- Email: [email protected]
Surface roughening and erosion rate change at low energy SIMS depth profiling of silicon during oblique O2+ bombardment
Applied Surface Science 253 (2006) 2662–2670 www.elsevier.com/locate/apsusc
Surface roughening and erosion rate change at low energy SIMS depth profiling of silicon during oblique O2þ bombardment B. Fares a, B. Gautier a,*, Ph. Holliger b, N. Baboux a, G. Prudon a, J.-Cl. Dupuy a a
Laboratoire de Physique de la Matie`re, Institut National des Sciences Applique´es de Lyon, UMR CNRS 5511, Baˆtiment Blaise Pascal, 7 Avenue Capelle, F-69621 Villeurbanne Cedex, France b CEA-DRT, LETI/DST, CEA/GRE, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France Received 6 February 2006; received in revised form 18 May 2006; accepted 18 May 2006 Available online 30 June 2006
Abstract Surface roughening of boron d-doped Si samples under low energy (0.5 keV/ O2 þ , 44 and 54 , and 1.0 keV/ O2 þ , 48 ) O2 þ bombardment at oblique incidence with and without oxygen flooding was studied with atomic force microscopy (AFM) and secondary ion mass spectrometry (SIMS). The erosion rate, the surface topography and the depth resolution as a function of depth have been measured. Changes in secondary ion yields have been correlated with changes in surface topography. It is found that the surface roughness depends on impact energy and incidence angle without flooding. The roughness decreases with decreasing impact energy. For the same energy (0.5 keV/ O2 þ ), the wavelength increases slightly with increasing angle of incidence and the roughness increases with increasing angle of incidence. With flooding, the roughness can be efficiently avoided. The best conditions to avoid roughness when analysing ultra shallow profiles with our magnetic sector instrument is 0.5 keV/ O2 þ , 44 with flooding. A procedure for the depth calibration of the profiles revealed that surface roughness causes an erosion rate change as measured using the shift of the position of the measured B peaks with and without flooding. The consequences of the roughness in terms of depth resolution of the profiles are analysed with and without flooding. Moreover, we show that the value of the Gaussian broadening parameter of the depth resolution function is closely related to the final dispersion of the heights in the crater bottom. # 2006 Elsevier B.V. All rights reserved. Keywords: SIMS; AFM; d-Doped; RMS roughness; Depth resolution
1. Introduction The use of low energy ion bombardment in secondary ion mass spectrometry (SIMS) depth profiling is a mandatory step for the characterisation of ultra shallow junctions [1]. During the past years, the depth resolution has been enhanced, mainly by means of reduced primary ion energy and of deconvolution algorithms [2,3]. At present, the development of a severe roughness at the crater bottom remains one of the main obstacle toward the ultimate resolution [4], which makes mandatory the understanding and the experimental control of its appearance. Several past studies have aimed at this understanding: Ng et al. [5], for example, suggest a direct relation between surface
* Corresponding author. E-mail address: [email protected] (B. Gautier). 0169-4332/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.apsusc.2006.05.034
composition of the crater bottom and the development of surface roughness at low energy 0.5–2 keV/O2 þ without flooding. For the same energies and incidence angles between 45 and 80 , Jiang and Alkemade [6] show that surface roughening occurs at an eroded depth of only a few tens of nanometres. According to these authors, the inhomogeneous incorporation of oxygen, and the sputtering rate dependence on both surface topography and oxygen content, determines the occurrence of roughening [6]. Liu et al. [7] show that the highest depth resolution is achievable at 0.5 keV/O2 þ , 69 and that the onset of roughening occurs earlier for smaller incidence angles (in the 46–69 range). This angular dependence is explained using the heterogeneous layer model [8,9]. Another way to minimise roughness is the sample’s rotation [10,11]. Under flooding conditions, surface roughness at low energies occurs quickly and heavily at intermediate flooding pressures [6,8,12]. However, several authors did not observe
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2663
roughness at a saturation flooding pressure compared to UHV [5,13–15]. A number of studies [13,14] have shown that the depth resolution for B in Si bombarded by 3–8 keV/O2 þ is better with oxygen flooding, and that no roughness is observed at O2 þ saturation pressure. To explain this phenomenon, Ng et al. [5] suggest that the oxygen flooding leads to the formation of a homogeneous stoichiometric silicon dioxide on the crater bottom using 1 keV/O2 þ primary energy and 56 incidence angle. However, Wittmaack and Corcoran [12] suggest that with oxygen flooding, the erosion rate in Si for oblique 2 keV/ O2 þ beam changes significantly after removal of a layer of ’ 20–40 nm depth, which shows that flooding may not solve all the problems. When the roughness appears, the depth resolution degrades rapidly [16] as demonstrated, e.g. by Jiang and Alkemade [6], who worked in silicon with a boron multi-d layer structure and a 1 keV/O2 þ , 60 O2 þ beam. Eventually, when the roughness appears, the quantification of the SIMS analysis cannot be precise because of the variation of important parameters like the erosion rate, the ionisation yield and the sputter yield. The present paper explores in more detail the formation of surface roughness during sputter depth profiling of Si using 1 keV/O2 þ , 48 , and 0.5 keV/O2 þ , 44 and 54 with and without oxygen flooding. The erosion rate change, the variations in secondary ion intensities, the surface roughness, and the depth resolution for B deltas are measured. Multiple d layers are used to establish an intrinsic depth scale for ion beam sputter profiling and to measure the depth resolution. The aim of this paper is to explore the consequences of the roughness on very low primary beam energy analysis, needed for a very high depth resolution, using a direct measurement of the topography by atomic force microscopy (AFM). We also aim at establishing the optimal experimental conditions which leads to the best depth resolution with no roughness. 2. Experimental setup All SIMS depth profiles were performed on a Cameca IMS 5f instrument. The experiments reported here were carried out using an oxygen (O2 þ ) primary beam with a primary current in the 18.4–19 nA range at impact energy of 1 keV/O2 þ and incidence angle of 48 , and 0.5 keV/O2 þ at two different incidence angles (44 and 54 ). When oxygen flooding was used, the pressure was set at the saturation pressure of 2:0 106 Torr (we have checked that the saturation was effectively reached by varying the flow of the oxygen leak). The final crater depths were measured with a Tencor P10 surface profilometer and the depth scale was established assuming a constant erosion rate. The B d-doped Si sample [17] (abbreviated Si:B in the rest of the text) used in this work consists of one (mono) d layer at a depth of 17.9 nm under the surface, followed by five double d layers separated by approximately 1, 2, 3, 6 and 10 nm, respectively, and eventually another (mono) d. The position of the different layers, and the distance between them can be estimated from Fig. 1b. Surface topography measurements of the crater bottoms were performed using a Digital Instrument
Fig. 1. SIMS profiles of secondary ions from the B deltas analysed with: (a) Ep ¼ 1 keV/ O2 þ , 48 ; (b) 0.5 keV/ O2 þ , 44 ; (c) 0.5k eV/ O2 þ , 54 under UHV.
Dimension 3100 atomic force microscopy, operated in tapping mode. All AFM images were collected using the same silicon tips ( ’ 300 kHz resonance frequency, 40 N/m stiffness, radius < 10 nm), with scan sizes of 1 mm 1 mm.
