Round Robin: measurement of H implantation distributions in Si by elastic recoil detection

Transcription

1 Nuclear Instruments and Methods in Physics Research B 222 (2004) Round Robin: measurement of H implantation distributions in Si by elastic recoil detection G. Boudreault a,1, R.G. Elliman b,r.gr otzschel c, S.C. Gujrathi d, C. Jeynes a, *, W.N. Lennard e, E. Rauhala f, T. Sajavaara f,2, H. Timmers b,g,3, Y.Q. Wang h,4, T.D.M. Weijers b,g a Ion Beam Centre, The Nodus Laboratory, University of Surrey, Guildford, Surrey GU2 7XH, UK b Department of Electronic Materials Engineering, Research School of Physical Sciences and Engineering, Australian National University, Canberra, ACT 0200, Australia c Institute of Ion Beam Physics and Materials Research, Forschungszentrum Rossendorf, P.O. Box , D Dresden, Germany d Departement de Physique, Universite de Montreal, C.P. 6128, succ. centre-ville, Montreal, Que., Canada H3C 3J7 e Department of Physics and Astronomy, University of Western Ontario, 1151 Richmond Street, Suite 2, London, Ont., Canada N6A 4B8 f Accelerator Laboratory, Department of Physical Sciences, University of Helsinki, P.O. Box 43, 00014, Finland g Department of Nuclear Physics, Research School of Physical Sciences and Engineering, Australian National University, Canberra, ACT 0200, Australia h Ion Beam Analysis Laboratory, IT Characterization Facility, University of Minnesota, 100 Union Street, SE, Minneapolis, MN 55455, USA Received 26 August 2003; received in revised form 17 February 2004 Abstract A 200 mm amorphised Si wafer was implanted with 6-keV H þ ions at a nominal fluence of atoms/cm 2. The uniformity of the implant was better than 2% over the wafer. Samples of the wafer were analysed for absolute H fluence by nuclear reaction analysis and elastic recoil detection (ERD) analysis, including both helium and heavy ion beams, using various types of detector (Si with range foil, time of flight ERD, and a position-sensitive gas ionisation DE E detector), various ion beams (He, Cl, Cu, I, Au) and independent analytical procedures. The results are compared and the inter-lab reproducibility is evaluated. The surface H, unstable under heavy ion beams, was resolved and accounted for throughout the analysis. Estimates of total combined uncertainties are about 6% for all participants, but the interlab reproducibility of the measurements was found to be 2.2%. Correct quantification of the H data from the gas ionisation detector is demonstrated. The uncertainty budget is discussed in detail. Ó 2004 Elsevier B.V. All rights reserved. * Corresponding author. Tel.: ; fax: address: c.jeynes@surrey.ac.uk (C. Jeynes). 1 Present address: Ion Beam Facility, Faculty of Sciences, Division of Physics and Astronomy, Vrije Universiteit, De Boelelaan 1081, 1081 HV Amsterdam, The Netherlands. 2 Present address: Physics Department, Instituut voor Kern- en Stralingsfysica, Katholieke Universiteit Leuven, Celestijnenlaan 200 D, B-3001 Leuven, Belgium. 3 Present address: School of Physics, University of New South Wales at the Australian Defence Force Academy, Canberra, ACT 2600, Australia. 4 Present address: Materials Science and Technology Division, Los Alamos National Laboratory, Los Alamos, NM X/$ - see front matter Ó 2004 Elsevier B.V. All rights reserved. doi: /j.nimb

2 548 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) PACS: )a; w; )s; Tt Keywords: ERDA; NRA; Hydrogen; Silicon; Implant standards 1. Introduction Hydrogen is an essential element in a wide variety of important thin film materials, from the hydrogenated amorphous silicon used in photoelectrics to polymers of interest to the coatings industry. However, analysis for hydrogen is troublesome. X-ray analysis on electron microscopes is sensitive to most of the Periodic Table, and thinwindow detectors can now analyse at least down to B, but H appears to be permanently out of reach. In principle, SIMS (secondary ion mass spectrometry) can analyse for the whole Periodic Table, but quantitative analysis for H is often difficult. Ion beam analysis (IBA) techniques are therefore especially valuable since the quantification of H with IBA is straightforward and sensitive. Both forward recoil and nuclear reaction analyses work well. Nuclear reaction analysis (NRA) can be very sensitive, has an excellent depth resolution, and is of course isotope specific [1,2]. But the NRA depth profile is obtained by scanning the incident beam energy; this is always relatively time consuming. The forward (elastic) recoil spectrometry methods (referred to as ERD, elastic recoil detection 5 ) [3], on the other hand, are able to analyse recoils of all the elements in the sample at all accessible depths simultaneously with a heavy ion incident beam (HI-ERD). On the other hand, with a He beam suitable for RBS, ERD spectra (thus: He-ERD) can be collected simultaneously with RBS spectra, and again all elements of the sample can be analysed. ERD is therefore a depth-profiling method of more general applicability than NRA. The kinematics and geometry involved in ERD sometimes mean that multiple and plural scattering affect the depth resolution and produce significant 5 ERDA (elastic recoil detection analysis) is also a widely used acronym. Both ERD and ERDA are recommended by Amsel in CUTBA: Cleaning Up the Tower of Babel of Acronyms in IBA [48]. low-energy background (and limit the sensitivity of hydrogen detection) [4 6]. Because of the variety of IBA methods that can be used and because of the importance of H profiling, we believe that it is timely to present an inter-laboratory comparison of a suitable sample. It is surprising that, even though these methods are now very well established in the literature and despite the importance of hydrogen, no such interlab comparison has been carried out since the work of Ziegler et al. in 1978 [7]. 6 Also, no attempt has yet been made to demonstrate the absolute accuracy of ERD measurements traceable to international standards such as has been done for RBS (Rutherford backscattering spectrometry) [8 10]. This is much more difficult for ERD with its more complicated analytical setup than for RBS, especially since ERD is often used in regimes where the interactions are not Rutherford (that is, the cross-sections are not given by the Coulomb potential). However, following the success of the present Round Robin, we hope that the IBA community will soon demand a hydrogen standard certified by national standards labs to match the new Sb implanted RBS standard recently certified (with an uncertainty of 0.6%!) 7 [9]. Such a certified standard sample is quite likely to be a hydrogen implant distribution in silicon such as we have used in this work. This is because hydrogen implanted in silicon has been demonstrated to be very stable since the H is firmly trapped at implantation damage sites [11]. Of course, silicon itself is now easily available with extremely high purity and at extraordinary reproducibility [12], and is therefore an ideal material for standards. The implantation process is funda- 6 A paper by Banks et al. has also recently been presented at the IBA-16 conference in Albuquerque (July 2003) [49]. 7 This certified standard is known as: IRMM-302/BAM- L001, where IRMM is the Institute for Reference Materials and Measurements, Geel; and BAM is the Bundesanstalt f ur Materialforschung, Berlin.

