectrum synthesis: An assumed sample composition is iteratively modified to obtain a good match of the predicted and the measured energy spectrum . This method, which is commonly used with conventional RBS analysis, has the disadvantage that it will become very difficult for complex samples and quickly varying cross sections. The second method is resonance scanning whereby the incident beam energy is step by step incremented to obtain resonant scattering at an ever increasing depth . The area of the resonance peak is measured at each energy to obtain a yield versus beam energy curve . This curve can then be converted to a concentration versus depth profile of the resonant element . The second method requires that a sequence of RBS spectra is taken, out of which each time the yield is extracted . We will discuss here other parameters that can be extracted from the spectrum and which carry also information about the sample. The dependence of these parameters on the incident beam energy will be discussed. It will be shown that energy spread and especially the detector resolution tend to set a limit to the application of the parameters other than the yield.

0168-583X/92/$05.00 0 1992 - Elsevier Science Publishers B .V. All rights reserved

Ill . CONTRIBUTED PAPERS

284

ltd De Coster et al. / Depth profiling offight-Z elements with elastic resonances

2. Resonance probing

2.2. Peak parameters versus beam energy

2.1 . Parameters

The dependence of the peak parameters on beam energy will first be discussed from first principles, A commonly used resonance to "probe" the oxygen assuming ideal conditions . Afterwards the contribucontent is the well known î60(a, «)16 0 resonance at tions of beam energy spread contributions and detector E R = 3.045 MeV with a resonance width rR = 10 keV resolution will be discussed. We have worked without [2]. For a given incident beam energy, the presence of subtraction of the nonresonant background for the the light element at the depot where the beam energy parameters because the presence of this nonresonant is reduced to the resonance energy, will give rise to a yield is the cause of some of the features which are backscattering spectrum . This light-elepeak in the observed . Also the oxygen profile is assumed to be ment signal is usually superposed on a high backcontained in the top layer of thickness t. ground from the high-Z substrate, so that background With no energy spread of the beam and infinitely subtraction is necessary to isolate the light-element small detector resolution, the (integral) yield A and signal . With the resulting peak four parameters can be the channel height H for the element are given by (see associated : (1) The peak position E P, the energy at chapters 3 and 4 of ref. [11): which the resonance peak is detected . This is given by a channel number which has to be converted to energy. A(EO)= f`dA(x)dx (2) The peak height HP, the counts in the channel corresponding to the peak position . (3) The peak intea ENx E _ Q12 E,1 gral AP, the area under the resonance peak or of the dE, (1) cos 0, IE,,%E, (dE/dx),((E) total light-element signal . To obtain only the contribution due to resonant scattering, the nonresonant conN(x) dx tribution has to be subtracted . (4) The peak width rp, H(E1) =o,(Ex)QI2 the FWHM of the resonance peak or of the total cos 0 1 signal. To obtain the FWHM of the resonance itself, e(KEx) the nonresonant background has to be subtracted. (2) the FWHM of the total signal with the Measuring [E(Ex)l cos 0, E(E,) nonresonant contribution gives an additional increase of the FWHM . Fig. 1 shows measured RBS spectra of Here dA(x)=(Q .f2/cos0 1 ) o, (Ex )N(x) dx is the SiO, layers of different thickness on a silicon subscattering yield from the element in the slab between strate . The inset shows the oxygen signal of the 5000 %1 depth x and x + dx . N(x) is the atomic density of the thick layer after (quadratic) background subtraction element in the slab, Ex is the mean beam energy in the with the relevant parameters indicated . slab. AE, is the energy loss corresponding to the layer thickness AE,,, = fo'i-",(dE/dx) dx . E, is the energy corresponding to the channel, K is the kinematic factor, e is the stopping cross section and [e(Ex)] the stopping cross section factor. !2 is the solid angle subtendedby the detector, Q is the number of incident particles, 01 is the tilt angle and .1 is the width of a MCA channel. The backscattering angle 0 is generally fixed in the case of resonance profiling : 0 = 165° for 160(

Fig. 1 . Normalized 3.045 MeV °He spectra at 166.1° of different SiO2 films on Si: 204 Â (solid line) and 892 Â (dashed line). The inset shows the simulated 50(10 Â oxygen signal after background subtraction. Relevant parameters are indicated .

