- Email: [email protected]
Hydrogen in amorphous Si and Ge during solid phase epitaxy
Thin Solid Films 518 (2010) 2317–2322
Contents lists available at ScienceDirect
Thin Solid Films j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / t s f
Hydrogen in amorphous Si and Ge during solid phase epitaxy B.C. Johnson a,⁎, P. Caradonna a, D.J. Pyke a, J.C. McCallum a, P. Gortmaker b a b
School of Physics, University of Melbourne, Victoria 3010, Australia Department of Electronic Materials Engineering, The Australian National University, A.C.T. 0200, Australia
a r t i c l e
i n f o
Available online 20 October 2009 Keywords: Solid phase epitaxy Hydrogen diffusion Device modeling
a b s t r a c t Studies into the effect of hydrogen on the kinetics of solid phase epitaxy (SPE) in amorphous Si (a-Si) and Ge (a-Ge) are presented. During SPE, H diffuses into surface amorphous layers from the surface and segregates at the crystalline–amorphous interface. Some of the H crosses the interface and diffuses into the crystalline material where it either leaves the sample or is trapped by defects. H segregation at concentrations up to 2.3 × 1020 H/cm3 is observed in buried pha-Si layers with the SPE rate decreasing by up to 20%. H also results in a reduction of dopant-enhanced SPE rates and is used to explain the asymmetry effects between the SPE velocity profile and the dopant concentration profile observed with shallow dopant implants. Conversely, H diffusion is enhanced by dopants in a-Si. These studies suggest that H diffusion and SPE may be mediated by the same defect. The extent of H in-diffusion into a-Ge surface layers during SPE is about one order of magnitude less that that observed for a-Si layers. This is thought to be due to the lack of a stable surface oxide on a-Ge. However, a considerably greater retarding effect on the SPE rate in a-Ge of up to 70% is observed. A single unifying model is applied to both dopant-enhanced SPE and H diffusion processes. © 2009 Elsevier B.V. All rights reserved.
1. Introduction In the fabrication of a broad range of Si and Ge complementary metal-oxide-semiconductor (CMOS) devices an amorphous layer may form through the implantation of heavy ions with keV energies. Alternatively, self-amorphization implants may be used before implantation of lighter ions in order to avoid implantation channeling. Regrowth of the crystal layer via solid phase epitaxy (SPE) has been identified as a pathway for achieving high dopant activation with a low thermal budget [1]. Stringent demands are placed on device fabrication modeling where devices must be made efficiently in order to meet the requirements of future technology nodes. Accordingly, an extensive SPE literature exists (for comprehensive reviews see Refs. [2–8]). The velocity of the c–a interface during SPE has a strong dependence on many parameters, all of which need to be known for process modeling to be accurate. These parameters include the substrate crystallographic orientation [9], pressure [10], and the presence of dopants [6]. SPE is also thermally activated, the c–a interface velocity being described by an Arrhenius-type equation of the form, v = vo expð−Ea = kTÞ
ð1Þ
where vo and Ea are the pre-exponential factor and activation energy, respectively. For SPE in Si, vo = 4.64 × 108 cm/s and Ea = 2.7 eV has
⁎ Corresponding author. E-mail address: [email protected] (B.C. Johnson). 0040-6090/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.tsf.2009.09.145
been determined up to the melting point [3]. The corresponding Ge values are vo = 2.6 × 109 cm/s and Ea = 2.15 eV determined in a temperature range of 300–540 °C [7]. The SPE rate is often unknowingly retarded by the presence of non-doping impurities such as hydrogen [4]. Hydrogen is an interesting case as it is often present during SPE unless special steps are taken. The behaviour of H during SPE then will also need to be understood and incorporated into fabrication models. During thermal treatments on surface a-Si layers, H diffuses from the native oxide into the layer. Once the H meets the c–a interface it strongly segregates in the amorphous phase and retards the SPE rate by up to ~50% [4]. This in-diffusion occurs whenever there is water vapor in the ambient or a surface oxide present. For thin a-Si layers (b400 nm), such as those produced during shallow junction processing, a nearly constant concentration of ~ 2 ×109 H/cm is expected at the c–a interface throughout the SPE process [4]. Amorphous layers formed by cluster implantation of decaborane (B10H14) [11] also contain H and therefore may affect the SPE rate. For thick a-Si layers, H can infiltrate several microns into the a-Si layer before meeting the c–a interface. The SPE rate is found to decrease linearly with H concentration up to [H] ≈ 3 × 1019 cm− 3. For greater concentrations up to 7 × 1019 cm− 3 the SPE rate depends only weakly on the H concentration. This threshold value has been correlated with the density of dangling bonds (DB) in a-Si formed by ion implantation and has been cited as evidence for the possible involvement of DBs in the SPE process [4,12]. Hydrogen diffusion is also thermally activated and for low H concentrations ([H]) has been described by an equation similar to
2318
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