2664
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
3. Results and discussion 3.1. Surface roughness and variations of secondary ion intensities 3.1.1. Ultra high vacuum (UHV) The appearance of the surface roughness depends on the eroded depth [18,19]. Fig. 1 shows the depth profile of the þ þ þ þ secondary ions (30 Si , 28 Si16 O , 28 Si2 16 O and 11 B ), measured under UHV for different impact energies and angles of incidences: (a) 1 keV/O2 þ , 48 ; (b) 0.5 keV/O2 þ , 44 ; (c) 0.5 keV/O2 þ , 54 . The apparent depth was established via the measured final crater depth: 220 nm for 1 keV/O2 þ , 48 ; 101 nm for 0.5 keV/O2 þ , 54 ; and 166 nm for 0.5 keV/O2 þ , 44 , assuming a constant erosion rate. In Fig. 2a, the onset of a significant surface roughening induces a variation in the secondary ion intensities of the matrix signal [20,21] and also distortions in the measured SIMS depth profiles [16] (the depth resolution degrades significantly). The matrix signals remain constant until the depth of 50 nm and then change dramatically þ þ þ (increase for 30 Si and 28 Si16 O , and decrease for 28 Si2 16 O ). After a transition phase, the signals stabilise again. An explanation for the intensity changes has been given by Wittmaack [22] who takes into account the local or microscopic angle of incidence which might not be the same (when roughness appears) as the macroscopic angle defined by the initial (plane) surface of the sample. In Fig. 1b and c, after a short surface transient ( ’ 4 nm for 0.5 keV/O2 þ , 44 , and ’ 8 nm for 0.5 keV/O2 þ , 54 ), the intensities of the matrix signals stabilise. Note that for the same impact energy (but different impact angle) the result is very different in Fig. 1b and c, the depth resolution degrading very quickly in the latter. Fig. 2 shows the AFM image topographies recorded at the bottom of craters corresponding to the SIMS analysis of Fig. 1a–c. When a topography appears, ridge lines (ripples) perpendicular to the direction of the primary beam can be observed. Table 1 provides a summary of the crater depth dependence of the root-mean-square (RMS) roughness and the wavelength, i.e., the distance between adjacent ripples, measured in each crater bottoms. The values must not be compared quantitatively since the SIMS analysis has not been stopped at the same depth, but the difference between the RMS roughness in each case allows us to compare the experimental conditions which lead to the development (or not) of the roughness and highlights the role of the incident angle for profiles performed using the same energy. The AFM pictures in Fig. 2a reveal the presence of a regular pseudo sinusoidal shaped topography with a wave vector parallel to the incident beam. The surface is rather rough (the root-mean-square roughness value is 3.85 nm and the wavelength is 66 nm). The typical peak-to-valley height of the ripples is ’ 10 nm. In Fig. 2c, the topography is composed of regular fine ripples superimposed on rather smooth bumps. The surface is not as rough as in the preceding case (the RMS roughness value is 1.54 nm and the wavelength is 33 nm). The typical peak-to-valley height of the ripples is ’ 2 nm. In
Fig. 2. AFM images 1 mm 1 mm of the crater surface after SIMS profiling with: (a) Ep ¼ 1 keV/ O2 þ , 48 ; (b) 0.5 keV/ O2 þ , 44 ; (c) 0.5 keV/ O2 þ , 54 under UHV. The z data scales were set at maximum values of 30 nm for (a), 2 nm for (b) and 10 nm for (c). Beam direction is from left to right.
contrast, the surface remains relatively smooth in Fig. 2b: the measured RMS roughness value is fairly close (0.21 nm) to the roughness of the initial (non-bombarded) silicon (0.1 nm). Thus, depending on experimental conditions, the topography changes from a smooth (Fig. 2b) to a rippled surface (Fig. 2a and c).
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2665
Table 1 Impact energy and angle of incidence (UHV)
Flooding
Depth of crater (nm)
RMS roughness (nm)
Wavelength (nm)
ld (nm)
s (nm)
FWHM (nm)
1 keV/ O2 þ , 48 0.5 keV/ O2 þ , 44 0.5 keV/ O2 þ , 54 1 keV/ O2 þ , 48 0.5 keV/ O2 þ , 44 0.5 keV/ O2 þ , 54
No No No Yes Yes Yes
220 166 101 220 277 174
3.85 0.21 1.54 0.17 0.14 0.40
66 23 33 15 10 15
’ 1.76 0.82 ’1 1.19 0.70 0.70
0.63–4.08 0.48–0.76 1.20–2.80 0.51–1.03 0.36–0.47 0.76–1.92
2.42–10.8 1.75–2.19 3.7 2.12–3.26 1.26–1.39 2.31–5.02
Again, we note that the surface roughness depends on impact energy and incident angle. The roughness decreases with decreasing impact energy: RMS ¼ 3:85 nm for 1 keV/O2 þ , 48 , and 0.21 nm for 0.5 keV/O2 þ , 44 . For the same energy (0.5 keV/O2 þ ), the wavelength increases slightly with increasing angle incidence (from 23 to 33 nm) and the RMS roughness increases dramatically when increasing of the incidence angle: 0.21 nm for 44 and 1.54 nm (seven times higher!) for 54 . This confirms that the incident angle plays a major role in the appearance of the roughness, in a context where the angle of incidence and the primary energy are linked in the apparatus used for the analysis. 3.1.2. Oxygen flooding Depth profiles have been measured with flooding at different impact energies and incident angle: (a) Ep ¼ 1 keV/ O2 þ , 48 O2 þ ; (b) 0.5 keV/O2 þ , 44 ; (c) 0.5 keV/O2 þ , 54 (see Fig. 3). With oxygen flooding at saturation, the intensity þ þ þ of 30 Si , 28 Si16 O and 28 Si2 16 O reaches equilibrium very rapidly after the beginning of the sputtering and the sputter rate remains constant throughout the entire depth profiling, except for a short surface transient phase (5 nm) for 0.5 keV/ O2 þ , 54 . The AFM images measured at the bottom of the craters produced in the corresponding experimental conditions (not shown) do not show any special features (no waves). Table 1 summarises the crater depth dependence of the root-mean-square roughness and the wavelengths. In any cases, the roughness is far less pronounced using oxygen flooding, although the morphology of the surface may differ from one case to another. In the case of 0.5 keV/O2 þ , 44 , the roughness is close to the roughness of the initial surface (closer than under UHV for the same energy and angle of incidence), leading to the conclusion that the best conditions (to avoid roughness) to analyse ultra shallow profiles in our study may be 0.5 keV/O2 þ , 44 with flooding. This will be confirmed by a further study of the depth resolution. 3.2. Change in apparent boron dose It seems interesting to measure the possible variation of the apparent boron dose of each d layer as a function of the analysis conditions and of the possible presence of the roughness. From the expression of the secondary ionic current I Bþ : I Bþ ¼ Y tot CðBÞt Bþ hI 0
(1)
and from the amount of target atoms dN s sputtered during the time dt dN s ¼ Y tot
I0 dt q
(2)
it is easy to show that: Z t2 Z z2 I Bþ dt ¼ hnqS0 CðBÞtBþ dz t1
(3)
z1
With Y tot : total sputter rate; CðBÞ: boron atomic concentration (depends on the depth: CðzÞ); tBþ : ionisation yield for the boron; h: transmission of the instrument; I 0 : primary intensity; S0 : analysed surface; q: elementary charge; t1 ; t2 ; z1 ; z2 are, respectively, the time (the depth) needed to analyse the whole d layer. We have plotted in Fig. 4 the apparent doses for each d or bid layer. These doses are normalised with respect to the first d layer, and the with respect to the successive d layers of the analysis at 0.5 keV/O2 þ , 44 . Thus, we suppose that t B þ does not vary during this analysis, which seems reasonable. The variations of the dose and consequently of the ionisation yield for the other analysis are then referenced to 1 for the first peak of each analysis. Following this procedure, a clear tendency can be detected: For the three analysis operated with flooding, the ionisation yield does not vary significantly, even when the roughness reaches 0.4 nm RMS. This is the signature of a surface saturated with oxygen, this saturation does not vary during the formation of the roughness. For the analysis without flooding (UHV), the ionisation rate varies with depth in both cases where the roughness appears. This is consistent with the existence of two different facets showing different incidence angles with respect to the primary beam, the facet exposed to the beam (the incidence angle is close to the normal) is the most oxidised as in the case of the analysis using higher energies [23], which increases the global ionisation rate in the same time it decreases the average erosion speed. 3.3. Change in erosion rate The onset of a significant surface roughening causes a change in erosion rate. Wittmaack [24] assumes that the erosion rate decreases in two steps, a rapid initial fall-off due to oxygen
2666
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
Fig. 4. Relative ionisation yield vs. depth for different incident energies/angles with or without flooding.