., a)160.

It is always possible to identify a maximum signal height in the spectrum . If a resonance is occurring in the sample, it will give rise to a peak in the signal height, which in any way will present a local maximum. To be able to assign the absolute maximum to the resonance, three energy regions have to be considered : (a) Eo 5 E R. There is no resonance andsince the cross section increases with E, the signal maximum will be situated at the front face of the oxygen containing layer. The position of this maximum is E M KEa and the height is HM a tr(Eo)N(0) dx . Both height HM and position E M will increase with energy . When E = ER the resonant cross section

W. De Coster et al. / Depth profiling of light-Z elements with elastic resonances occurs in the top slab: H M = Hp a a,NO dx and EM = E p = KER . (b) EO > ER . The resonance will occur at the depth x R where the incident beam energy is reduced to E R . The resonance will present a local maximum with height Hp a ?RN(XR) dx and E p - KE R f0fal"" J'(dE/dy)I oat dy, where the integral is the energy lost by the ion on its outward path and 0 is the angle between the backsrattered beam and the sample surface. Because of this additional energy loss the resonance will occur in the spectra at an energy Ep < KER . However, the height Hp will remain constant, neglecting the influence of straggling which will be discussed further on . If E0 increases, the resonance will occur at ever larger depths and the peak position will shift to ever lower values, until the backside of the layer, containing the element, is reached at an energy E0 -E R + AEeq. (c) E0 > ER + 0 Eeq. Resonant scattering no longer occurs in the film . Since for energies above E R the cross sections decrease, the highest cross section now occurs at the backside of the layer: H M a Q(Eo -DEeq )N(t)dx and EM -K[E0 -A Eeq ]_ f0/sin 0(dE/d y) I oat d y . Thus the height HM decreases while the position E M increases with energy. The integral A p is given by eq . (1) and this for every energy region . The FWHM Tp is determined by the convolution of the oxygen profile with the scattering cross section curve and with the energy loss function corresponding with the layer . In the ideal case the energy loss function would have a zero-width distribution . Even in this simplified case, there is no analytical solution for fp(E o), so that it has to be extracted from numerical calculations of the oxygen signal for given beam energy . The dependence of the parameters on the beam energy will be discussed further in section 2 .4. 2.3. Influence of energy spread and detector resolution It is hereby assumed that, to an approximation, the energy loss function can be described by a Gaussian distribution function . However, for the small energy loss close to the surface the use of Vavilov distributions would be more correct [7]. The different contributions to the energy spread are also assumed to be described by independent Gaussian distribution functions, of which the variances at each depth can be calculated by the theory as outlined by Williams and Mdller [8]. The inherent energy width of the beam together with straggling and multiple scattering along the inward path causes the particles at each depth x to be spread out over a number of energies. Resonant excitation will therefore occur from a certain depth interval

285

8xexo, around the mean depth XR. Since the different contributions to the energy spread SE(x) at the depth x can be added in quadrature, this depth interval is equal to sXexc

SE( z R ) (1/cos 0,)(dE/dx) I in

=

SEbeam + SEstr,in + SEBeo.in (1/cos 8,)(dE/dx) I in Here is BE, the energy spread of the incident team, SE-.i n is the energy spread due to straggling with inclusion of multiple scattering along the incident ton trajectory and SEgeo.,n is the geometrical contribution for the incident path (generally negligible) . The scattering yield dA(x) from a slab between x and x + d x then becomes dA(x) o