Eq. (1) with v and vo replaced by a pre-exponential factor, Do = 2.2 × 104 cm2/s and an activation energy Ea = 2.7 eV [13]. Since a common activation energy is shared by SPE and H diffusion, one is led to suspect that the rate limiting step in these two seemingly unrelated processes may be the same [13]. Hence, studies into H diffusion and its role during SPE can also provide important insight into the SPE mechanism. In this paper, experiments concerning the role of H during SPE are presented. Firstly, H implantation into buried a-Si layers is used to study the SPE retardation at H concentrations greater than 7 × 1019 cm− 3. H is also shown to have a temperature dependent segregation coefficient. The interaction of H with dopants is also observed through SPE and H diffusion studies presented here. The presence of dopants can enhance the diffusion of both H in-diffusing from the surface and implanted H. The SPE rates in a-Ge are also shown to suffer from H contamination. These observations are drawn together into a single model, the generalized Fermi level shifting model, which links structural changes at the interface or H diffusion to the local electronic structure [6,10]. The impact of these results on device modeling for fabrication processes is considered. 2. Theoretical background The generalized Fermi level shifting (GFLS) model is used in this work to describe the effect of H on SPE and dopant-enhanced SPE and H diffusion. The GFLS model has been successful in describing the dopant dependence of SPE in both a-Si and a-Ge [6,7,10]. The effect of pressure has also been recently incorporated into the model [14]. According to the model, the number of events, R, is determined by the concentration of neutral defects, X 0 and its positively or negatively charged counterparts, X±. The events could be either the transport of H from one site to another or the rearrangement of an atom at the c–a interface in SPE. It is assumed that the defects are in thermal and electronic equilibrium determined by the local band structure density of states. The event is then expected to be proportional to the concentration of these defects. For an n-type semiconductor and its intrinsic counterpart the rate of the event is given by R = αð½X 0 + ½X − j doped Þ
ð2aÞ
and Ri = αð½X 0 + ½X − j intrinsic Þ
ð2bÞ
respectively, where α is a constant and the square brackets denote a concentration. The charged fraction of defects is determined by Fermi–Dirac statistics and, for an n-type semiconductor, is expressed as the ratio of charged to neutral defect concentrations, ½X − j doped ½X 0
E −E− − = g exp f kT
ð3Þ
where Ef is the Fermi level and E− represents the energy level within the band gap of the defect X−. If a DB defect is responsible for the SPE process then it is expected that the degeneracy is g− = 1/2 and g+ = 1 for n-type and p-type semiconductors, respectively [6]. Once Eq. (3) is substituted into Eq. (2) we obtain, ½X − E −E− − 1 + 0 j doped 1 + g exp f R ½X kT : = = E −E− ½X − Ri 1 + g − exp fi 1 + 0 j intrinsic ½X
ð4Þ
kT
This equation is used to fit the normalized SPE rates (v/vi) or H diffusion coefficient (D/Di) data as a function of temperature with g and E− being free parameters.
We now consider the effect of H at the c–a interface. A number of studies have suggested that H retards SPE through the passivation of DBs at the c–a interface thus reducing the number of crystallization sites available [4,12]. According to the GFLS model, the defect concentration would become [Xh0] = [X 0] − a[H]. The charged defect is reduced by an amount b[H] where a and b represent the fraction of H that passivates the neutral and charged defects, respectively. Eq. (4) then becomes
v = vi
2 − 3 1 + ½Xh0 a½H 6 ½Xh 7 1− 0 4 − 5 ½X 1 + ½X 0
ð5Þ
½X
where v and vi are the dopant/H-affected and intrinsic SPE velocities, respectively. The term in square brackets is the normalized rate enhancement due to doping. The first term in parentheses has a linear trend with [H]. Indeed, the SPE retardation is observed to have a linear dependence on [H] up to a concentration of 3 × 1019 H/cm3 where the normalized SPE rate is about 0.5 [4]. For the case without dopants (Ef = Ei), the dopant term is expected to be close to unity since [X−]/[X 0] ≈ 0 determined from Eq. (3). Therefore, the factor [X 0]/α ≈ 6 × 1019 H/cm3. This is close to the value of 1.5 × 1020 H/cm3 reported by Oberlin et al. [12]. Furthermore, H is found to reduce the pre-exponential factor in Eq. (1) but not the activation energy, suggesting that H passivates crystallization sites while not affecting the energy associated with a crystallization event [7,15]. Various studies are described in the following sections to test this model and develop a more complete understanding of the role of H during SPE. 3. Experiment 3.1. Sample preparation The kinetics of SPE were measured in amorphous layers formed by self-ion implantation into b100N wafers. A National Electrostatics Corp. 1.7 MV tandem accelerator was used for all implants. The samples were tilted 7° off the incident beam axis to avoid channeling and also rotated about the surface normal by a similar amount to prevent any remaining possibility of planar channeling [16]. Substrates were affixed with Ag paste to the implanter stage, which was held at 77 K. Good thermal contact was especially important for a-Ge formation in order to avoid the porous structure that forms in highfluence room temperature implanted Ge [17,18]. Multiple ion beam energies were chosen to form homogeneous amorphous layers ranging between 1.5 and 4.2 μm thick. 