In this section, we analyse the surface roughness influence on the depth calibration procedure. A similar analysis was carried out by Wittmaack [24]. As bombardment takes place, the erosion rate varies at first time from a value nt (beginning of the transient state) to a value np (end of the transient state) because of the incorporation of the primary ions (finc , where f is the instantaneous beam dose and finc is the dose needed to reach the permanent state after the incorporation of the primary ions of the matrix). Then the erosion rate varies a second time when f reaches the value where roughness appears fr (finc fr ) to stabilise at a value nr when the roughness stops increasing. The depth z varies with primary dose in the following way: z ¼ Dzinc þ np f;
for f > finc
(4)
z ¼ Dztot þ nr f;
for f fr
(5)
with Dzinc is the shift arising from beam incorporation and Dztot is the shift arising from beam incorporation (Dzinc ) and from the roughness (Dzrug ): Dztot ¼ Dzinc þ Dzrug . The conventional depth calibration procedure supposes that the erosion rate is constant during all the analysis. Thus, it provides only the apparent depth: zapp ¼ napp f (napp is the mean-apparent-erosion speed). As the erosion rate is not constant, this leads to a shift, i.e., a difference between the true and the apparent depths, defined by: zshift ¼ zapp ztrue Generally, the apparent shift (whatever the value given to napp ) is: Fig. 3. SIMS profiles of secondary ions from the B deltas, analysed with: (a) Ep ¼ 1 keV/ O2 þ , 48 ; (b) 0.5 keV/ O2 þ , 44 ; (c) 0.5 keV/ O2 þ , 54 under oxygen flooding.
incorporation in the sample (less pronounced), and a long-term change associated with ripple formation. Tian and Vandervorst [25] suggested a different mechanism for the erosion rate change under flooding conditions.
zshift ¼ Dzinc þ ðnapp np Þf;
for f finc
(6)
for f fr
(7)
and zshift ¼ Dztot þ ðnapp nr Þf;
We supposed in this study that the apparent depths of the d layers measured with flooding (0.5 keV/O2 þ , 44 ) are the real depths because no roughness is observed using these conditions
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2667
with the position of the peak measured at 44 with flooding) for 0.5 keV/O2 þ , 54 with oxygen flooding. The shift suggests a fast variation of the erosion rate at the beginning of the sputtering, which varies quickly at the beginning, then stabilises, until it is equal to the rate measured at 0.5 keV/ O2 þ , 44 . Figs. 5 and 6 show that the erosion rate is different under UHV and using oxygen flooding: the difference is due to the development of roughness under UHV, leading to two different regimes for the erosion speed, as developed at the beginning of this section, the first one corresponding to the development of the altered layer, the second corresponding to the development of the roughness (note that the depth of the transition between both regimes corresponds to the change in the 30 Si16 O signal indicating the development of the roughness). Fig. 5. Shift vs. the apparent depth at 1 keV/ O2 þ , 48 with (A) and without flooding (B).
of analysis (see Section 3.1.2). From this ‘‘true depth’’ of the layers, we determined the apparent shift for the SIMS profiling at 1 keV/O2 þ , 48 . Fig. 5 shows the apparent shift versus the apparent depth with (A) and without flooding (B). The linearity of the curve corresponding to analysis A confirms that erosion rate remains constant with flooding. On the other hand, the curve corresponding to the analysis B is complex. Two different erosion rates are detected before and after 60 nm. This confirms the hypothetical behavior presented at the beginning of this section (Eqs. (3) and (4)): two zones where the behavior is linear, corresponding to the phases where the erosion rate is, respectively, np and nr . From the slopes of theses lines, we can deduce that np =napp ¼ 1:17 and nr =napp ¼ 0:9. Thus, the variation of erosion rate without oxygen flooding is such that np =nr = 1.17, which means that a drop of ’ 30% has occurred. Fig. 6 shows the variation of the peak shift with respect to the ‘‘real’’ value of the peak depth (still measured by comparing
Fig. 6. Measurement of the apparent shift vs. the real depth for 0.5 keV/ O2 þ , 54 with flooding.
3.4. Depth resolution It is well-known that the depth resolution of SIMS is influenced by atomic mixing [26], which induces a profile broadening and a shift from the true position, and by surface roughening [6]. We use the description of the Depth Resolution Function (DRF) of the SIMS analysis proposed by Dowsett et al. [27], now widely used by people who wish to characterise objectively the depth resolution [28,29,2,30,3], and which is described by three independent parameters: lu (the rising exponential part of the profile), s 0 (the Gaussian top) and ld (distance over which the intensity drops by a factor of e, arising from the collisional mixing). The decay length ld of the profiles measured by Ng et al. [5] is 1.2 nm and the broadening parameter s 0 is 0.75 nm using 0.5 keV/O2 þ , 56 with flooding. Takano et al. [31] and Hayashi et al. [32] used a theoretical model, the mixing-roughness-information model (MRI, initially developed by Hofmann [26]) to extract the depth resolution parameters. This model takes into account a parameter s relative to the roughness of the surface and a parameter w which quantifies the collisional mixing. Hayashi et al. [32] showed that the depth resolution degrades when the impact energy is less than 0.5 keV/O2 þ . The parameter s of the MRI model was calculated: s ¼ 3 nm without flooding, 0.5 keV/O2 þ , and 1.4 nm for 1 keV/O2 þ , because of the surface roughening. Takano et al. [31] showed that the best depth resolution is obtained with 1.5 keV/O2 þ at 45 (s ¼ 1:4 nm and w ¼ 1:7 nm) which are relatively high primary energy compared to this work. Moreover, the use of the oxygen flooding has sometimes been found to impede the accurate depth calibration of the SIMS analysis, and does not always lead to the best resolution [6]. Thus, there is a need to clarify this point. In order to better understand the influence of the oxygen flooding on the appearance of the roughness and the depth resolution, we extracted the parameters of the DRF ld, lu and s 0 by fitting the shallower peak, i.e., the peak measured at a depth where the roughness is not present. The values of the parameters are consistent with previous results [17]: lu is very small, lu ¼ 0:7 nm, ld ¼ 0:71 nm and s 0 ¼ 0:3 nm for a primary energy of 0.5 keV/O2 þ and a 44 incidence angle with
2668
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
flooding. Each double d layer is fitted with a function corresponding to the sum of two DRFs, ld and lu being set to the value corresponding to the roughness-free DRF, but the other parameters (position, sigma) remaining free. The fit is excellent most of the time, which means that the profile of the d layers when the roughness is present is fairly well described by a function of the parameters ld , lu and s which is the convolution of the DRF (ld , lu and s 0 ) with a Gaussian which standard deviation is s r , such that s 2 ¼ s 2r þ s 20 . This also shows that the Gaussian describes the induced roughness fairly well. For the d layers located at ’ 135 nm, and for 0.5 keV/O2 þ , 44 with flooding, the result of the fitting is: lu ¼ 0:48 nm, ld ¼ 0:71 nm, s 0 ¼ 0:32 nm and s r ¼ 0:41 nm, the positions z1 and z2 of the double peak are 128.7 and 139.2 nm. For 1 keV/ O2 þ , 48 , UHV: lu ¼ 0:43 nm, ld ¼ 1:75 nm, s 0 ¼ 0:64 nm and s r ¼ 3:99 nm, the positions z1 and z2 of the double peak are 127.3 and 138.0 nm. This leads to an additional broadening at the depth ’ 130 nm: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s r ð0:5 keV=O2 þ ; 130 nmÞ ¼ 0:412 0:322 ¼ 0:25 nm (8) s r ð1 keV=O2 þ ; 130 nmÞ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3:992 0:642 ¼ 3:94 nm
(9)
We plotted the fitted broadening parameter s as function of depth in Fig. 7. It shows that sputtering with oxygen flooding obviously results in smaller values of s. When flooding is used at 0.5 keV/O2 þ , 44 , s remains nearly constant with increasing eroded depth. Under UHV in the same conditions, s increases slightly but remains small (0.48–0.76 nm). We also note that with oxygen flooding at 0.5 keV/O2 þ , 54 , s increases linearly after the third peak, which is the signature of the development of the roughness which appears around 50 nm, in agreement with the AFM results of Fig. 5b. Under UHV (0.5 keV/O2 þ , 54 and 1 keV/O2 þ , 48 ), a strong and linear increase of s is found (from 1.2 to 2.8 nm) and the depth resolution is poor: ld ’ 1:76 nm (measured using the deepest d layer). The depth at which the depth resolution starts
to degrade is 60 nm for 1 keV/O2 þ , 48 . For the same energy with flooding, a slight increase was observed for s (0.51– 1.03 nm). It must be pointed out that measuring s is a very sensitive way to evaluate the roughness without the need for an AFM image. Even a small development of the roughness in the crater bottom can be detected by plotting the evolution of this parameter with depth. The values of the decay length ld and of the Gaussian broadening parameter s, for different impact energies and incident angle with and without oxygen flooding are summarised in Table 1. Relating the measurements of s and ld and the AFM measurements in the crater bottom (Fig. 3a) we can conclude that the distortions of the profiles are directly related to the onset of surface roughening. The profile broadening with depth is far more severe under UHV conditions. This means that oxygen flooding leads to the best depth resolution. In Table 1, the measured peak depths and measured Full Width at Half Maximum (FWHM) of the d layers are summarised. From the literature, Jiang and Alkemade [6] indeed performed a very important work dealing with the roughness at low energy primary bombardment. However, this topic cannot be considered as being entirely understood. As an example, Jiang et al. obtained their best depth resolution in the 4 nm range FWHM in the case of the analysis of d-doped layers. It is possible that such a (bad) apparent resolution originates from a bad quality of the sample obtained by MBE. The depth resolution obtained in our work in the best experimental conditions (500 eV, no roughness, flooding) is 1.5 nm FWHM. 3.5. Relation between s and the topography In this section, we show that the RMS roughness scales with the value of the Gaussian broadening parameter s. Fig. 8a shows a cross-section of the ripple-like topography for 1 keV/ O2 þ , 48 at UHV. According to Table 1, the value of the RMS roughness is 3.85 nm. Then we plot a histogram gathering the effective depths of the cross section with a step of 0.6 nm (Fig. 8b). This histogram is fitted with a Gaussian function: y ¼ y0 þ A eððxxc Þ=ð2seff ÞÞ
2
The best fit is obtained with a value of about 4.17 nm. Finally, we plot the fit of the last peak in Fig. 2a by the depth resolution function DRF [27] (Fig. 8c). The value of the broadening parameter is about 4.1 nm. If the statistical broadening of the profile is equal to the value of the broadening parameter s obtained using 0.5 keV/O2 þ , 44 with flooding, again we have [33]: s 2 ¼ s 2r þ s 20
Fig. 7. The fitted broadening parameter s as a function of depth.