- c

% 6 1 {'0

fbeam(E0 , E,) dEi

f

X 0 (fspread(Ei , E, x)rr(E)) dE IN(x)) dx . The first integral accounts for the incident beam energy spread with a Gaussian distribution function: 2 1 (Ei - Eo) - exp fbeam(E0 , E,) = 2 2S,,eam 1, Sbeam 2~r with E0 the mean beam energy and Sbeam the standard deviation of the Gaussian with FWHM = SEbeam The second integral accounts for the energy-loss straggling of a given component Ei of the incident beam . The energy spectrum inside the target is also described by a gaussian distribution function : 2 1 (E - É(x)) exp , fspread(Ei, E, x) = 2 S2spread(X) , Sspread 2Tr where É(x) = Ei - fô,eos e,(dE/d y) I oat dy is the mean energy at depth x to which an incident ion with energy E i is reduced, Sspread(X) is the depth dependent standard deviation, which includes energy-loss straggling as well as multiple scattering and geometrical contributions. Calculation of eq. (4) at each depth will give us the excitation function for the light element (oxygen): this is the amount of backscattering at each depth . Integration of eq . (4) over the whole layer containing the light element, would give the integral yield of that element. The effect of energy spread is a smearing out: the oxygen contribution at a certain energy comes from a certain depth interval and the excitation function will broaden . III . CONTRIBUTED PAPERS

286

W. De Coster et al. / Depth profiling of light-Z elements with elastic resonances

Since the elastically scattered particles have to travel backwards through the sample to reach the detector, contributions along the outward path as well as the

40f,

Ideallzed 0keV

finite energy resolution of the detector have to be taken into account to describe the actual shape in the spectrum. The additional energy spread of the outgo-

Peak maximum . i

10 keV 20 keV

ing particles, induced by straggling, multiple scattering

and geometrical contributions from the effective detector acceptance angle, combined with the detector reso-

lution SEde,, which will be under normal incidence the major component, can be described by a Gaussian energy distribution. Convolution of this gaussian with the excitation function gives an additional smearing out of the oxygen signal in the spectrum. The dependence of the peak parameters on the detector resolution is

10

discussed in section 2.4. a---1120 1140 1060 1080 1100

2.4. Calculations and results Spectrum simulations are performed with a modified version of the simulation package RUMP [9]. The program has been in-house adapted to handle the resonant scattering of 4He 2+ from oxygen . All sharp and broad resonances between 2.4 and 6.5 MeV have been included.The adaptation follows the scheme outlined in refs . [10,11]. To obtain the spectrum with

energy spread inclusion, the idealized spectrum has to be convoluted with a Gaussian with depth(-energy)-

Backscattered Energy (keV)

Fig. 2. Calculated contribution of the oxygen signal to 3.045 MeV °He backscattering spectra of a 892 A Si02 film on Si for different BE,: 0 keV (solid line), 10 keV (dashed line) and 20 keV (dotted line). The smearing out of the resonant signal is visible, yielding a decrease of the peak height and a shift in position to lower detected energies, as well as the increase in FWHM. However, the area of the signal remains constant.

dependent variance . The variance at each depth is

calculated according to the theory as outlined by Williams and M81ler [8j . For the geometrical contribu-

It is important to notice that the shape of the resonance peak in the elastic cross section curve is not

tions the beam diameter is taken to be 2 mm, the distance of the detector to the target is 10 cm and the

symmetrical. There is also a nonresonant `background' to the sides of the resonance peak. Therefore the

detector width in the scattering plane is 7 mm . This corresponds to an effective detector acceptance angle

resonance can not be described by a Breit-Wigner dispersion formula, which is symmetrical. In fact a thin film with constant oxygen concentration will sample the cross section curve from the beam energy Eo down

of > 4°, which is too large to achieve good resolution, considering the rapid variation of the resonant cross

section with backscattering angle 8. In fig. 2 the calculated oxygen signals are shown for resonant scattering at 3.045 MeV from 892

A

to the energy Eo- AEeq . This is clearly seen in fig. 2 where the reference signal (without energy spread) is

thick

proportional to the oxygen cross section curve, sampled

Si02 layers on Si, and this for different values of the detector resolution. The height and yield are given in normalized units, according to the RUMP notation [9].