3.2. Time resolved reflectivity The SPE rates of the c–a interface were determined from time resolved reflectivity (TRR) measurements by acquiring reflectivity data simultaneously using two HeNe lasers at wavelengths of λ = 1.152 μm to probe the thin amorphous layers and at λ = 1.523 μm for the thicker layers. As the c–a interface moves through the sample, peaks in the TRR reflectivity trace occur every λ/2n. By combining the measured TRR traces and a theoretical reflectivity versus c–a interface position curve the velocity of the interface was determined. These data were collected while the samples were held on a resistively heated vacuum chuck and annealed in air over the temperature range of 460–660 °C for Si and 300–540 °C for Ge. The temperature of the samples during the anneals was calibrated by comparing the reading of a type-K thermocouple embedded in the sample stage with the melting points of various suitably encapsulated metal films evaporated onto Si wafers. The error associated with the temperature reading was found to be ±1 °C. Measurements were performed in air to match the experimental conditions of other studies and so that the effects of H infiltration could
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
be examined and quantified. Further details on the experimental apparatus are presented elsewhere [19]. 3.3. Secondary ion mass spectrometry Secondary ion mass spectroscopy (SIMS) was performed on selected samples to measure the H concentration profile in the amorphous layer after a partial anneal. A Cameca IMS-5F secondary ion mass spectrometer equipped with a Cs ion source. A primary ion beam of Cs+ with a net impact energy of 14.5 keV was applied for the depth profiling. The Cs+ beam was rastered over an area of 250 μm2. Negative secondary ions were collected from the central part of the rastered region (55 μm in diameter) in order to avoid crater side wall effects. The SIMS detection limits were 2 × 1017 cm–3 and 5 × 1015 cm–3 for H and B, respectively. The depth scale was quantified by a stylus profilometer with an accuracy of within 2%. 4. Results and discussion 4.1. Hydrogen retardation of SPE Fig. 1a) and b) are schematics of the expected H concentration ([H]) at two different times during an anneal of a surface a-Si layer at 660 °C. In Fig. 1a) the c–a interface has moved 0.25 μm from its original position. During this time H has infiltrated into the amorphous layer from the surface oxide. In Fig. 1b), the H has reached the c–a interface and causes the SPE rate to decrease while being strongly segregated on the amorphous side of the interface. During segregation a fraction of the H passes over to the crystalline side of the interface where it rapidly diffuses until trapped at defect sites as denoted by the Gaussian distribution around the original c–a interface. Fig. 1c) shows the corresponding SPE rate as a function of the interface depth as determined by TRR. Between depths of 1.2 and 2.2 μm the SPE rate is the intrinsic value which can be determined with Eq. (1) above. Once the H reaches the c–a interface, the rate decreases. To study this decrease beyond the H concentrations previously considered, H was implanted into buried a-Si layers. The buried layers
Fig. 1. Schematic of an a-Si surface layer with the expected [H] profiles after a partial anneal for a) 50 s and b) 140 s at 660 °C. The interface velocities, va and vb are the intrinsic and H-retarded values, respectively. The dashed line shows the position of the original a–c interface. c) Corresponding SPE rate with the velocities, va and vb indicated.
2319
were formed by two Si implants at 0.6 and 2 MeV while the substrates were held at −10 °C. This implant temperature ensures that a surface crystalline capping remains after implantation which inhibits H infiltration from the surface into the amorphous layer to a concentration below at least 1017 cm− 3[19]. Fig. 2a shows the evolution of the implanted H profile after partial anneals at 580 °C for 400, 2550 and 5400 s. The original c–a interfaces of the buried layer are indicated by the dashed lines at 0.2 and 2.2 μm. After the first partial anneal the H profile decreases in concentration and broadens. As expected, H then segregates at the c–a interfaces which refines the profile as SPE proceeds. A fraction of the implanted H diffuses out of the amorphous layer and is gettered by defects which have formed around the original position of both the front and back interfaces. The H at these positions increases with annealing time with [H] in the near-surface region increasing at a greater rate. The original front interface position acts as a sink for both the implanted H escaping from the amorphous region and H diffusing in from the surface. This is especially clear for the 400 s anneal case. The implanted H has not yet reached the interface as shown by the lack of segregation but a relatively high [H] value is present in the crystalline capping layer. This could be due to the presence of implantation induced damage in the capping layer and could aid in inhibiting H diffusion into the amorphous layer. The refined [H] profile grows to a concentration similar to that of the original implanted concentration. Initially, it was hoped that the refinement of the [H] profile by the c–a interface would lead to a more efficient way to form a high density of H-related defects as used in the SmartCut™ technique where H is implanted into c-Si [20]. In this technique, H clusters together to form bubbles with an anneal at around 800 °C. However, too much H is lost
Fig. 2. a) SIMS of the implanted [H] profile in buried a-Si layers after partial anneals of 400, 2550 and 5400 s at 580 °C. b) The dependence of the SPE rate on the [H] determined from a). Data from Roth et al. is also shown [13].