where s is the total broadening of the profile and s r is the broadening due to the roughness. As a result, the statistical distribution of the heights in the crater bottom is Gaussian as a first approximation. A random distribution has to be
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2669
correlation at such a low scale (see also a paper dealing with SiGe [34]). The Gaussian distribution is then a correct first order description of the phenomena (second order centred moments). 4. Conclusion In this study, we have studied the occurrence of roughness in the SIMS crater bottom when different experimental conditions are used (impact energy, angle of incidence, with and without oxygen flooding) leading to a very high depth resolution. The roughness is always very weak using oxygen flooding compared to UVH in the experimental conditions we explored. Oxygen bombardment without flooding during low energy (1 keV/O2 þ , 48 and 0.5 keV/O2 þ , 54 ) induces surface roughening, distortions of the profiles and erosion rate changes compared with oxygen flooding. At low energy, the profile broadening with depth is more severe under UHV compared with oxygen flooding. Moreover, using AFM images, we show that the RMS roughness is closely related to the value of the Gaussian broadening parameter. This study has shown that oblique incidence O2 þ bombardment with oxygen flooding can be used to obtain accurate, high depth resolution shallow junction profiles in silicon, with almost no roughness in the crater. In our study, the optimal experimental conditions (no surface roughening, optimal depth resolution) seem to be 0.5 keV/O2 þ 44 with flooding. References
Fig. 8. (a) The cross-section of the ripple topography for 1 keV/ O2 þ , 48 , UHV, (b) fit of the last peak using an analytical expression of the DRF convolved with a Gaussian and (c) histogram of the depth in the crater bottom (step: 0.6 nm).
considered, described by a Gaussian which standard deviation is directly correlated to the broadening of the SIMS profiles (s), with the same order of magnitude. This is to our knowledge the first direct demonstration of the role of the roughness on the shape of the profiles in silicon, with a direct quantitative
[1] K. Wittmaak, J Vac. Sci. Technol. B 16 (5) (1998) 2776–2785. [2] B. Gautier, R. Prost, G. Prudon, J.-C. Dupuy, Surf. Interface Anal. 24 (1996) 733–745. [3] B. Gautier, R. Prost, G. Prudon, J.-C. Dupuy, Surf. Interface Anal. 26 (1998) 974–983. [4] B. Fares, B. Gautier, N. Baboux, G. Prudon, P. Holliger, J.-C. Dupuy, Appl. Surf. Sci. 231–232 (2004) 136–140. [5] C. Ng, A. Wee, C. Huan, A. See, Appl. Surf. Sci. (2001) 557–560. [6] Z. Jiang, P. Alkemade, J. Vac. Sci. Technol. B 16 (4) (1998) 1971–1982. [7] R. Liu, C. Ng, A. Wee, Appl. Surf. Sci. 203–204 (2003) 256–259. [8] K. Elst, W. Vandervorst, J. Alay, J. Vac. Sci. Technol. B 11 (6) (1993) 1968–1993. [9] K. Elst, W. Vandervorst, J. Vac. Sci. Technol. B 12 (6) (1994) 3205–3216. [10] R. Liu, A. Wee, D. Shen, H. Takenaka, Surf. Interface Anal. 36 (2004) 172–176. [11] R. Liu, A. Wee, Appl. Surf. Sci. 231–232 (2004) 653–657. [12] K. Wittmaak, S. Corcoran, J. Vac. Sci. Technol. B 16 (1) (1998) 272–279. [13] P. Zalm, C. Vriezema, Nucl. Instrum. Meth. Phys. B 67 (1992) 495–499. [14] J. Erickson, R. Brigham, J. Vac. Sci. Technol. B 14 (1) (1996) 353–357. [15] F. Jahnel, R.V. Criegern, Appl. Surf. Sci. 203–204 (2004) 367–370. [16] P. Alkemade, Z. Jiang, J. Vac. Sci. Technol. B 19 (5) (2001) 1699–1705. [17] N. Baboux, J. Dupuy, G. Prudon, P. Holliger, F. Laugier, A. Papon, J. Hartmann, J. Cryst. Growth 245 (2002) 1–8. [18] E. Cirlin, J. Vajo, T. Hasenberg, J. Vac. Sci. Technol. B 12 (1) (1994) 269– 275. [19] Z. Jiang, P. Alkemade, Appl. Phys. Lett. 73 (3) (1998) 315–317. [20] J. Vajo, R. Doty, E. Cirlin, J. Vac. Sci. Technol. A 14 (5) (1996) 2709– 2720. [21] G. Lau, E. Tok, W. ATS, L. R, L. SL, Surf. Rev. Lett. 8 (5) (2001) 453–457. [22] K. Wittmaack, J. Vac. Sci. Technol. A 8 (3) (1990) 2246–2250. [23] B. Gautier, B. Fares, G. Prudon, J.-C. Dupuy, Appl. Surf. Sci. 131–132 (2004) 136–140.
2670
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
[24] K. Wittmaack, J. Vac. Sci. Technol. B 18 (1) (2000) 1–6. [25] C. Tian, W. Vandervorst, J. Vac. Sci. Technol. A 15 (3) (1997) 452– 459. [26] S. Hoffmann, J. Vac. Sci. Technol. A 9 (3) (1991) 1466–1476. [27] M. Dowsett, G. Rowlands, P. Allen, R. Barlow, Surf. Interface Anal. 21 (1994) 310–315. [28] J. Lee, K. Kim, H. Kim, D. Moon, Surf. Interface Anal. 37 (2) (2005) 176– 180. [29] D. Chu, M. Dowsett, Phys. Rev. B 56 (23) (1997) 15167–15170.
[30] B. Gautier, J.-C. Dupuy, R. Prost, G. Prudon, Surf. Interface Anal. 25 (6) (1997) 464–477. [31] A. Takano, Y. Homma, Y. Higashi, Appl, Surf. Sci. 203–204 (2003) 294– 297. [32] S. Hayashi, A. Takano, H. Takenaka, Y. Tomma, Appl. Surf. Sci. 203–204 (2003) 298–301. [33] B. Gautier, G. Prudon, B. Semmache, J.-C. Dupuy, Nucl. Instrum. Meth. Phys. B 142 (1998) 361–376. [34] G. Lau, E. Tok, R. Liu, Nucl. Instrum. Meth. Phys. B 215 (2004) 76–82.