3 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) mental to semiconductor processing and is also thoroughly understood and widely available [13]. The only difficulty with using Si as an IBA standard is its crystallinity, which results in very strong channelling effects with the well-collimated beams used in IBA; this is overcome merely by amorphising the material by implantation. Alternative methods of amorphous-silicon (a-si) production, such as sputter deposition [14] or CVD (chemical vapour deposition) [15,16], are less attractive due to their high intrinsic H content. Marwick has reviewed IBA methods of H analysis in semiconductors [17]. In this work we use Kapton (polyimide) foils as hydrogen standards; these have a well defined chemistry, are readily available, and are remarkably resistant to ion beam damage [18]. Polyaniline [19], hydrogenated tantalum [20] and titanium hydride [21] have also been proposed. We report the analysis by seven laboratories of samples from a single amorphised and H-implanted silicon wafer. Both helium and heavy ion beams are used, with various detection systems including a novel gas ionisation detector. It turns out to be non-trivial to obtain H profiles simultaneously with profiles of heavy elements with this detector, indeed the initial reason for this Round Robin was to validate it. We determine both bulk and surface hydrogen, and estimate the instability of the surface hydrogen under ion irradiation. We will discuss the results and evaluate the inter-lab reproducibility. The estimated uncertainties will be presented as recommended in GUM [22], together with the uncertainty budget [23]. 2. Procedures and results Table 1 summarises the analytical procedures for all participants. Table 2 summarises the uncertainty budget in a unified way for all the data and Table 3 summarises the results. In Table 3 the Type A uncertainties [22] are entered where they are available. This section will amplify and explain these tables. The data analysis is only indicated in this section: it is analysed in detail in the next section. The uncertainty of the present analysis is dominated by uncertainties in stopping powers. The stopping power tabulations ( TRIM85 ) of Ziegler et al. [24] were used, together with updates: TRIM95, SRIM96 and SRIM2000. The updates do not have full documentation. These tabulations have an unknown uncertainty in general: we optimistically assume an uncertainty of 5% (although it could be much larger: the tabulations have an error larger than 10% for He in Si [25]). Recent measurements of the stopping of He in Si and C by Konac et al. ( KKKNS ) [26] have been confirmed with a 2% uncertainty [27,28], parameterised by Barradas et al. [29] (2% higher than KKKNS) and validated against the new IRMM/BAM Sb implanted standard [9] by Boudreault et al. [10]. The natural unit of thickness for IBA work is areal density: that is, atoms/cm 2, but for convenience thicknesses are usually given in nm, using for conversion an atom density of atoms cm 3 for Si Sample preparation The samples shared among the participants for absolute H fluence measurements were from a Si wafer amorphised to a depth of 200 nm and implanted by Axcelis Technologies Inc. with 6 kev H þ ions at a nominal fluence of atoms/ cm 2. There was no compensation for beam neutralisation, so the actual fluence may exceed the nominal by as much as 15%. 2% uniformity of the implant over the wafer has been determined directly by a set of comparative ERD measurements at London Australian National University (ANU): HI- ERD and DE E detector Measurements carried out at the ANU employed a heavy ion beam for analysis, and a position-sensitive gas ionisation DE E detector [30 32]. The pressure inside the gas ionisation detector can either be optimised for hydrogen depth-profiling or for that of heavier elements [33]. When optimised for hydrogen depth-profiling, so that protons are stopped in the sensitive volume of the detector, the information obtained for heavy ion recoils is only very limited. Since the strength

4 Table 1 Analytical details for each participant in the ERD Round Robin Participant Method Beam Angles Solid angle Incident [ ] Recoil [ ] [msr] Australian National University Surrey Detector HI-ERD 200 MeV 197 Au 16þ Position-sensitive DE E gasionisation detector + Mylar window Conventional 1.5 MeV 4 He þ 13.3; Si detector + 6-lm ERD Mylar range foil Standard a Si (SRIM2000) Kapton (TRIM85); Si (KKKNS b ) London Conventional ERD 1.6 MeV 4 He þ Si detector + 6-lm Mylar range foil Kapton, Mylar (TRIM95, except C); Si and C (KKKNS); Faraday cup Helsinki ToF-ERD 53 MeV 127 I 10þ ToF E detector Si (SRIM96) Rossendorf HI-ERD 35 MeV 35 Cl 7þ Si detector + 18-lm Al range foil D-implanted reference target NRA MeV 15 N Forward direction 4 00 NaI(Tl) scintillator Kapton (SRIM2000) For Si, O, C: ToF E Montreal ToF-ERD 40 MeV 63 Cu 8þ ; Si (TRIM95) 30 MeV 35 Cl 5þ detector; for H: Si detector + 13-lm Mylar/17-lm Al range foils NRA MeV 15 N Forward direction BGO scintillator Minnesota Conventional ERD 1.3 MeV 4 He þ Si detector + 5-lm Mylar range foil a Stopping powers used are specified in parentheses. b Barradas et al. [29] stopping powers actually used are 2% higher than KKKNS [25]. 1-nm Pt-coated Kapton (SRIM2000); Si (KKKNS) 550 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004)