by the 892 oxide from ER= 3.045 MeV down to ER -A Eeq = 3.027 MeV. Thus the oxygen signal has clearly an asymmetrical shape with a median at an

The 892 Â Si02 gives an equivalent energy loss AEeq = 19 keV for beam energies around 3 MeV. Also

A

shown is a reference signal, calculated with inclusion of

energy lower than ER. Since the convolution of this signal with a Gaussian to include the energy spread and detector resolution, acts as a kind of smoothing

only the straggling contribution . As can be seen the signal tends to be smeared out with increasing detector

and averaging through the nonzero tails of the Gaussian, the maximum of the convoluted signal will shift

signal integral yield, however, is independent of the detector resolution and thus remains constant. The

on the beam energy for different detector resolutions SE&, The signal height and the integral yield are

the layer containing oxygen . A is thus only dependent on contributions along the inward path .

decreases with increasing SEd_ With increasing 8Ed,t also the shape of the curves becomes more rounded

resolution: the signal maximum shifts towards lower energies while the maximum height decreases . The

towards the median and will decrease in height . Figs. 3a-3c show the dependence of the parameters

total yield A depends only on the scattering excitation function. and is obtained by integration of eq. (4) over

shown in fig. 3a. Whereas the yield is independent of the detector resolution, the maximum signal height

W. De Coster et al. /Depthprofiling of light-Z elements with elastic resonances

and the plateau, corresponding to the constant oxygen concentration, decreases in width . The yield curve, on the other hand, shows no plateau at all because the yield corresponds to a much larger "sampled" depth interval than for the height . The position of the signal maximum exhibits the behaviour described in section 2.2 (fig . 3b) . Around E R the resonance occurs in the sampl, and the position decreases with increasing beam energy as the resonance is at fixed energy. Outside the energy interval [ER, E R + AEeq ] the position of the maximum increases . For larger detector resolution, the position is

287

shifted towards the median and thus the backscattered energy interval between the two turning points, corresponding to the two interfaces, decreases. These turning points correspond to the interfaces and occur at fixed beam energy independent of SEd _ Fig . 3c shows the FWHM of the oxygen signal measured without nonresonant background subtraction . This FWHM is a complex function of beam energy, determined by the convolution of the concentration profile with the scattering cross section curve. Although no measure of the resonance width, it gives an indication about the general shape of the oxygen

U N E

11151 1110

40

- ow low

C

35

10bV

1105

S d

30

Y

~ 25

m 1100

w

1095

20

1090

15

1085

3020

3040

3060

Beam Energy (keV)

3080

103020

3060 3040 Beam Energy (keV)

3080

Fig. 3 . Calculated peak parameters versus beam energy for a 892 Â Si0 2 film on Si for different Md., : 0 keV (solid line), 10 keV (dashed line) and 20 keV (dotted line). (a) Resonance height and integral yield. The probing resolution decreases with increasing detector FWHM. The yield curve is independent of the detector resolution . (b) Resonance peak position . The decrease of the energy range between the turning points due to increasing detector is clear. (c) FWHM of the oxygen signal . The convolution of resonance and profile becomes more obscured with increasing 8Edet. III . CONTRIBUTED PAPERS

288

W. De Coster et al. / Depth profiling of light-Z elements with elastic resonances

signal in the spectrum. For SEEdt = 0, the decrease of the FWHM for Ea < 3045 keV and the increase for E > 3065 keV with increasing energy is due to respectively the occurrence and the disappearance of resonant scattering in the oxide layer. Between these energies the signal is very peaked. With increasing SEde, the features are smeared out and the oxygen signal will have less sharp features. The influence of the detector resolution on the parameters is shown in fig. 4 . Various oxide thicknesses have been considered for a 3 .045 MeV He beam at normal incidence. At low SEde, the FWHM is little influenced while at high SEde, it increases with the detector resolution (fig . 4c). With increasing SEde, the