2320
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
through the interfaces during the SPE anneal to successfully form bubbles using this method [21]. Such a study, however, did allow the [H] dependence of the SPE rate to be studied to [H] values up to 2.3 × 1020 cm− 3. Fig. 2b) shows the retardation of the SPE rate as a function of [H] determined from Fig. 2a). This data compares well with previous studies by Roth et al. [4]. Below [H]=3 × 1019 cm− 3 the normalized SPE rate has a linear dependence on the [H] as predicted by Eq. (5). Above 3 × 1019 cm− 3 the SPE rate has a weak dependence on [H] and continues to decrease to 20% of the intrinsic value at a concentration of 2.3 × 1020 cm− 3. 4.2. Hydrogen segregation As shown in Fig. 2, H segregates at the interface implying a segregation coefficient, k, of less than one. SPE rate profiles such as that shown in Fig. 1 can be used to study the segregation of H further since the dependence of SPE retardation on [H] is known. Fig. 3 shows the normalized SPE rate for surface a-Si layers for anneal temperatures between 500–640 °C. The retardation is linearly proportional to [H] as determined in Fig. 2b) and is indicated on the right axis. The SPE retardation has a strong temperature dependence with the greatest retardation occurring for the lowest anneal temperatures. This can be explained by a temperature dependent segregation coefficient. For example, during the 640 °C anneal, H more readily crosses the c–a interface resulting in less segregation and therefore less SPE retardation. Such a parameter is extremely important if the extent of retardation for a particular SPE anneal is to be determined during device fabrication. A numerical model is currently being developed to study this effect. 4.3. Hydrogen-dopant interactions It is well-known that dopants enhance the SPE rate in both a-Si and a-Ge [6,7]. The behaviour and effect of H during the SPE anneals required to activate dopants is not so well-known. Fig. 4 summarizes a recent experiment involving the implantation of B or P into the nearsurface region of a 2.2 μm thick surface a-Si layer [22]. Fig. 4a) shows the B concentration profile measured by SIMS. The B profile was formed by multiple energy B implantations in order to broaden the profile. Overlaid are two calculated H profiles for a 560 °C anneal for 90 min. To schematically illustrate the in-diffusion of H, simulated profiles are also included. The dashed line is the H profile in an intrinsic sample whereas the solid line is the H profile in the B implanted sample. Segregation was not considered in the calculation and it was assumed that B enhances the H diffusion coefficient by a factor of 20. This factor was determined from Eq. (4) and g and E−
Fig. 3. Normalized SPE rate for surface a-Si layers showing H retardation. [H] at the c–a interface on the right axis is calculated using Eq. (5) with [X0]/a = 6 × 1019 H/cm3.
Fig. 4. a) Implanted B profile in a-Si determined by SIMS and numerically calculated H concentration profiles for H diffusion from the surface. The B was assumed to enhance the H diffusion coefficient by a factor of 20. b) SIMS concentration profiles of H in intrinsic (circles), B implanted (triangles) and P implanted (squares) a-Si layers after an anneal of about 90 min at 560 °C. The resulting c–a interface depth is 1.2 μm.
values discussed in Section 4.4. Such values could not be extracted from this data due to insufficient knowledge of segregation. The H is transported through the B implanted region at a much faster rate than in the intrinsic case leading to an increased [H] value in the amorphous layer beyond the B implanted region. Fig. 4b shows the SIMS profiles for partially annealed intrinsic, B implanted and P implanted samples and shows that dopant-enhanced H diffusion is indeed occurring. The peak P concentration was greater than the peak B concentration at 5 × 1020/cm3 however the H in-diffusion is not as significant. This behaviour is similar to that found in SPE data where B enhancement is much greater than P enhancement [6]. The [H] profile around the dopant implanted region has a complex shape which is attributed to a matrix type effect of the SIMS measurement but beyond the dopant region [H] is clearly enhanced. The normalized SPE rates at a depth of 1.2 μm for these samples are shown in Fig. 5. The SPE rates corresponding to the SIMS profiles in Fig. 4b are also plotted in Fig. 2b and agree well with Roth et al. [13] The retarded rates from each sample decrease linearly as a function of 1/kT due to the temperature dependence of the segregation coefficient, k, discussed above and observed in Fig. 3. In intrinsic a-Si, we see a 20% reduction of the normalized SPE rate over the temperature range studied. The SPE retardation for the dopant implanted layers is much greater than in the intrinsic case due to the enhancement of [H]. The B implanted layers show the greatest retardation. The H-retarded SPE rates for a layer containing both B and P are also shown. These rates are much less than the B implanted sample suggesting that P compensates the doping. This results in an H diffusion coefficient that is closer to the intrinsic value.