5 Table 2 Uncertainty budget for all participants with coverage k ¼ 1 (The estimated uncertainty for each parameter is indicated. Where the uncertainty for a parameter has been estimated statistically (Type A uncertainty) this is indicated with an asterisk. Stopping powers used are given in Table 1 with uncertainty of 2% for KKKNS, otherwise 5%) H content Q (%) X d (%) cos h ½eŠ QX d (%) cos h ½eŠ [e] (%) Kapton or cos h (%) (%) Instability (%) (%) [e] (%) Counting statistics (%) Surface: bulk ratio (%) H loss from sample (%) Reproducibility (%) ANU Large Surrey London Helsinki Rossendorf (NRA) ? Rossendorf (ERD) Large Montreal Minnesota ? ? 5.9 cos h ½eŠ or cos h: uncertainty due to 0.1 uncertainty in the geometry. H loss, Kapton instability and surface/bulk ratio: uncertainty is taken as 10% of the estimated correction. Reproducibility is taken from repeated measurements of bulk H content (total H for ANU).? indicates no data. Combined standard uncertainty (%) G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004)

6 552 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Table 3 Global summary of the results from all the participants together with the estimated uncertainties Participant H content [ atoms/cm 2 ] Combined standard Surface Implant Total uncertainty (u c ) for implanted H determination [%] ANU 63.5(10) 5.2 Surrey 14.0(13) 57.6(10) London (10) Helsinki Rossendorf (ERD) 66(3) 3.2 Rossendorf (NRA) Montreal 11.4(30) 55.5(30) 66.9(10) 6.0 Minnesota Mean Standard deviation 4.8 {47%} 1.25 {2.2%} Numbers in { } represent the uncertainty (standard deviation) of the measurements in %. Numbers in ( ) are the reproducibility of the last figure where available. of this particular detection system is the simultaneous quantification of all elements present in a sample, for the purpose of the present study, the gas pressure was optimised for the detection of heavy recoil ions. In this mode the heavy ions are stopped in the sensitive volume of the detector, whereas the high energy protons from the sample are transmitted, with only energy-loss information being obtained. While the hydrogen content of the sample is thus quantified simultaneously with the heavy elements, the achievable depth resolution for hydrogen is so much reduced that the surface hydrogen is not resolved. However, one purpose of this work was to demonstrate that the gas ionisation detector could be quantified correctly, since this is not trivial, and we will show that the total H content measured this way is very similar to the other HI- ERD measurements. The position-sensitive gas ionisation detector used has a subdivided anode with two DE-electrodes and a residual energy electrode (E res ). The pulse amplitudes from these electrodes can be combined to obtain the total ion energy. The detector also features a grid electrode between Frisch grid and anode, which provides an independent energy signal [31,32]. This signal (E g ), amplified with high gain, also provides the trigger for the electronic acquisition of events, including proton events. Two separate samples were analysed. The energy of the incident 197 Au 16þ beam was ( ± 0.10) MeV. It was collimated in front of the sample using four slits, which were 200 mm apart. The first and third slits had the nominal dimensions of 0.5 mm 3 mm and the second and fourth had the dimensions of 1 mm 4 mm, respectively. The angle between sample normal and beam was The detector was located within the plane defined by beam and sample normal at a scattering angle of 45.9, 278 mm from the sample. Recoil ions entered the detector through a (0.50 ± 0.03) lm thick Mylar window supported by a rectilinear grid of gold-coated tungsten wires. The transmission of the grid is better than 97% and has been considered in the determination of the detection solid angle which was (3.50 ± 0.05) msr. Propane gas was passed through the detector at a constant pressure of 80 mbar. As illustrated in Fig. 1, all detected recoil ions heavier than protons are evident in a two-dimensional projection of the DE versus E g histogram. Apart from the Si recoil ions, O and C events associated with the sample surface are apparent. As part of the sorting process, position information for each event was used to correct the kinematic energy spread over the acceptance angle of 3.9, thus retaining the energy resolution of the detector [31,32]. The horizontal position of the

7 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Fig. 1. Two-dimensional projection of the spectrum of DE versus energy (E g ) (for sample #1), after correction of kinematic energy broadening across the acceptance angle. The relative ion yields are indicated by the grey-scale (z-axis). In addition to Si ions, C and O ions from the sample surface can be identified. The low intensity haze of events above Si corresponds to electronic pile-up (ANU). ions on entry into the detector was obtained from the relative response of two sawtooth subdivisions of the second DE-electrode on the anode. For both samples energy spectra for C, O and Si were extracted from the two-dimensional projection. The energy scale of the spectrum was calibrated by associating the position of the half-maximum of the high-energy edge for Si and the centroids of the surface peaks for O and C with surface scattering. Energy loss in the detector window and the pulse height deficit of the detector signal [34] were taken into account. This gave an energy interval of 85.1 kev per digital channel and an offset of 1.5 MeV. The hydrogen events were identified in the extreme low-energy part of the two-dimensional projection of the E res -versus-e g histogram where the protons are completely separated from other low-energy ions [33]. This low-energy part of the projection is shown in Fig. 2. The E res -versus-e g spectra were calibrated by modelling the detector response using tabulated (SRIM2000) stopping powers for protons in propane. Since higher-energy protons pass through the sensitive volume of the detector, only a fraction of their energy is detected. The E res E g projection is therefore complex, containing contributions from both stopped and transmitted protons with the response curve bending over at the point of maximum energy loss under both electrodes. The indicated region corresponds to hydrogen recoils from near the sample surface and shows a large yield of detected protons. This region was therefore taken to represent the implanted hydrogen and integrated. Apart from this intense signal a smaller number of counts along the response curve indicate the presence of a uniform trace concentration of hydrogen throughout the wafer. The charge solid angle product (see below, Eq. (1)) is determined from the amorphous silicon yield, Fig. 2. The extreme low-energy part of the two-dimensional projection of the spectrum which relates the E res signal with that from the grid electrode and allows the identification of protons. The implanted hydrogen can be identified as intense yield in the near-surface region. The response curve for hydrogen is indicated in the direction of increasing sample depth (ANU).