height decreases, the more rapidly the thinner the oxide is because then the tails of the Gaussian contribute less to the signal (fig . 4a). Also the peak position decreases with increasing SEde, (fig. 4b). It decreases towards the median of the profile until at high SEddt these become identical : the shape then is Gaussian, dominated by the detector resolution. The former results are obtained in the assumption of perfectly flat samples. Surface topography will also influence the shape of the spectrum . Surface roughness (with random distribution) will contribute to a further degradation of the energy resolution. It can roughly be described by an additional Gaussian for both inward and outward path, to be included in the convolution . A

Fig. 4. The calculated influence of the detector resolution is shown for backscattering at 3.045 MeV from 3000 .4 (solid line), 892A (dashed line) and 204 Â (dotted line). (a) Resonance peak height, (b) peak position

oxygen signal.

three oxide thickness : and (c) FWHM of the

W. De Coster et al. / Depth profiling of light-Z elements with elastic resonances

detailed discussion is not attempted in this paper. Surface roughness is further not considered . 3. Methodology In view of the previous, one could consider depth profiling methodologies using the different parameters . Both peak height and peak area (yield) can be used as a probe to determine the element concentration as a function of depth. The peak position would in principle allow to calculate the depth at which the resonance occurs. However, the depth scale can be determined more conveniently and more accurately from the incident beam energy. The peak position could be used to assign local maxima to the occurrence of a resonance in the case of a peaked profile . If the resonance occurs at low concentrations, the profile maximum can yield a higher nonresonant signal. Following the position as a function of beam energy allows to discard points that do not correspond to resonant scattering. The FWHM is of no direct use for determining a depth profile and will not be discussed further. Compared to the integral yield, the use of the peak height has the following advantage: whereas the yield always samples a depth interval corresponding to FR (+the energy resolution), the height allows in principle to probe a smaller depth interval (about the energy resolution along the inward path). The previous advantage is obstructed by the finite detector resolution (and outward path contributions), which are of no concern for the yield. The probing accuracy decreases with increasing detector resolution, due to shifts and a decrease in height. A clear disadvantage is that the statistics are much lower, so that scatter is greater and a larger dose has to be collected . This larger scatter can induce errors in the assignment of the signal maximum. Also is it more difficult to extract the correct height and position than to extract the yield from the spectrum. For actual depth profiling, subtraction of the nonresonant background may be done, if one bears in mind the influence of it on the parameters . For the height, the subtraction would imply the subtraction of a constant from the determined value, for the integral this would yield subtraction of a nearly constant value and for the FWHM this certainly would give a much better measure of the convolution of the resonance peak with the profile. Tracing the peak position as a function of energy would allow to unambiguously identify the resonant peak . However, the position can only be extracted with limited accuracy . The recorded spectrum should correspond as closely as possible to the true spectrum. Since the spectrum is discretisized into a series of channels and each channel corresponds to a certain energy