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
Fig. 5. Normalized SPE rates determined at a depth of 1.2 μm in intrinsic (circles), B (triangles), P (squares) and compensated implanted Si layers. The solid lines are linear fits. Errors are shown on the intrinsic data only for clarity.
The dopant-enhanced SPE rates are also affected by the H. The Benhanced SPE rates were 63–47% lower in the temperature range 500–640 °C than the expected enhanced rates in H-free material [6,22]. This is a dramatic effect and fabrication codes will need to take this into account. Eq. (5) can be used in this situation to predict the SPE rate but only if the value of [H] at the interface and the passivation behaviour of H at the interface is known. This then becomes quite a complex problem with many interdependences. However, Eq. (5) in the context of the so-called asymmetric effect offers some insight into this interdependence as discussed below [3]. In previous studies an asymmetry between the dopant-enhanced SPE rate and the arsenic concentration profile has been reported. This can also be understood in terms of the infiltration of H from the surface into the amorphous layer [23,24]. By a comparison between surface and buried amorphous layers the effect of H infiltration on the SPE kinetics can be discerned. It has been found that for the nearsurface side of the As-enhanced SPE profile in surface a-Si layers SPE is retarded by the infiltrating H causing an apparent asymmetry between the enhanced SPE rate and the As profile. The SPE rate profile in buried a-Si layers does not show any asymmetry. This asymmetry was also modeled using simple H in-diffusion calculations and Eq. (5) [24]. The model showed good agreement with the observed rates if it was assumed that both donors and SPE defects (DBs) were passivated by H. The As is expected to enhance the H diffusion coefficient but this was not considered in this earlier work. Similar asymmetries have been observed in both B and P implanted surface a-Si layers [25]. We suggest that the detailed kinetics of dopantenhanced H diffusion need to be considered in order to fully account for the observed effects.
2321
Fig. 6. [H] profiles in a-Si after an anneal of 455 °C for 5 min in intrinsic (circles) and in samples implanted with 1 × 1020 B/cm3 (crosses) and 1.5 × 1020 B/cm3 (diamonds). The B profile determined by SIMS is also shown.
normalized H diffusion coefficients are fit well with Eq. (2) when E+ − Ev = (0.44 ± 0.06) eV and g+ = (0.1 ± 0.1) for a [B] = 1 × 1020 B/cm3 is used. We note that the value of g is much smaller than expected and a value of zero also lies within its error bounds. We believe this to arise from the difficulty in accurately calculating the Fermi level in a-Si. However, Fermi level shifting in a-Si is plausible once the dopant concentration exceeds the midgap defect concentration [26]. The ability of the model to fit H diffusion data well supports the proposition that both SPE and H diffusion are governed by the same defect. Furthermore, a single model can be used to describe both processes during thermal anneals. 4.5. Hydrogen in a-Ge during SPE In a-Ge, H has also been shown to affect the SPE rates but the behaviour is quite different to that observed in a-Si [7]. For ~3.25 μm thick surface a-Ge layers no large scale velocity reductions are observed until the final ~ 0.4 μm of regrowth. This had previously been reported by a number of workers who speculated that the retardation was either due to surface impurities driven into the sample during amorphization implants or H in-diffusion [4,10]. Fig. 7 shows [H] in partially annealed a-Ge layers. In the asimplanted state, H was found only in the near-surface region. With an anneal of 121 s at 420 °C the H is shown to diffuse into the sample but has not reached the c–a interface which is at a depth of 0.79 μm. In a-
4.4. Dopant-enhanced H diffusion To determine the value of [H] at the c–a interface during SPE the dependence of H diffusion on dopant concentration needs to be determined. To study this, multiple B implants were used to form a constant concentration profile over which a single H profile was implanted. Fig. 6 shows this [H] profile after an anneal for 5 min at 455 °C in intrinsic and B implanted samples. At the surface, the in-diffusion of H is similar for each sample. When B is present, the H profiles are much broader signifying B-enhanced H diffusion. The solid curves were determined by solving the diffusion equation numerically using an implicit finite difference algorithm (to be published elsewhere). The
Fig. 7. SIMS profiles of the hydrogen content in thick a-Ge samples: in the as-implanted state (○), and for partial anneals at 420 °C for durations of 121 s (△) and 227 s (◇). The expected c–a interface depths for each of these partial anneals based on the TRR data were 0.79 μm and 0.24 μm, respectively.