8 554 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Fig. 3. The E res spectrum of the detected protons for sample #1, in comparison with the response function of this (residual energy) electrode for high-energy protons. The inset shows the reduction of the hydrogen content with increasing fluence (ANU). with SRIM2000 stopping powers, and the total H is given by the total number of counts (Eq. (2)). The total H content determined for the two samples was and atoms/cm 2, respectively, assuming scattering cross-sections for surface scattering. The total fluence of Au ions incident on the material was ions/cm 2.In order to establish if any hydrogen was desorbed during the analysis, the hydrogen yield was integrated for sequential, equal fluence intervals. The results, displayed in the inset of Fig. 3, are consistent for both samples and show that the hydrogen concentration decreases by 5% during the measurement. Fitting and extrapolation to zero fluence increases the measured concentration for both samples, then the corrected H content for the two samples was and atoms/cm 2, respectively. The ratio of the two measured concentrations is not affected by the systematic uncertainty of n Si, but only limited by counting statistics. This implies a non-uniformity of the hydrogen implant across the wafer of the order of (3.1 ± 1.5)% which agrees within uncertainties with the expectation that the implant non-uniformity is less than 2%. The best estimate is the average of both measurements, which is (61.5 ± 1) atoms/cm 2 where the estimated uncertainty just represents the measurement reproducibility. Using the same procedure as in the analysis of the hydrogen content, the surface concentrations of carbon and oxygen are and atoms/cm 2, respectively. In Fig. 3, the E res spectrum measured for protons from one of the samples is compared with the response of the residual energy electrode to protons recoiling from the surface region of a uniform standard [33]. The response function was obtained by recording a hydrogen spectrum for a Kapton standard. It is apparent that the centroid of the hydrogen distribution is inside the sample (E res ffi 0:44 MeV) implying that the concentration of any surface hydrogen present is small compared to the implant concentration. The energy difference of 30 kev between the E res signal of surface scattering and the centroid of the hydrogen spectrum is equivalent to a difference in recoil energy of DE ffi 70 kev. For the experimental geometry used, this corresponds to a thickness interval of (670 ± 70) atoms/cm 2 (134 nm), similar to the projected range of 6-keV protons in silicon which is atoms/cm 2 (106 nm, using SRIM2000). Correcting the scattering cross-sections for this depth (106 nm) a hydrogen concentration of atoms/cm 2 is obtained Surrey: conventional ERD The Surrey ERD analysis of the hydrogenatedsilicon implants was carried out using a MeV 4 He probe beam of about 30 na and nominally 1 mm diameter in a conventional setup. The

9 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) accelerator energy was calibrated using three points ( 27 Al(p,c) 28 Si at and kev, and 19 F(p,ac) 16 O at kev); the uncertainty in the energy was about 0.3%. Measurements were made at 13.3 and 15.0 incident angles. A precision 6 movement goniometer [35] was used. Recoil measurements were made using a 3 29 mm 2 (2.19 msr) Hamamatsu photodiode mounted vertically, with a recoil angle of 26.6 and a atoms/cm 2 (6 lm) Mylar (C 10 H 8 O 4 ) range foil. RBS measurements were also simultaneously carried out with two ion-implanted Si detectors, nominally 50 mm 2 (1.334 msr) and 25 mm 2 (0.737 msr) and scattering angles of and 118.8, respectively. The detector angle was determined using the goniometer to find the deflection angles required to move a laser beam from normal to the detector. The normal position could be determined to about one step (0.005 ). The electronics calibration for the RBS detectors was performed with a Au/Ni/SiO 2 /Si sample [36]. The ERD detector was calibrated using two beam energies (1506 and 1405 kev) and interpreting the energy shift in the spectrum assuming TRIM85 stopping powers for the energy lost in the range foil. The RBS detector charge solid angle product was determined from the silicon sample using Barradas et al. stopping powers [29] (2% higher than KKKNS [26]). To verify this the Bi content of a Harwell series standard [37] was measured from the Si yield. This was for three beam incidence angles (90, 13.3 and 15.0 ). A value of 4.51(11) Bi/cm 2 was found, which compares well with the certified value (4.72(10)) and also with the value determined by reference to the IRMM-302/BAM-L001 Sb certified standard [9] by Boudreault et al. (4.64(7)) [10]. A Kapton sample, mm thick CR grade from Goodfellow Cambridge Ltd., and whose film composition was assumed to be a mixture of PI (polyimide: C 22 H 10 N 2 O 5 ) and alumina, was used to calibrate the ERD solid angle. The PI/alumina ratio was determined by RBS using four different areas of the sample and the same two different glancing incidence angles, assuming the C/H ratio is 2.2. Note that the collected charge was determined by the spectral height of the RBS and agreed with the measurements on Si to 4 ± 2%. The alumina content was found to be 8.8(5) at%, using all 16 RBS spectra (4 areas, 2 angles, 2 detectors) together with the IBA spectral fitting code Data- Furnace [38,39]. The composition of the Kapton (assuming C/H ¼ 2.2) was found to be (C, H, N, O, Al) ¼ (51.4, 23.4, 4.7, 17.0, 3.5). Then the H signal from Kapton was used to determine the ERD detector solid angle ratio with the RBS detector. From the 8 sets of spectra (4 areas, 2 beam incident angles), a value of 2.19(14) msr was obtained as a solid angle for the ERD detector. From the Kapton spectra a fluence effect has been observed. The first and second measurements (2 beam incident angles) of the 4 pairs give 2.26(15) and 2.13(11) msr solid angle values, respectively. As H is lost with increasing bombardment the number of counts for a given incident charge falls. This is the same effect as reducing the solid angle for a fixed H content. Thus the effect observed is as expected. Hence we estimate an ERD solid angle value corrected for H loss in the Kapton as being 2.39(20) msr. Seven simultaneous RBS and ERD spectra from the H implant were obtained (see Table 4). One typical measurement is shown in Fig. 4. The fit as obtained using DataFurnace is also shown in this figure. As can be observed, the RBS signal is not fitted well; this is because the charge is fixed at an approximate value for the fit. The correct charge value is determined from the Si signal. The fitted structure obtained for the H implant is illustrated in Fig. 5. All the seven spectra gave a similar result. It is interesting that for the seven independent fits the Simulated Annealing algorithm (DataFurnace) found the surface H peak; this is quite unambiguous for this data. The mode of the implanted depth profile is at 90 ± 2 nm depth. Combining the seven measurements and using the corrected solid angle of 2.39 msr, values of (57.8 ± 1.0) /cm 2 and (13.9 ± 1.3) /cm 2 are found for the implant and the surface peaks, respectively, which gives a total H amount of (71.7 ± 2.2) /cm 2. Note that the multiple scattering tail was ignored in this analysis. The measurements are listed in chronological order.