Surface roughening and erosion rate change at low energy SIMS depth profiling of silicon during oblique O2þ bombardment B. Fares a, B. Gautier a,*, Ph. Holliger b, N. Baboux a, G. Prudon a, J.-Cl. Dupuy a a
Laboratoire de Physique de la Matie`re, Institut National des Sciences Applique´es de Lyon, UMR CNRS 5511, Baˆtiment Blaise Pascal, 7 Avenue Capelle, F-69621 Villeurbanne Cedex, France b CEA-DRT, LETI/DST, CEA/GRE, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France Received 6 February 2006; received in revised form 18 May 2006; accepted 18 May 2006 Available online 30 June 2006
Abstract Surface roughening of boron d-doped Si samples under low energy (0.5 keV/ O2 þ , 44 and 54 , and 1.0 keV/ O2 þ , 48 ) O2 þ bombardment at oblique incidence with and without oxygen flooding was studied with atomic force microscopy (AFM) and secondary ion mass spectrometry (SIMS). The erosion rate, the surface topography and the depth resolution as a function of depth have been measured. Changes in secondary ion yields have been correlated with changes in surface topography. It is found that the surface roughness depends on impact energy and incidence angle without flooding. The roughness decreases with decreasing impact energy. For the same energy (0.5 keV/ O2 þ ), the wavelength increases slightly with increasing angle of incidence and the roughness increases with increasing angle of incidence. With flooding, the roughness can be efficiently avoided. The best conditions to avoid roughness when analysing ultra shallow profiles with our magnetic sector instrument is 0.5 keV/ O2 þ , 44 with flooding. A procedure for the depth calibration of the profiles revealed that surface roughness causes an erosion rate change as measured using the shift of the position of the measured B peaks with and without flooding. The consequences of the roughness in terms of depth resolution of the profiles are analysed with and without flooding. Moreover, we show that the value of the Gaussian broadening parameter of the depth resolution function is closely related to the final dispersion of the heights in the crater bottom. # 2006 Elsevier B.V. All rights reserved. Keywords: SIMS; AFM; d-Doped; RMS roughness; Depth resolution
1. Introduction The use of low energy ion bombardment in secondary ion mass spectrometry (SIMS) depth profiling is a mandatory step for the characterisation of ultra shallow junctions [1]. During the past years, the depth resolution has been enhanced, mainly by means of reduced primary ion energy and of deconvolution algorithms [2,3]. At present, the development of a severe roughness at the crater bottom remains one of the main obstacle toward the ultimate resolution [4], which makes mandatory the understanding and the experimental control of its appearance. Several past studies have aimed at this understanding: Ng et al. [5], for example, suggest a direct relation between surface
* Corresponding author. E-mail address: [email protected] (B. Gautier). 0169-4332/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.apsusc.2006.05.034
composition of the crater bottom and the development of surface roughness at low energy 0.5–2 keV/O2 þ without flooding. For the same energies and incidence angles between 45 and 80 , Jiang and Alkemade [6] show that surface roughening occurs at an eroded depth of only a few tens of nanometres. According to these authors, the inhomogeneous incorporation of oxygen, and the sputtering rate dependence on both surface topography and oxygen content, determines the occurrence of roughening [6]. Liu et al. [7] show that the highest depth resolution is achievable at 0.5 keV/O2 þ , 69 and that the onset of roughening occurs earlier for smaller incidence angles (in the 46–69 range). This angular dependence is explained using the heterogeneous layer model [8,9]. Another way to minimise roughness is the sample’s rotation [10,11]. Under flooding conditions, surface roughness at low energies occurs quickly and heavily at intermediate flooding pressures [6,8,12]. However, several authors did not observe
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2663
roughness at a saturation flooding pressure compared to UHV [5,13–15]. A number of studies [13,14] have shown that the depth resolution for B in Si bombarded by 3–8 keV/O2 þ is better with oxygen flooding, and that no roughness is observed at O2 þ saturation pressure. To explain this phenomenon, Ng et al. [5] suggest that the oxygen flooding leads to the formation of a homogeneous stoichiometric silicon dioxide on the crater bottom using 1 keV/O2 þ primary energy and 56 incidence angle. However, Wittmaack and Corcoran [12] suggest that with oxygen flooding, the erosion rate in Si for oblique 2 keV/ O2 þ beam changes significantly after removal of a layer of ’ 20–40 nm depth, which shows that flooding may not solve all the problems. When the roughness appears, the depth resolution degrades rapidly [16] as demonstrated, e.g. by Jiang and Alkemade [6], who worked in silicon with a boron multi-d layer structure and a 1 keV/O2 þ , 60 O2 þ beam. Eventually, when the roughness appears, the quantification of the SIMS analysis cannot be precise because of the variation of important parameters like the erosion rate, the ionisation yield and the sputter yield. The present paper explores in more detail the formation of surface roughness during sputter depth profiling of Si using 1 keV/O2 þ , 48 , and 0.5 keV/O2 þ , 44 and 54 with and without oxygen flooding. The erosion rate change, the variations in secondary ion intensities, the surface roughness, and the depth resolution for B deltas are measured. Multiple d layers are used to establish an intrinsic depth scale for ion beam sputter profiling and to measure the depth resolution. The aim of this paper is to explore the consequences of the roughness on very low primary beam energy analysis, needed for a very high depth resolution, using a direct measurement of the topography by atomic force microscopy (AFM). We also aim at establishing the optimal experimental conditions which leads to the best depth resolution with no roughness. 2. Experimental setup All SIMS depth profiles were performed on a Cameca IMS 5f instrument. The experiments reported here were carried out using an oxygen (O2 þ ) primary beam with a primary current in the 18.4–19 nA range at impact energy of 1 keV/O2 þ and incidence angle of 48 , and 0.5 keV/O2 þ at two different incidence angles (44 and 54 ). When oxygen flooding was used, the pressure was set at the saturation pressure of 2:0 106 Torr (we have checked that the saturation was effectively reached by varying the flow of the oxygen leak). The final crater depths were measured with a Tencor P10 surface profilometer and the depth scale was established assuming a constant erosion rate. The B d-doped Si sample [17] (abbreviated Si:B in the rest of the text) used in this work consists of one (mono) d layer at a depth of 17.9 nm under the surface, followed by five double d layers separated by approximately 1, 2, 3, 6 and 10 nm, respectively, and eventually another (mono) d. The position of the different layers, and the distance between them can be estimated from Fig. 1b. Surface topography measurements of the crater bottoms were performed using a Digital Instrument
Fig. 1. SIMS profiles of secondary ions from the B deltas analysed with: (a) Ep ¼ 1 keV/ O2 þ , 48 ; (b) 0.5 keV/ O2 þ , 44 ; (c) 0.5k eV/ O2 þ , 54 under UHV.
Dimension 3100 atomic force microscopy, operated in tapping mode. All AFM images were collected using the same silicon tips ( ’ 300 kHz resonance frequency, 40 N/m stiffness, radius < 10 nm), with scan sizes of 1 mm 1 mm.