2322
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
Si the H would have segregated at the interface by this stage of regrowth [15]. After a longer anneal of 227 s, the H has come into contact with the interface at 0.24 μm and shows strong segregation (again k b 1 as for the Si system). The segregated [H] value is of a similar order to that observed in a-Si [15]. The in-diffusion of H in a-Ge is about one order of magnitude less than that observed in a-Si. This may be due to the inability of a-Ge to form a stable native oxide. Implanted H profiles have also been studied in a-Ge layers [7]. A saturation of the SPE rate retardation was observed for concentrations between 6 × 1018 and 6 × 1019 H/cm3. This agrees with the value of 3 × 1019 H/cm3 observed for a-Si. The SPE rate around this saturation level is a factor of ~3.7 times slower than the intrinsic rate, as compared to the factor of ~ 2 for a-Si. 5. Conclusions The effect of H on the SPE rate is complex. The H infiltrates from the surface into the amorphous surface layer during thermal processing. The H diffusion coefficient is dependent not only on the temperature but also the dopant concentration in the amorphous layer. Once H meets the c–a interface it strongly segregates on the amorphous side of the interface. The segregation coefficient is found to have a temperature dependence. The in-diffusion of H and its dopant-enhancement can be used to explain the so-called asymmetric effects of the SPE velocity profiles with the implanted dopant profiles. The infiltration of H into a-Ge is also observed but to a much smaller extent than in a-Si. For H implanted a-Ge layers the Hretarded SPE rate decreases by up to a factor of 3.7. The generalized Fermi level shifting model provides a convenient and accurate way of describing these effects under a single framework. In the model, both H diffusion and SPE occur via DBs. H affects SPE by passivating the DBs at the c–a interface. When dopants are also present H tends to passivate both the DB and the dopant atom which would otherwise take part in Fermi level shifting. Dopant and H effects are important parameters to consider in device modeling.
Acknowledgment The Department of Electronic Materials Engineering at the Australian National University is acknowledged for their support by providing access to SIMS and ion implanting facilities. This work was supported by a grant from the Australian Research Council. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26]
International Technology Roadmap for Semiconductors (2003 Edition). J.W. Mayer, L. Eriksson, S.T. Picraux, J.A. Davies, Can. J. Phys. 46 (1968) 663. G.L. Olson, J.A. Roth, Mat. Sci. Rep. 3 (1988) 1. G. Olson, J. Roth, in: D. Hurle (Ed.), Handbook of Crystal Growth, vol. 3, Elsevier Science B.V, 1994, p. 255, chap. 7. O. Hellman, Mater. Sci. Eng. R16 (1996) 1. B.C. Johnson, J.C. McCallum, Phys. Rev. B. 76 (2007) 045216. B.C. Johnson, P. Gortmaker, J.C. McCallum, Phys. Rev. B 77 (2008) 214109. N.G. Rudawski, K.S. Jones, S. Morarka, M.E. Law, R.G. Elliman, J. Appl. Phys. 105 (2009) 081101. L. Csepregi, E.F. Kennedy, J.W. Mayer, T.W. Sigmon, J. Appl. Phys. 49 (1978) 3906. G.-Q. Lu, E. Nygren, M.J. Aziz, J. Appl. Phys. 70 (1991) 5323. J. Matsuo, D. Takeuchi, T. Aoki, I. Yamada, Ion Implantation Technology. Proceedings of the 11th International Conference, 1996, p. 768. J.C. Oberlin, A.C. Chami, E. Ligeon, C. Prunier, Nucl. Instr. Meth. B 19/20 (1987) 462. J.A. Roth, G.L. Olson, D.C. Jacobson, J.M. Poate, Mat. Res. Soc. Symp. Proc. 297 (1993) 291. D. D'Angelo, L. Romano, I. Crupi, E. Carria, V. Privitera, M.G. Grimaldi, Appl. Phys. Lett. 93 (2008) 231901. J.A. Roth, G.L. Olson, D.C. Jacobson, J.M. Poate, Kinetics of Phase Transformations, MRS Symposia Proceedings, vol. 205, Materials Research Society, Pittsburgh, 1992. A. Armigliato, R. Nipoti, G. Bentini, M. Mazzone, M. Bianconi, A.N. Larsen, A. Gasparotto, Mater. Sci. Eng. B2 (1989) 63. B.R. Appleton, O.W. Holland, J. Narayan, O.E.S. III, J.S. Williams, K.T. Short, E. Lawson, Appl. Phys. Lett. 41 (1982) 711. O. Holland, B. Appleton, J. Narayan, J. Appl. Phys. 54 (1983) 2295. J.C. McCallum, Appl. Phys. Lett. 69 (1996) 925. M. Bruel, Electron. Lett. 31 (1995) 1201. D. J. Pyke, Master's thesis, University of Melbourne (2006). B.C. Johnson, P. Caradonna, J.C. McCallum, Mater. Sci. Eng. B 157 (2009) 6. B.C. Johnson, J.C. McCallum, J. Appl. Phys. 95 (2004) 4427. B.C. Johnson, J.C. McCallum, J. Appl. Phys. 96 (2004) 2381. R.M. Walser, Y.-J. Jeon, Mater. Res. Soc. Symp. Proc. 205 (1992) 27. S. Coffa, J.M. Poate, Appl. Phys. Lett. 63 (1993) 1080.