10 556 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Table 4 Measurements of total H content (bulk + surface) in the Si:H sample (Surrey) Measurement H content [ atoms/cm 2 ] Charge [lc] Area Incident angle [ ] Bulk Surface Total M a 15 M b 15 M c 15 M c 13 M d 13 M d 15 M c 15 Mean Standard deviation (%) measurements from the same irradiated area on the sample. A charge between 8 and 9 lc was collected for each measurement. No significant bulk H loss due to beam irradiation could be detected London: conventional ERD Fig. 4. Typical simultaneous RBS/ERDA spectra from the H-implanted Si sample, with fitted spectra (Surrey). Fig. 5. Hydrogen fitted profile (Surrey). Bulk H loss by beam damage can be investigated by comparing the results from the same series of Two pieces of the hydrogenated-silicon wafer were measured using a 1.6-MeV 4 He beam in a conventional ERD setup. Both Kapton and Mylar targets were used as standards to determine the detector solid angle. These two agreed at about 2.2%. Stopping powers used were TRIM95, except for C and Si where KKKNS was used. Charge collection was monitored by means of an intermittent Faraday cup that intercepts the beam in front of the target with a duty cycle (beam-on fraction) of approximately 63%. A second Faraday cup was put downstream to calibrate the beam monitor, which has been shown to be accurate at better than 1% [28]. The measurements from the two samples gave and atoms/cm 2 as amount of implanted hydrogen, which results in 59 ± atoms/cm 2 assuming uniformity over the wafer. No apparent hydrogen loss with incident fluence has been observed. The surface hydrogen peak was resolved, as can be seen in Fig. 6. A H content of and atoms/cm 2 were found for samples #1 and #2, respectively, giving an average of atoms/cm 2. This is slightly higher than

11 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Fig. 6. Hydrogen ERDA spectrum obtained from sample #1 (London). the surface areal density of H/cm 2 generally found for any clean sample (Au, GaAs, Si, etc.) using the system at London. Note that some efforts were made to etch (using HF) the surface hydrogen off the implanted samples without success; it seems that the surface hydrogen is rather tightly bonded Helsinki: ToF-ERD A ToF-ERD instrument with two timing gates and an energy detector has been used in the analysis of over 500 samples each year and is described elsewhere [40]. A beam of 53-MeV 127 I 10þ ions was used and the angle between sample normal and beam was 70. The detector telescope was located at an angle of 40.0 ± 0.1 with respect to the incoming beam and it consisted of two identical carbon foil timing gates of thickness 5 lg/cm 2, and having diameters of 12 and 18 mm and distances from the sample 488 and 1172 mm, respectively. An ion-implanted detector was situated 71 mm after the last timing detector. The solid angle of the telescope is 0.19 msr, and the effective solid angle with accelerating and deflecting grids in the timing gates taken into account is 0.08 msr. The timing signal from the first timing gate was delayed about 500 ns and a time-window of 500 ns was used in the time-toamplitude converter. The beam spot size was 10 mm 2, beam current of 0.2 particle-na, and data were collected event by event (list mode) with a collection time of 4600 s. The raw E ToF histogram is plotted in Fig. 7(a) with a logarithmic z-scale and every event visible. The detection efficiency of the timing detector for hydrogen is dependent on the stopping power and therefore also the energy, and it is quite low (around 15%) for recoiled hydrogen atoms. However, the timing gate detection efficiency for heavier recoil atoms is very close to 100%, and because of these two things the events seen only at the energy detector result almost entirely from H recoils. It must be noted, that long time windows (500 ns or more) are required to avoid background Fig. 7. (a) 3D ToF histogram which represents raw measurement data. (b) Coincident events with the E detector (coincident TOF and E) and the events observed only by the E detector (non-coincident E) are drawn with dotted and dashed lines, respectively. The sum of the two events is plotted with solid line. No background subtraction was performed for plotted data (Helsinki).