2664
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
3. Results and discussion 3.1. Surface roughness and variations of secondary ion intensities 3.1.1. Ultra high vacuum (UHV) The appearance of the surface roughness depends on the eroded depth [18,19]. Fig. 1 shows the depth profile of the þ þ þ þ secondary ions (30 Si , 28 Si16 O , 28 Si2 16 O and 11 B ), measured under UHV for different impact energies and angles of incidences: (a) 1 keV/O2 þ , 48 ; (b) 0.5 keV/O2 þ , 44 ; (c) 0.5 keV/O2 þ , 54 . The apparent depth was established via the measured final crater depth: 220 nm for 1 keV/O2 þ , 48 ; 101 nm for 0.5 keV/O2 þ , 54 ; and 166 nm for 0.5 keV/O2 þ , 44 , assuming a constant erosion rate. In Fig. 2a, the onset of a significant surface roughening induces a variation in the secondary ion intensities of the matrix signal [20,21] and also distortions in the measured SIMS depth profiles [16] (the depth resolution degrades significantly). The matrix signals remain constant until the depth of 50 nm and then change dramatically þ þ þ (increase for 30 Si and 28 Si16 O , and decrease for 28 Si2 16 O ). After a transition phase, the signals stabilise again. An explanation for the intensity changes has been given by Wittmaack [22] who takes into account the local or microscopic angle of incidence which might not be the same (when roughness appears) as the macroscopic angle defined by the initial (plane) surface of the sample. In Fig. 1b and c, after a short surface transient ( ’ 4 nm for 0.5 keV/O2 þ , 44 , and ’ 8 nm for 0.5 keV/O2 þ , 54 ), the intensities of the matrix signals stabilise. Note that for the same impact energy (but different impact angle) the result is very different in Fig. 1b and c, the depth resolution degrading very quickly in the latter. Fig. 2 shows the AFM image topographies recorded at the bottom of craters corresponding to the SIMS analysis of Fig. 1a–c. When a topography appears, ridge lines (ripples) perpendicular to the direction of the primary beam can be observed. Table 1 provides a summary of the crater depth dependence of the root-mean-square (RMS) roughness and the wavelength, i.e., the distance between adjacent ripples, measured in each crater bottoms. The values must not be compared quantitatively since the SIMS analysis has not been stopped at the same depth, but the difference between the RMS roughness in each case allows us to compare the experimental conditions which lead to the development (or not) of the roughness and highlights the role of the incident angle for profiles performed using the same energy. The AFM pictures in Fig. 2a reveal the presence of a regular pseudo sinusoidal shaped topography with a wave vector parallel to the incident beam. The surface is rather rough (the root-mean-square roughness value is 3.85 nm and the wavelength is 66 nm). The typical peak-to-valley height of the ripples is ’ 10 nm. In Fig. 2c, the topography is composed of regular fine ripples superimposed on rather smooth bumps. The surface is not as rough as in the preceding case (the RMS roughness value is 1.54 nm and the wavelength is 33 nm). The typical peak-to-valley height of the ripples is ’ 2 nm. In
Fig. 2. AFM images 1 mm 1 mm of the crater surface after SIMS profiling with: (a) Ep ¼ 1 keV/ O2 þ , 48 ; (b) 0.5 keV/ O2 þ , 44 ; (c) 0.5 keV/ O2 þ , 54 under UHV. The z data scales were set at maximum values of 30 nm for (a), 2 nm for (b) and 10 nm for (c). Beam direction is from left to right.
contrast, the surface remains relatively smooth in Fig. 2b: the measured RMS roughness value is fairly close (0.21 nm) to the roughness of the initial (non-bombarded) silicon (0.1 nm). Thus, depending on experimental conditions, the topography changes from a smooth (Fig. 2b) to a rippled surface (Fig. 2a and c).
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2665
Table 1 Impact energy and angle of incidence (UHV)
Flooding
Depth of crater (nm)
RMS roughness (nm)
Wavelength (nm)
ld (nm)
s (nm)
FWHM (nm)
1 keV/ O2 þ , 48 0.5 keV/ O2 þ , 44 0.5 keV/ O2 þ , 54 1 keV/ O2 þ , 48 0.5 keV/ O2 þ , 44 0.5 keV/ O2 þ , 54
No No No Yes Yes Yes
220 166 101 220 277 174
3.85 0.21 1.54 0.17 0.14 0.40
66 23 33 15 10 15
’ 1.76 0.82 ’1 1.19 0.70 0.70
0.63–4.08 0.48–0.76 1.20–2.80 0.51–1.03 0.36–0.47 0.76–1.92
2.42–10.8 1.75–2.19 3.7 2.12–3.26 1.26–1.39 2.31–5.02
Again, we note that the surface roughness depends on impact energy and incident angle. The roughness decreases with decreasing impact energy: RMS ¼ 3:85 nm for 1 keV/O2 þ , 48 , and 0.21 nm for 0.5 keV/O2 þ , 44 . For the same energy (0.5 keV/O2 þ ), the wavelength increases slightly with increasing angle incidence (from 23 to 33 nm) and the RMS roughness increases dramatically when increasing of the incidence angle: 0.21 nm for 44 and 1.54 nm (seven times higher!) for 54 . This confirms that the incident angle plays a major role in the appearance of the roughness, in a context where the angle of incidence and the primary energy are linked in the apparatus used for the analysis. 3.1.2. Oxygen flooding Depth profiles have been measured with flooding at different impact energies and incident angle: (a) Ep ¼ 1 keV/ O2 þ , 48 O2 þ ; (b) 0.5 keV/O2 þ , 44 ; (c) 0.5 keV/O2 þ , 54 (see Fig. 3). With oxygen flooding at saturation, the intensity þ þ þ of 30 Si , 28 Si16 O and 28 Si2 16 O reaches equilibrium very rapidly after the beginning of the sputtering and the sputter rate remains constant throughout the entire depth profiling, except for a short surface transient phase (5 nm) for 0.5 keV/ O2 þ , 54 . The AFM images measured at the bottom of the craters produced in the corresponding experimental conditions (not shown) do not show any special features (no waves). Table 1 summarises the crater depth dependence of the root-mean-square roughness and the wavelengths. In any cases, the roughness is far less pronounced using oxygen flooding, although the morphology of the surface may differ from one case to another. In the case of 0.5 keV/O2 þ , 44 , the roughness is close to the roughness of the initial surface (closer than under UHV for the same energy and angle of incidence), leading to the conclusion that the best conditions (to avoid roughness) to analyse ultra shallow profiles in our study may be 0.5 keV/O2 þ , 44 with flooding. This will be confirmed by a further study of the depth resolution. 3.2. Change in apparent boron dose It seems interesting to measure the possible variation of the apparent boron dose of each d layer as a function of the analysis conditions and of the possible presence of the roughness. From the expression of the secondary ionic current I Bþ : I Bþ ¼ Y tot CðBÞt Bþ hI 0
(1)
and from the amount of target atoms dN s sputtered during the time dt dN s ¼ Y tot
I0 dt q
(2)
it is easy to show that: Z t2 Z z2 I Bþ dt ¼ hnqS0 CðBÞtBþ dz t1
(3)
z1
With Y tot : total sputter rate; CðBÞ: boron atomic concentration (depends on the depth: CðzÞ); tBþ : ionisation yield for the boron; h: transmission of the instrument; I 0 : primary intensity; S0 : analysed surface; q: elementary charge; t1 ; t2 ; z1 ; z2 are, respectively, the time (the depth) needed to analyse the whole d layer. We have plotted in Fig. 4 the apparent doses for each d or bid layer. These doses are normalised with respect to the first d layer, and the with respect to the successive d layers of the analysis at 0.5 keV/O2 þ , 44 . Thus, we suppose that t B þ does not vary during this analysis, which seems reasonable. The variations of the dose and consequently of the ionisation yield for the other analysis are then referenced to 1 for the first peak of each analysis. Following this procedure, a clear tendency can be detected: For the three analysis operated with flooding, the ionisation yield does not vary significantly, even when the roughness reaches 0.4 nm RMS. This is the signature of a surface saturated with oxygen, this saturation does not vary during the formation of the roughness. For the analysis without flooding (UHV), the ionisation rate varies with depth in both cases where the roughness appears. This is consistent with the existence of two different facets showing different incidence angles with respect to the primary beam, the facet exposed to the beam (the incidence angle is close to the normal) is the most oxidised as in the case of the analysis using higher energies [23], which increases the global ionisation rate in the same time it decreases the average erosion speed. 3.3. Change in erosion rate The onset of a significant surface roughening causes a change in erosion rate. Wittmaack [24] assumes that the erosion rate decreases in two steps, a rapid initial fall-off due to oxygen
2666
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
Fig. 4. Relative ionisation yield vs. depth for different incident energies/angles with or without flooding.