Contents lists available at ScienceDirect
Thin Solid Films j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / t s f
Hydrogen in amorphous Si and Ge during solid phase epitaxy B.C. Johnson a,⁎, P. Caradonna a, D.J. Pyke a, J.C. McCallum a, P. Gortmaker b a b
School of Physics, University of Melbourne, Victoria 3010, Australia Department of Electronic Materials Engineering, The Australian National University, A.C.T. 0200, Australia
a r t i c l e
i n f o
Available online 20 October 2009 Keywords: Solid phase epitaxy Hydrogen diffusion Device modeling
a b s t r a c t Studies into the effect of hydrogen on the kinetics of solid phase epitaxy (SPE) in amorphous Si (a-Si) and Ge (a-Ge) are presented. During SPE, H diffuses into surface amorphous layers from the surface and segregates at the crystalline–amorphous interface. Some of the H crosses the interface and diffuses into the crystalline material where it either leaves the sample or is trapped by defects. H segregation at concentrations up to 2.3 × 1020 H/cm3 is observed in buried pha-Si layers with the SPE rate decreasing by up to 20%. H also results in a reduction of dopant-enhanced SPE rates and is used to explain the asymmetry effects between the SPE velocity profile and the dopant concentration profile observed with shallow dopant implants. Conversely, H diffusion is enhanced by dopants in a-Si. These studies suggest that H diffusion and SPE may be mediated by the same defect. The extent of H in-diffusion into a-Ge surface layers during SPE is about one order of magnitude less that that observed for a-Si layers. This is thought to be due to the lack of a stable surface oxide on a-Ge. However, a considerably greater retarding effect on the SPE rate in a-Ge of up to 70% is observed. A single unifying model is applied to both dopant-enhanced SPE and H diffusion processes. © 2009 Elsevier B.V. All rights reserved.
1. Introduction In the fabrication of a broad range of Si and Ge complementary metal-oxide-semiconductor (CMOS) devices an amorphous layer may form through the implantation of heavy ions with keV energies. Alternatively, self-amorphization implants may be used before implantation of lighter ions in order to avoid implantation channeling. Regrowth of the crystal layer via solid phase epitaxy (SPE) has been identified as a pathway for achieving high dopant activation with a low thermal budget [1]. Stringent demands are placed on device fabrication modeling where devices must be made efficiently in order to meet the requirements of future technology nodes. Accordingly, an extensive SPE literature exists (for comprehensive reviews see Refs. [2–8]). The velocity of the c–a interface during SPE has a strong dependence on many parameters, all of which need to be known for process modeling to be accurate. These parameters include the substrate crystallographic orientation [9], pressure [10], and the presence of dopants [6]. SPE is also thermally activated, the c–a interface velocity being described by an Arrhenius-type equation of the form, v = vo expð−Ea = kTÞ
ð1Þ
where vo and Ea are the pre-exponential factor and activation energy, respectively. For SPE in Si, vo = 4.64 × 108 cm/s and Ea = 2.7 eV has
⁎ Corresponding author. E-mail address: [email protected] (B.C. Johnson). 0040-6090/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.tsf.2009.09.145
been determined up to the melting point [3]. The corresponding Ge values are vo = 2.6 × 109 cm/s and Ea = 2.15 eV determined in a temperature range of 300–540 °C [7]. The SPE rate is often unknowingly retarded by the presence of non-doping impurities such as hydrogen [4]. Hydrogen is an interesting case as it is often present during SPE unless special steps are taken. The behaviour of H during SPE then will also need to be understood and incorporated into fabrication models. During thermal treatments on surface a-Si layers, H diffuses from the native oxide into the layer. Once the H meets the c–a interface it strongly segregates in the amorphous phase and retards the SPE rate by up to ~50% [4]. This in-diffusion occurs whenever there is water vapor in the ambient or a surface oxide present. For thin a-Si layers (b400 nm), such as those produced during shallow junction processing, a nearly constant concentration of ~ 2 ×109 H/cm is expected at the c–a interface throughout the SPE process [4]. Amorphous layers formed by cluster implantation of decaborane (B10H14) [11] also contain H and therefore may affect the SPE rate. For thick a-Si layers, H can infiltrate several microns into the a-Si layer before meeting the c–a interface. The SPE rate is found to decrease linearly with H concentration up to [H] ≈ 3 × 1019 cm− 3. For greater concentrations up to 7 × 1019 cm− 3 the SPE rate depends only weakly on the H concentration. This threshold value has been correlated with the density of dangling bonds (DB) in a-Si formed by ion implantation and has been cited as evidence for the possible involvement of DBs in the SPE process [4,12]. Hydrogen diffusion is also thermally activated and for low H concentrations ([H]) has been described by an equation similar to
2318
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