289

interval over which the true spectrum is summed, this implies a small channel width .1. The channel containing the resonance, will present a local maximum. The smallerthe channel width, the higher the accuracy with which the peak position can be identified . The uncertainty is in fact given by .1 . On the other hand, the smaller e, the lower the height per channel per beam dose . Thus statistical scatterwill be larger and obscure the position of the maximum. Fof given statistical scatter, the required dose will be higher for smaller channel widths. A compromise between dose and accuracy is necessary. 4. Experimental results The experiments are carried out on a tandem type accelerator, NEC 5SDH-2 1.7 MV terminal . The 3.045 MeV '6 0(a, a)r6 0 resonance is applied for the profiling of oxygen in Si02 layers, with various thicknesses : 204 A (DEeq = 4.3 keV), 892 A (DEeq =19 keV) and 5000 A (AEeq = 109 keV). The thicknesses have 1+yea obtained by spectroscopic ellipsometry. The samples have negligible surface roughness. The energy was calibrated by employing the 160(a, a) î60 resonance itself, due to the absence of other possibilities. The energy was determined from the energy scans which were performed. Inspection afterwards has revealed that this might have induced a systematic offset of a few keV. However, this only shifts the energy scale, without distorting it. The calibration was reproducible within the accelerator setting uncertainty for independent runs. Two different setups were used for the experiments, both with a scattering geometry of 9, = 92 = 166°, solid angle ,fl y = 1.6 msr and 02 = 3.8 msr and effective acceptance angle d9, = 2.8° and d92 = 4.2°. The data acquisition presents the main difference: in setup 1 the MCA has a fixed division of 1024 channels with .1, = 1.9 keV/channel and in setup 2 the division is 512 channels with .12 = 3.66 keV/channel . In view of the remark made on the channel width k' in section 3 more accurate measurements of the energy position can be obtained with setup 1. The energy variation for the scan was carried out by manually varying the terminal setting (setup 1) or by application of the target biasing technique [12]. In fig. 5 the measured peak height and integral yield versus beam energy is shown for a 892 A oxide on Si (experiment with setup 1). The yield curve is peaked around 3050 keV and as such does not give a good indication of the constant profile . The signal height, however, exhibits a kind of plateau which is more in correspondence with the constant oxygen profile. The detector resolution, extracted from the spectra is found to be about 25 keV. This high value is mainly due to III. CONTRIBUTED PAPERS

290

W. De Coster et at / Depth profiling of light-Z elements with elastic resonances

5000

-----50000

4000

10000

-+-

°

40000

530

Height Position

i

9

8000

526

11

m 3000 u O

30000 

é

2000

20000

1000

O

10000

3030

3040

3050

3060

3070

0

Beam Energy (keV)

/

o

e 6000 Go Û 4000

528

524

/

C

522

o,oso ô

U

520

20001

518 o/

0

516 3030

3040

3050

3060

3070

3080

Fig . 5 . Measured peak height and peak yield versus beam energy for a 892 1 SiO2 film on Si with perpendicular beam incidence. In contrast to the yield curve, the height exhibits a kind of plateau in correspondence with the contrast oxygen concentration .

Beam Energy (keV) Fig. 7. Measured peak height and peak position versus beam energy for a 204 A Si0 2 film on Si with perpendicular beam incidence. The height curve is very peaked, a plateau is not reached due to the to high energy increment . The position is a straight line, a decrease in energy not being measurable .

degradation of the long-used detectors . The measured peak position is not shown: there was a large scatter on the data due to the fact that .1=4 .7 keV/channel, therefore the expected behaviour (fig . 36) was hardly discernable . Fig. 6 shows the measured peak height and position for the same 892 Â oxide, but now measured with setup 1 . The height curve is roughly the same, apart from the data scatter. The position clearly exhibits the expected behaviour, though also the scatter is relatively large . The discrepancy at 3055 keV is caused by a discontinuity in the manual energy scan,

which could also be evidenced in the shift of the Si-leading edge . The extracted detector resolution is 14.8 keV. The peak height and position versus beam energy of a thin 204 t1 oxide are given in fig. 7 (experiment with setup 1 .) No plateau is visible in the height. This is partly due to the influence of the detector resolution which is 14 .8 keV . The spectrum for such a thin oxide was always nearly Gaussian without any distinct features . Apart from this, the expected plateau width is so narrow (4 keV) that it would not show up with the used

7000

° Height °°° ° Position~o / ° °

60001

4000 3000, 20001

°

0 0°0

50001

c

8 °

°a

0

1100

o

1080 8\

1060

8V

0ô

No

1000 0

Fig. 6. Measured peak height and peak position versus beam energy for a 892 A Si02 film on Si with perpendicular beam incidence. The position follows globally the expected be haviour, the large scatter is due to inaccurate position extraction limited by the finite energy scale 8'.

d w

1040

1 1020 3020 3030 3040 3050 3060 3070 3080

Beam Energy (keV)

Fig . 8 . Measured peak height and peak position versus beam energy for a 5000 A Si02 film on Si with perpendicular beam incidence over a limited energy range . The peak position follows the expected behaviour. The variations in height around ER are due to normalization errors.