In this section, we analyse the surface roughness influence on the depth calibration procedure. A similar analysis was carried out by Wittmaack [24]. As bombardment takes place, the erosion rate varies at first time from a value nt (beginning of the transient state) to a value np (end of the transient state) because of the incorporation of the primary ions (finc , where f is the instantaneous beam dose and finc is the dose needed to reach the permanent state after the incorporation of the primary ions of the matrix). Then the erosion rate varies a second time when f reaches the value where roughness appears fr (finc fr ) to stabilise at a value nr when the roughness stops increasing. The depth z varies with primary dose in the following way: z ¼ Dzinc þ np f;
for f > finc
(4)
z ¼ Dztot þ nr f;
for f fr
(5)
with Dzinc is the shift arising from beam incorporation and Dztot is the shift arising from beam incorporation (Dzinc ) and from the roughness (Dzrug ): Dztot ¼ Dzinc þ Dzrug . The conventional depth calibration procedure supposes that the erosion rate is constant during all the analysis. Thus, it provides only the apparent depth: zapp ¼ napp f (napp is the mean-apparent-erosion speed). As the erosion rate is not constant, this leads to a shift, i.e., a difference between the true and the apparent depths, defined by: zshift ¼ zapp ztrue Generally, the apparent shift (whatever the value given to napp ) is: Fig. 3. SIMS profiles of secondary ions from the B deltas, analysed with: (a) Ep ¼ 1 keV/ O2 þ , 48 ; (b) 0.5 keV/ O2 þ , 44 ; (c) 0.5 keV/ O2 þ , 54 under oxygen flooding.
incorporation in the sample (less pronounced), and a long-term change associated with ripple formation. Tian and Vandervorst [25] suggested a different mechanism for the erosion rate change under flooding conditions.
zshift ¼ Dzinc þ ðnapp np Þf;
for f finc
(6)
for f fr
(7)
and zshift ¼ Dztot þ ðnapp nr Þf;
We supposed in this study that the apparent depths of the d layers measured with flooding (0.5 keV/O2 þ , 44 ) are the real depths because no roughness is observed using these conditions
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2667
with the position of the peak measured at 44 with flooding) for 0.5 keV/O2 þ , 54 with oxygen flooding. The shift suggests a fast variation of the erosion rate at the beginning of the sputtering, which varies quickly at the beginning, then stabilises, until it is equal to the rate measured at 0.5 keV/ O2 þ , 44 . Figs. 5 and 6 show that the erosion rate is different under UHV and using oxygen flooding: the difference is due to the development of roughness under UHV, leading to two different regimes for the erosion speed, as developed at the beginning of this section, the first one corresponding to the development of the altered layer, the second corresponding to the development of the roughness (note that the depth of the transition between both regimes corresponds to the change in the 30 Si16 O signal indicating the development of the roughness). Fig. 5. Shift vs. the apparent depth at 1 keV/ O2 þ , 48 with (A) and without flooding (B).
of analysis (see Section 3.1.2). From this ‘‘true depth’’ of the layers, we determined the apparent shift for the SIMS profiling at 1 keV/O2 þ , 48 . Fig. 5 shows the apparent shift versus the apparent depth with (A) and without flooding (B). The linearity of the curve corresponding to analysis A confirms that erosion rate remains constant with flooding. On the other hand, the curve corresponding to the analysis B is complex. Two different erosion rates are detected before and after 60 nm. This confirms the hypothetical behavior presented at the beginning of this section (Eqs. (3) and (4)): two zones where the behavior is linear, corresponding to the phases where the erosion rate is, respectively, np and nr . From the slopes of theses lines, we can deduce that np =napp ¼ 1:17 and nr =napp ¼ 0:9. Thus, the variation of erosion rate without oxygen flooding is such that np =nr = 1.17, which means that a drop of ’ 30% has occurred. Fig. 6 shows the variation of the peak shift with respect to the ‘‘real’’ value of the peak depth (still measured by comparing
Fig. 6. Measurement of the apparent shift vs. the real depth for 0.5 keV/ O2 þ , 54 with flooding.
3.4. Depth resolution It is well-known that the depth resolution of SIMS is influenced by atomic mixing [26], which induces a profile broadening and a shift from the true position, and by surface roughening [6]. We use the description of the Depth Resolution Function (DRF) of the SIMS analysis proposed by Dowsett et al. [27], now widely used by people who wish to characterise objectively the depth resolution [28,29,2,30,3], and which is described by three independent parameters: lu (the rising exponential part of the profile), s 0 (the Gaussian top) and ld (distance over which the intensity drops by a factor of e, arising from the collisional mixing). The decay length ld of the profiles measured by Ng et al. [5] is 1.2 nm and the broadening parameter s 0 is 0.75 nm using 0.5 keV/O2 þ , 56 with flooding. Takano et al. [31] and Hayashi et al. [32] used a theoretical model, the mixing-roughness-information model (MRI, initially developed by Hofmann [26]) to extract the depth resolution parameters. This model takes into account a parameter s relative to the roughness of the surface and a parameter w which quantifies the collisional mixing. Hayashi et al. [32] showed that the depth resolution degrades when the impact energy is less than 0.5 keV/O2 þ . The parameter s of the MRI model was calculated: s ¼ 3 nm without flooding, 0.5 keV/O2 þ , and 1.4 nm for 1 keV/O2 þ , because of the surface roughening. Takano et al. [31] showed that the best depth resolution is obtained with 1.5 keV/O2 þ at 45 (s ¼ 1:4 nm and w ¼ 1:7 nm) which are relatively high primary energy compared to this work. Moreover, the use of the oxygen flooding has sometimes been found to impede the accurate depth calibration of the SIMS analysis, and does not always lead to the best resolution [6]. Thus, there is a need to clarify this point. In order to better understand the influence of the oxygen flooding on the appearance of the roughness and the depth resolution, we extracted the parameters of the DRF ld, lu and s 0 by fitting the shallower peak, i.e., the peak measured at a depth where the roughness is not present. The values of the parameters are consistent with previous results [17]: lu is very small, lu ¼ 0:7 nm, ld ¼ 0:71 nm and s 0 ¼ 0:3 nm for a primary energy of 0.5 keV/O2 þ and a 44 incidence angle with
2668
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
flooding. Each double d layer is fitted with a function corresponding to the sum of two DRFs, ld and lu being set to the value corresponding to the roughness-free DRF, but the other parameters (position, sigma) remaining free. The fit is excellent most of the time, which means that the profile of the d layers when the roughness is present is fairly well described by a function of the parameters ld , lu and s which is the convolution of the DRF (ld , lu and s 0 ) with a Gaussian which standard deviation is s r , such that s 2 ¼ s 2r þ s 20 . This also shows that the Gaussian describes the induced roughness fairly well. For the d layers located at ’ 135 nm, and for 0.5 keV/O2 þ , 44 with flooding, the result of the fitting is: lu ¼ 0:48 nm, ld ¼ 0:71 nm, s 0 ¼ 0:32 nm and s r ¼ 0:41 nm, the positions z1 and z2 of the double peak are 128.7 and 139.2 nm. For 1 keV/ O2 þ , 48 , UHV: lu ¼ 0:43 nm, ld ¼ 1:75 nm, s 0 ¼ 0:64 nm and s r ¼ 3:99 nm, the positions z1 and z2 of the double peak are 127.3 and 138.0 nm. This leads to an additional broadening at the depth ’ 130 nm: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s r ð0:5 keV=O2 þ ; 130 nmÞ ¼ 0:412 0:322 ¼ 0:25 nm (8) s r ð1 keV=O2 þ ; 130 nmÞ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3:992 0:642 ¼ 3:94 nm
(9)
We plotted the fitted broadening parameter s as function of depth in Fig. 7. It shows that sputtering with oxygen flooding obviously results in smaller values of s. When flooding is used at 0.5 keV/O2 þ , 44 , s remains nearly constant with increasing eroded depth. Under UHV in the same conditions, s increases slightly but remains small (0.48–0.76 nm). We also note that with oxygen flooding at 0.5 keV/O2 þ , 54 , s increases linearly after the third peak, which is the signature of the development of the roughness which appears around 50 nm, in agreement with the AFM results of Fig. 5b. Under UHV (0.5 keV/O2 þ , 54 and 1 keV/O2 þ , 48 ), a strong and linear increase of s is found (from 1.2 to 2.8 nm) and the depth resolution is poor: ld ’ 1:76 nm (measured using the deepest d layer). The depth at which the depth resolution starts
to degrade is 60 nm for 1 keV/O2 þ , 48 . For the same energy with flooding, a slight increase was observed for s (0.51– 1.03 nm). It must be pointed out that measuring s is a very sensitive way to evaluate the roughness without the need for an AFM image. Even a small development of the roughness in the crater bottom can be detected by plotting the evolution of this parameter with depth. The values of the decay length ld and of the Gaussian broadening parameter s, for different impact energies and incident angle with and without oxygen flooding are summarised in Table 1. Relating the measurements of s and ld and the AFM measurements in the crater bottom (Fig. 3a) we can conclude that the distortions of the profiles are directly related to the onset of surface roughening. The profile broadening with depth is far more severe under UHV conditions. This means that oxygen flooding leads to the best depth resolution. In Table 1, the measured peak depths and measured Full Width at Half Maximum (FWHM) of the d layers are summarised. From the literature, Jiang and Alkemade [6] indeed performed a very important work dealing with the roughness at low energy primary bombardment. However, this topic cannot be considered as being entirely understood. As an example, Jiang et al. obtained their best depth resolution in the 4 nm range FWHM in the case of the analysis of d-doped layers. It is possible that such a (bad) apparent resolution originates from a bad quality of the sample obtained by MBE. The depth resolution obtained in our work in the best experimental conditions (500 eV, no roughness, flooding) is 1.5 nm FWHM. 3.5. Relation between s and the topography In this section, we show that the RMS roughness scales with the value of the Gaussian broadening parameter s. Fig. 8a shows a cross-section of the ripple-like topography for 1 keV/ O2 þ , 48 at UHV. According to Table 1, the value of the RMS roughness is 3.85 nm. Then we plot a histogram gathering the effective depths of the cross section with a step of 0.6 nm (Fig. 8b). This histogram is fitted with a Gaussian function: y ¼ y0 þ A eððxxc Þ=ð2seff ÞÞ
2
The best fit is obtained with a value of about 4.17 nm. Finally, we plot the fit of the last peak in Fig. 2a by the depth resolution function DRF [27] (Fig. 8c). The value of the broadening parameter is about 4.1 nm. If the statistical broadening of the profile is equal to the value of the broadening parameter s obtained using 0.5 keV/O2 þ , 44 with flooding, again we have [33]: s 2 ¼ s 2r þ s 20
Fig. 7. The fitted broadening parameter s as a function of depth.