12 558 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) from low energy heavy recoils which can miss the time window of the ToF detector and therefore generate non-coincident events to the energy detector. The final energy spectrum for hydrogen is obtained by adding the coincident events seen both in ToF and energy detectors and the noncoincident events collected only from the energy detector. This is illustrated in Fig. 7(b). The small background can be subtracted from the final results. Due to this analysis procedure for hydrogen the energy spectrum is obtained from the energy detector, whereas for other elements the energy spectrum calculated from the timing signal gives a better resolution and a linear calibration. By means of this approach for hydrogen analysis a 100% detection efficiency is achieved and hydrogen analysed accurately among all other elements in the Periodic Table using only one measurement. With the ¼ 40 geometry used, the surface peak of the hydrogen (located at energy channel 210) could not be separated from the implantation distribution. It should be noted that this is not only due to the detector energy resolution but mostly due to multiple scattering. The separation could be achieved by tilting the sample more. The surface peaks of C and O did not have an effect on the results. The depth profiles were calculated using Rutherford scattering cross-sections and SRIM1996 stopping powers. As the measurement was made in a list mode, the hydrogen profile could be monitored continuously. The hydrogen energy spectrum of the sample is plotted in Fig. 8(a) with 6 times magnified H surface peak from an unimplanted Si sample. The total hydrogen yield is plotted as a function of the ion fluence in Fig. 8(b) and diminished by about 10% throughout the analysis. It was not possible to resolve whether the hydrogen loss occurred in the surface or in the implanted area. Fig. 9 shows the H profile as a function of time in the first part of the analysis. In addition to hydrogen loss, the peak profile changed slightly over the measurement. At first H seems to concentrate at the depth of highest concentration, but this peak disappeared as the measurement went on. The H surface peak is not resolved from the implant, but is estimated by fitting a smoothed surface signal to the total spectrum (see Fig. 8(a)). Since the surface is about 8% of the total, and with an uncertainty of (say) 50% on the estimate of its area, the resulting uncertainty of the implant signal area is estimated at about 4%, see Table Rossendorf: HI-ERD + NRA Both NRA and conventional HI-ERD analyses were carried out at Rossendorf for the characterisation of the hydrogen content of the sample. Fig. 8. Total H energy spectrum from the implanted sample plotted with 6 times magnified surface hydrogen peak from another, hydrogen-free sample. Due to the peak broadening and irradiation losses, H surface peak cannot be separated from the implanted sample. A 10% hydrogen loss can be seen in (b) where the total hydrogen amount is plotted as a function of the irradiation fluence. The beam spot size was about 10 mm 2 and current 200particle-pA (Helsinki).

13 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Fig. 9. 3% of the event list data, with the first time interval shown as a spectrum (Helsinki). The beam fluence measurement was done indirectly by backscattering from a rotating sector beam shutter (Au on Al) regularly calibrated against a Faraday cup. The charge collection uncertainty is estimated as 2%. The HI-ERD was performed using a 35-MeV 35 Cl 7þ beam of 1 mm 2 spot size, together with a silicon detector covered with an 18 lm Al stopper foil, with a recoil angle of 38 and a solid angle of 2.1 msr, and led to a measurement of a total of H/cm 2, but the energy resolution was not sufficient to resolve the surface peak. A total fluence of ions struck the sample. The data were calibrated using a local reference target implanted with D, where the fluence was determined by charge collection in the implanter. The standard uncertainty of 5% in the ERD measurements has been estimated from repeated evaluations over time of standard samples. The NRA was carried out using a 15 N beam with energy increasing from MeV in steps of 15 kev for a total span of about 250 kev, with a beam spot size of 20 mm 2. The narrow and isolated resonance in the reaction: 1 H+ 15 N fi 4 He + 12 C+c, at MeV and producing MeV gamma rays was then used [41]. The energy loss was calculated from SRIM2000 database assuming pure silicon (a 10 at% H content has no significant effect on the peak integral). A 4 00 NaI(Tl) detector was used in the forward direction. The charge was again collected from backscattering from a rotating beam shutter. The system is described in more detail elsewhere [18]. The data were normalized by using a Kapton standard sample and the SRIM2000 database. The NRA analysis allowed one to resolve the hydrogen surface contamination. The values found are and atoms/cm 2 for the implanted hydrogen and the surface hydrogen, respectively, which gives a total amount of H/cm 2. The H depth profile obtained is presented in Fig. 10. To determine the beam related H release during the NRA measurement, the yield of detected hydrogen was measured as a function of incident ion fluence just below the peak at a depth of 100 nm at a fresh spot. The results are shown in Fig. 11. As can be observed, the fluence-dependent decrease of the yield is relatively low and it is not evident whether there is significant hydrogen release during the measurement or only a peak broadening with a constant integral Montreal: ToF-ERD The detector was a pair of silicon surface barrier detectors with identical solid angle of 0.18 msr and cooled to )10 C, one of which had a range foil (Mylar or Al) to detect H recoil energies, and the other of which provided the stop signal for the time of flight of all other recoils. The start signal is provided by a 10 lg/cm 2 carbon foil with a

14 560 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Fig. 10. Hydrogen depth profile obtained using NRA (Rossendorf). Fig. 11. NRA detected hydrogen yield as a function of increasing incident fluence at a depth of 100 nm: study of beam related to hydrogen release (Rossendorf). multichannel plate electron detector. The distance between the MCP and the detectors was 65 cm. More details of this setup are given elsewhere [42,43]. A series of 8 different measurements were made from 2 samples of the hydrogenated-silicon wafer by using 2 different beams (40-MeV 63 Cu 8þ and 30-MeV 35 Cl 5þ ) and 2 different absorbers (13 lm Mylar and 17 lm Al). The Cu ions had the current of 3 na and the spot size of 4 mm 2 while the Cl ions had the current of 5 na and the spot size of 2 mm 2. In each measurement a total fluence of about ions struck the target. The thickness of Mylar and Al absorbers was deduced by using a precision microbalance (weighing method) as well as MeV a particles from 241 Am source (energy-loss method); the results from the two methods agreed with each