W. De Coster et al. / Depth profiling of light-Z elements with elastic resonances energy increment of 3 keV . The position is almost a straight line with a plateau around 3.045 MeV . The moving towards lower energy of the position is not discernable . Finally, results of position and height for a thick 5000 A oxide are shown in fig. 8 . Due to the large thickness, only a partial scan over 80 keV is performed. The results show the expected behaviour of peak height and position . The experiment is carried out in chamber 2 with the target biasing technique . The variations in height around ER are due to a systematic error in the normalization of the spectra, necessary with the target biasing technique.

291

larger interval can be probed with higher accuracy and depth resolution . However, energy spread contributions will also increase for the larger path lengths involved. Also the surface roughness will become the limiting factor for the application of tilting. Further theoretical and experimental investigation of this is necessary . Acknowledgement W. De Coster is indebted to the Belgian IWONL for financial support.

5. Conclusion

References

In depth profiling with elastic resonances, such as the '60(a, af 6 0 resonance, different parameters can be extracted form the recorded spectrum at each energy. The variation of the peak height, position, yield and FWHM with incident beam energy is discussed for normal incidence. The influence of energy spread in the sample and of the detector resolution on the parameters is shown. Especially the detector resolution has a major influence on the shape of the backscattering signal . A depth profiling methodology using the peak height as a probe has been proposed. Advantages and restrictions when compared with the use of the yield are discussed. It is clear that the detector resolution sets a practical limit to this application. With better detector resolution more accurate results can be obtained . The former has been experimentally illustrated for the depth profiling of Si0 z layers of different thickness . However, better results can be obtained when the experiments are performed with higher resolution than in these preliminary experiments. Tilting the target might yield an increase in accuracy since the profile will be stretched and will correspond to a greater energy interval. In principle this

[1] W.K . Chu, J .W . Mayer and M.A . Nicolet, Backscattering Spectrometry (Academic Press, New York, 1978) . [2] J.R. Cameron, Phys . Rev. 90 (1953) 839. [3] B.K. Patnaik, C .V. Barros Leite, G .B. Baptista, E.A . Schweikert, D .L. Cocke, L. Quinones and N. Magnussen, Nucl . Instr. and Meth . B35 (1988) 159. [4] J . Li, L .J. Matienzo, P . Revesz, Gy. Vizkelethy, S.Q. Wang, J .J . Kaufman and J.W. Mayer, Nucl. Instr. and Meth. B46 (1990) 287. [5] J .A . Martin, M . Nastasi, J.R. Tesmer and C.J. Maggiore, Appl . Phys. Lett. 52 (1988) 2177. [6] M . Bozoian, K.M. Hubbard and M . Nastasi, Nucl . Instr. and Meth. B51 (1990) 311 . [7] J.R. Bird, R .A. Brown, D .D . Cohen and J .S. Williams, in : Ion Beam Material Analysis, eds. J.R . Bird and J.S . Williams (Academic Press Australia, 1989) p . 618. [8] J .S . Williams and W . Möller, Nucl . Instr. and Meth. 157 (1978) 213 . [9] L .R. Doolittle, Nucl . Instr. and Meth. B9 (1985) 344 . [10] B. Blanpain, P Revesz, L .R. Doolittle, K.H. Purser and J .W . Mayer, Nucl . Instr. and Meth. B34 (1988) 459. [11] P . Berning and R .E. Benenson, Nucl . Instr. and Meth. B36 (1989) 335. [12] W. De Coster, B . Brijs, J. Goemans and W . Vandervorst, Proc. 10th Int. Conf. on Ion Beam Analysis, Eindhoven, The Netherlands, 1-5 July 1991, Nucl . Instr. and Meth. B64 (1992).

III . CONTRIBUTED PAPERS