where s is the total broadening of the profile and s r is the broadening due to the roughness. As a result, the statistical distribution of the heights in the crater bottom is Gaussian as a first approximation. A random distribution has to be
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
2669
correlation at such a low scale (see also a paper dealing with SiGe [34]). The Gaussian distribution is then a correct first order description of the phenomena (second order centred moments). 4. Conclusion In this study, we have studied the occurrence of roughness in the SIMS crater bottom when different experimental conditions are used (impact energy, angle of incidence, with and without oxygen flooding) leading to a very high depth resolution. The roughness is always very weak using oxygen flooding compared to UVH in the experimental conditions we explored. Oxygen bombardment without flooding during low energy (1 keV/O2 þ , 48 and 0.5 keV/O2 þ , 54 ) induces surface roughening, distortions of the profiles and erosion rate changes compared with oxygen flooding. At low energy, the profile broadening with depth is more severe under UHV compared with oxygen flooding. Moreover, using AFM images, we show that the RMS roughness is closely related to the value of the Gaussian broadening parameter. This study has shown that oblique incidence O2 þ bombardment with oxygen flooding can be used to obtain accurate, high depth resolution shallow junction profiles in silicon, with almost no roughness in the crater. In our study, the optimal experimental conditions (no surface roughening, optimal depth resolution) seem to be 0.5 keV/O2 þ 44 with flooding. References
Fig. 8. (a) The cross-section of the ripple topography for 1 keV/ O2 þ , 48 , UHV, (b) fit of the last peak using an analytical expression of the DRF convolved with a Gaussian and (c) histogram of the depth in the crater bottom (step: 0.6 nm).
considered, described by a Gaussian which standard deviation is directly correlated to the broadening of the SIMS profiles (s), with the same order of magnitude. This is to our knowledge the first direct demonstration of the role of the roughness on the shape of the profiles in silicon, with a direct quantitative
[1] K. Wittmaak, J Vac. Sci. Technol. B 16 (5) (1998) 2776–2785. [2] B. Gautier, R. Prost, G. Prudon, J.-C. Dupuy, Surf. Interface Anal. 24 (1996) 733–745. [3] B. Gautier, R. Prost, G. Prudon, J.-C. Dupuy, Surf. Interface Anal. 26 (1998) 974–983. [4] B. Fares, B. Gautier, N. Baboux, G. Prudon, P. Holliger, J.-C. Dupuy, Appl. Surf. Sci. 231–232 (2004) 136–140. [5] C. Ng, A. Wee, C. Huan, A. See, Appl. Surf. Sci. (2001) 557–560. [6] Z. Jiang, P. Alkemade, J. Vac. Sci. Technol. B 16 (4) (1998) 1971–1982. [7] R. Liu, C. Ng, A. Wee, Appl. Surf. Sci. 203–204 (2003) 256–259. [8] K. Elst, W. Vandervorst, J. Alay, J. Vac. Sci. Technol. B 11 (6) (1993) 1968–1993. [9] K. Elst, W. Vandervorst, J. Vac. Sci. Technol. B 12 (6) (1994) 3205–3216. [10] R. Liu, A. Wee, D. Shen, H. Takenaka, Surf. Interface Anal. 36 (2004) 172–176. [11] R. Liu, A. Wee, Appl. Surf. Sci. 231–232 (2004) 653–657. [12] K. Wittmaak, S. Corcoran, J. Vac. Sci. Technol. B 16 (1) (1998) 272–279. [13] P. Zalm, C. Vriezema, Nucl. Instrum. Meth. Phys. B 67 (1992) 495–499. [14] J. Erickson, R. Brigham, J. Vac. Sci. Technol. B 14 (1) (1996) 353–357. [15] F. Jahnel, R.V. Criegern, Appl. Surf. Sci. 203–204 (2004) 367–370. [16] P. Alkemade, Z. Jiang, J. Vac. Sci. Technol. B 19 (5) (2001) 1699–1705. [17] N. Baboux, J. Dupuy, G. Prudon, P. Holliger, F. Laugier, A. Papon, J. Hartmann, J. Cryst. Growth 245 (2002) 1–8. [18] E. Cirlin, J. Vajo, T. Hasenberg, J. Vac. Sci. Technol. B 12 (1) (1994) 269– 275. [19] Z. Jiang, P. Alkemade, Appl. Phys. Lett. 73 (3) (1998) 315–317. [20] J. Vajo, R. Doty, E. Cirlin, J. Vac. Sci. Technol. A 14 (5) (1996) 2709– 2720. [21] G. Lau, E. Tok, W. ATS, L. R, L. SL, Surf. Rev. Lett. 8 (5) (2001) 453–457. [22] K. Wittmaack, J. Vac. Sci. Technol. A 8 (3) (1990) 2246–2250. [23] B. Gautier, B. Fares, G. Prudon, J.-C. Dupuy, Appl. Surf. Sci. 131–132 (2004) 136–140.
2670
B. Fares et al. / Applied Surface Science 253 (2006) 2662–2670
[24] K. Wittmaack, J. Vac. Sci. Technol. B 18 (1) (2000) 1–6. [25] C. Tian, W. Vandervorst, J. Vac. Sci. Technol. A 15 (3) (1997) 452– 459. [26] S. Hoffmann, J. Vac. Sci. Technol. A 9 (3) (1991) 1466–1476. [27] M. Dowsett, G. Rowlands, P. Allen, R. Barlow, Surf. Interface Anal. 21 (1994) 310–315. [28] J. Lee, K. Kim, H. Kim, D. Moon, Surf. Interface Anal. 37 (2) (2005) 176– 180. [29] D. Chu, M. Dowsett, Phys. Rev. B 56 (23) (1997) 15167–15170.
[30] B. Gautier, J.-C. Dupuy, R. Prost, G. Prudon, Surf. Interface Anal. 25 (6) (1997) 464–477. [31] A. Takano, Y. Homma, Y. Higashi, Appl, Surf. Sci. 203–204 (2003) 294– 297. [32] S. Hayashi, A. Takano, H. Takenaka, Y. Tomma, Appl. Surf. Sci. 203–204 (2003) 298–301. [33] B. Gautier, G. Prudon, B. Semmache, J.-C. Dupuy, Nucl. Instrum. Meth. Phys. B 142 (1998) 361–376. [34] G. Lau, E. Tok, R. Liu, Nucl. Instrum. Meth. Phys. B 215 (2004) 76–82.