15 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) other within 1.5%. The energy-loss parameters from TRIM95 were used. To estimate the surface H concentration relative to the implanted H, and also to determine the implanted peak depth, a separate (low resolution) NRA experiment using the 15 N resonance reaction was performed (Fig. 12). The beam energy in range 6.3 to 6.8 MeV was varied in 20 kev steps from carefully determined surface H peak position, and the resonant 4.43-MeV gamma rays were detected with a BGO detector placed at 2 cm from the target in the forward direction. Assuming the half width of the surface H peak as 15 kev, the counts in the surface and the implanted H peak areas were found to be in the ratio This number is expected to have large uncertainty (of 25%) because of low resolution, and also because the beam related surface H loss was not accounted for. Fig. 12 also shows that the implanted H peak is at a depth of DE ¼ kev from the surface. Using the depth scale as 1.54 kev/nm for the energy range MeV of 15 N in Si, the implanted H peak appears to be at the depth 84.4 ± 6.5 nm. In the first experiment with a Cl beam, the H spectrum showed some indication of a surface hydrogen peak that was poorly resolved from the implanted peak. However, this surface H disappeared rapidly within first 5 min of beam irradiation with a beam fluence less than During the same interval there was no apparent change in the implanted peak. Therefore in the subsequent experiments using different beam spot positions the target was irradiated for 7 min before data collection. In each measurement the total counts in the implanted H peak as a function of beam fluence was monitored and found to be constant within 5%. The results obtained from the series of eight different sets of data (2 samples, 2 different beams, 2 different absorbers) are displayed in Table 5 and Fig. 13. The mean implanted (surface corrected) Table 5 Summary of the ToF-ERD results (Montreal) Beam Absorber Sample Implanted H fluence ( atoms/ cm 2 ) 40-MeV 63 Cu 8þ 13 lm Mylar # MeV 63 Cu 8þ 13 lm Mylar # MeV 35 Cl 5þ 17 lm Al # MeV 35 Cl 5þ 17 lm Al # MeV 63 Cu 8þ 17 lm Al # MeV 63 Cu 8þ 17 lm Al # MeV 35 Cl 5þ 13 lm Mylar # MeV 35 Cl 5þ 13 lm Mylar # Mean 55.5 Standard 1.0 (1.7%) deviation Fig. 12. H profile by NRA (Montreal). Fig. 13. H profile from energy spectra (Montreal).

16 562 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) hydrogen fluence is found to be (55.5 ± 1.0) atoms/cm 2. Concentrations of surface contaminants (C and O) as well as the changes in implanted H concentration in target matrix had a negligible effect on the deduced implanted fluence. However, in the reported results the relative concentrations of all the detected elements were deduced by using an iterative depth profile technique accounting for the layer-by-layer changes in the matrix composition [44]. All the depth profiles were similar. The maximum of the implanted peak was at a mean depth of 84.1 ± 6.5 nm consistent with the NRA Minnesota: conventional ERD Range foil ERD is used with the following experimental conditions: 1.3 MeV 4 He þ incident beam; 15 incident angle and 30 forward recoil angle in IBM geometry; 5-lm Mylar stopping foil; 1-nm Pt-coated Kapton as H standard. The 1-nm Pt coating on Kapton was used to reduce the beam charging on Kapton and also to correct for the charge measurement uncertainty by monitoring the Pt-signal with a RBS detector at 165. By using a lower energy 4 He þ beam (1.3 MeV) and a thinner Mylar range foil (5 lm), a better separation between the surface and implanted hydrogen was achieved by ERD, as shown in Fig. 14. The first step in the analytical method was to use two-gaussian functions to fit the ERD spectrum and determine the peak areas for both H surface and H implant. Then the H content was determined via relative measurements using the surface spectrum height of the Kapton spectrum. Since the apparent depth of the maximum H- concentration for the implant seen by the incident beam is about 360 nm below the surface, it is important that the mean energy approximation [45] be used to correct the recoil cross-section differences between Kapton and the sample in the calculation. The H-stability correction in Kapton and charge measurement in both sample and Kapton were carefully done. The detail of the experiment and calculations has already been presented [46]. The results are atoms/cm 2 implanted hydrogen if SRIM2000 stopping powers are used, but atoms/cm 2 if KKKNS stopping powers are used. A standard deviation of around 1.5%, mainly due to the statistical fluctuations in determining the Kapton spectrum height, is associated with this measurement. The H implant is located at 94 nm with a FWHM of 78 nm. Fig. 14. Energy spectrum showing Gaussian deconvolution (Minnesota).

17 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) Uncertainty analysis In RBS or ERD for a thick film sample the surface yield Y c (in counts/channel) from a particular atom species with proportion N c in the sample is given by Y c ¼ QXN c rd=ðcos h½ešþ; ð1þ where QX is the charge solid angle product (in sr), r is the differential cross-section for the ion of given energy into the detector (in cm 2 /sr), d is the energy width of an ADC channel (kev/channel), h is the beam incidence angle from normal and [e] is the stopping cross-section factor (ev cm 2 ) for the given geometry and beam energy. In ERD, [e] is evaluated for the incident beam on the inward path and the recoil beam on the outward path. Note that in Eq. (1), d can be routinely measured at 0.5% uncertainty for RBS [36,47], but for range foil ERD the ADC calibration depends on the knowledge of stopping through the foil which could potentially be measured directly at an uncertainty approaching 1%, but is usually estimated from the nominal thickness and the stopping power tables. Because these contribute largely to the offset of the calibration and not the gain, the uncertainty is estimated provisionally as 2%. For a thin film the total yield Y is given by Y ¼ QXNr= cos h; ð2þ where r is the differential cross-section at the average energy of backscattering from the film. Strictly, Eq. (2) should be written as an integral over the depth of the film since the cross-section r is a strong function of beam energy. Data analysis codes all do this numerically, but the mean energy approximation is accurate for sufficiently thin films (including this H implant). Five different data analysis procedures were used, and the uncertainty budget for all the participants is summarised in Table 2. For all participants the uncertainty is dominated by the uncertainty in the stopping powers, which are discussed above. The NRA analysis (Rossendorf) relies on good charge measurement with the detector solid angle and the H depth profile both being determined via Eq. (1), the former through the Kapton standard with polymer stopping powers and the latter using Si stopping powers. In this case the d is the energy step for the profile and h ¼ 0. The top row in Table 2 refers to Eq. (1) for the Rossendorf analysis: for all other cases it refers to Eq. (2). The Montreal NRA analysis was not quantified. With ERD the H content is obtained from the H spectrum via Eq. (2). In the simplest case (Rossendorf) comparison was made directly with a D implanted standard via Eq. (2), where the comparative charge measurement relied on a calibrated chopper. In this case there is no traceability since one implanted standard is being compared to another. However, if a traceable standard existed the uncertainty would be dominated by the charge collection uncertainty since the fluence is given by a comparison. In the second ERD case (ANU, Helsinki, Montreal) the QX factor was obtained directly from the Si signal via Eq. (1). Only one measurement was required and therefore this factor does not appear in Table 2. The cross-section ratio r H =r Si is needed for both of these cases, but these cross-sections are Rutherford with the heavy ion beams used, and have an uncertainty of about 1/4% due to the uncertainty of the screening correction [8]. As discussed above, the uncertainty of the energy calibration d of the ADC is estimated as 0.5% where range foils are not used and 2% otherwise. In the third ERD case (London) the charge is monitored with a calibrated chopper, and the solid angle of the ERD detector is determined directly from the Kapton (and Mylar) standards via Eq. (1). The Si stopping powers are used only to get the depth scale of the H profile, and have only a second order effect on the H content. The uncertainty in QX given in Table 2 is due to the uncertainty in Q since it is only the X of the ERD detector that must be determined. In the last ERD case (Surrey, Minnesota) QX is determined for the RBS detector via Eq. (1) from the Si stopping, the uncertainty of which dominates the uncertainty in QX given in Table 2. The solid angle ratio of the RBS and ERD detectors is determined from the Kapton standard also via Eq. (1). The geometrical uncertainty for the glancing geometries used in ERD is significant: 0.1 angular

18 564 G. Boudreault et al. / Nucl. Instr. and Meth. in Phys. Res. B 222 (2004) uncertainty at 15 incident and exit angles give an uncertainty in cos h ½eŠ of 2.2%. Geometrical uncertainty can arise not only from measurement of the scattering angle but also from uncertainty of the beam position on the sample. For example, for a detector distance of 100 mm and incidence/recoil angles of 15 /30 a displacement of only 175 lm is sufficient for 0.1 change in the recoil angle. A displacement of the sample surface relative to the eucentric point has an equivalent effect. The major systematic uncertainty in the present work is in the tabulated values for the stopping cross-section e and is at least 5%. The exception to this is the data for Si and C from Konac et al. [26], and see [25,27,28,30] which has about 2% uncertainty. Finally, H is lost from the sample during the measurement, and this has to be allowed for. The situation is complicated by the different behaviour of the surface and the bulk H. For all the measurements the ability to resolve the surface from the bulk is important, not only because the surface H behaves differently but also because the surface H varies strongly between the samples. For all the cases where the solid angle of the H detector must be determined separately, the instability of the Kapton H standard to the ion beam also has to be taken into account. 4. Discussion and conclusions The excellent inter-lab reproducibility of 2.2% for determination of the implanted H content of the samples is suspected to be accidental in view of the estimates of combined standard uncertainty (CSE) of about 6%. If the estimate of the uncertainty in the stopping powers is reduced to 2% then the estimated CSE falls to about 4%. The stopping powers used are for a wide variety of ion beams in both polymers (Kapton) and silicon and so there is no clear systematic bias in this set of results to account for this reproducibility. The three most similar participants (London, Surrey and Minnesota), with range foil light ion ERD, used different methods but essentially the same stopping powers and obtained closely similar results. We hope that the establishment of the KKKNS [25] stopping powers for He in Si as traceable with a CSE of 2% to certified standards [10] can rapidly be extended to other important cases, and a traceable standard for H with much better accuracy be developed. The H on the surface of these samples is not only sample and time dependent but also unstable to incident ion beams, and must be discriminated for reliable results. A portion of the surface film is very unstable to heavy ion beams and is rapidly removed. Consequently the HI-ERD results show less than half the surface H than the helium ERD. All of the HI-ERD analyses had difficulty resolving the surface signal of this low energy (6 kev) H implant, and a useful implanted standard would have a much deeper H profile. We should note that since the implant profile shape is well defined, a curve fitting algorithm (see Fig. 14) would greatly enhance the effective depth resolution. For this system it is not necessary to use a model-free analysis. Implanted samples are likely to be more stable than polymer ones and are therefore in principle capable of better accuracy. However, direct traceability of implanter dosimetry has yet to be established, and although the uncertainty of implanter dosimetry should be less than about 2% with care, traceable implanted standards still need separate certification. The uncertainty analysis makes it clear that after stopping powers, the most important contribution to the uncertainty for range foil ERD is in the ADC calibration. A proper analysis of this uncertainty is urgent. For all the ERD analyses the geometrical uncertainty is important and may be underestimated in the present work. For a traceable standard it will be necessary to determine an upper limit on this uncertainty by using various beam incident angles or otherwise. The original motivation for this Round Robin was to validate ANU s novel gas ionisation detector. This detector has the important advantage of combining relatively good resolution with complete lack of degradation under the beam. Quantitative results are obtainable even when this detector is not operated optimally for profiling H, although for this low energy implant the lack of resolution for the surface hydrogen limits the value of the data.