Eq. (1) with v and vo replaced by a pre-exponential factor, Do = 2.2 × 104 cm2/s and an activation energy Ea = 2.7 eV [13]. Since a common activation energy is shared by SPE and H diffusion, one is led to suspect that the rate limiting step in these two seemingly unrelated processes may be the same [13]. Hence, studies into H diffusion and its role during SPE can also provide important insight into the SPE mechanism. In this paper, experiments concerning the role of H during SPE are presented. Firstly, H implantation into buried a-Si layers is used to study the SPE retardation at H concentrations greater than 7 × 1019 cm− 3. H is also shown to have a temperature dependent segregation coefficient. The interaction of H with dopants is also observed through SPE and H diffusion studies presented here. The presence of dopants can enhance the diffusion of both H in-diffusing from the surface and implanted H. The SPE rates in a-Ge are also shown to suffer from H contamination. These observations are drawn together into a single model, the generalized Fermi level shifting model, which links structural changes at the interface or H diffusion to the local electronic structure [6,10]. The impact of these results on device modeling for fabrication processes is considered. 2. Theoretical background The generalized Fermi level shifting (GFLS) model is used in this work to describe the effect of H on SPE and dopant-enhanced SPE and H diffusion. The GFLS model has been successful in describing the dopant dependence of SPE in both a-Si and a-Ge [6,7,10]. The effect of pressure has also been recently incorporated into the model [14]. According to the model, the number of events, R, is determined by the concentration of neutral defects, X 0 and its positively or negatively charged counterparts, X±. The events could be either the transport of H from one site to another or the rearrangement of an atom at the c–a interface in SPE. It is assumed that the defects are in thermal and electronic equilibrium determined by the local band structure density of states. The event is then expected to be proportional to the concentration of these defects. For an n-type semiconductor and its intrinsic counterpart the rate of the event is given by R = αð½X 0 + ½X − j doped Þ
ð2aÞ
and Ri = αð½X 0 + ½X − j intrinsic Þ
ð2bÞ
respectively, where α is a constant and the square brackets denote a concentration. The charged fraction of defects is determined by Fermi–Dirac statistics and, for an n-type semiconductor, is expressed as the ratio of charged to neutral defect concentrations, ½X − j doped ½X 0
E −E− − = g exp f kT
ð3Þ
where Ef is the Fermi level and E− represents the energy level within the band gap of the defect X−. If a DB defect is responsible for the SPE process then it is expected that the degeneracy is g− = 1/2 and g+ = 1 for n-type and p-type semiconductors, respectively [6]. Once Eq. (3) is substituted into Eq. (2) we obtain, ½X − E −E− − 1 + 0 j doped 1 + g exp f R ½X kT : = = E −E− ½X − Ri 1 + g − exp fi 1 + 0 j intrinsic ½X
ð4Þ
kT
This equation is used to fit the normalized SPE rates (v/vi) or H diffusion coefficient (D/Di) data as a function of temperature with g and E− being free parameters.
We now consider the effect of H at the c–a interface. A number of studies have suggested that H retards SPE through the passivation of DBs at the c–a interface thus reducing the number of crystallization sites available [4,12]. According to the GFLS model, the defect concentration would become [Xh0] = [X 0] − a[H]. The charged defect is reduced by an amount b[H] where a and b represent the fraction of H that passivates the neutral and charged defects, respectively. Eq. (4) then becomes
v = vi
2 − 3 1 + ½Xh0 a½H 6 ½Xh 7 1− 0 4 − 5 ½X 1 + ½X 0
ð5Þ
½X
where v and vi are the dopant/H-affected and intrinsic SPE velocities, respectively. The term in square brackets is the normalized rate enhancement due to doping. The first term in parentheses has a linear trend with [H]. Indeed, the SPE retardation is observed to have a linear dependence on [H] up to a concentration of 3 × 1019 H/cm3 where the normalized SPE rate is about 0.5 [4]. For the case without dopants (Ef = Ei), the dopant term is expected to be close to unity since [X−]/[X 0] ≈ 0 determined from Eq. (3). Therefore, the factor [X 0]/α ≈ 6 × 1019 H/cm3. This is close to the value of 1.5 × 1020 H/cm3 reported by Oberlin et al. [12]. Furthermore, H is found to reduce the pre-exponential factor in Eq. (1) but not the activation energy, suggesting that H passivates crystallization sites while not affecting the energy associated with a crystallization event [7,15]. Various studies are described in the following sections to test this model and develop a more complete understanding of the role of H during SPE. 3. Experiment 3.1. Sample preparation The kinetics of SPE were measured in amorphous layers formed by self-ion implantation into b100N wafers. A National Electrostatics Corp. 1.7 MV tandem accelerator was used for all implants. The samples were tilted 7° off the incident beam axis to avoid channeling and also rotated about the surface normal by a similar amount to prevent any remaining possibility of planar channeling [16]. Substrates were affixed with Ag paste to the implanter stage, which was held at 77 K. Good thermal contact was especially important for a-Ge formation in order to avoid the porous structure that forms in highfluence room temperature implanted Ge [17,18]. Multiple ion beam energies were chosen to form homogeneous amorphous layers ranging between 1.5 and 4.2 μm thick. 3.2. Time resolved reflectivity The SPE rates of the c–a interface were determined from time resolved reflectivity (TRR) measurements by acquiring reflectivity data simultaneously using two HeNe lasers at wavelengths of λ = 1.152 μm to probe the thin amorphous layers and at λ = 1.523 μm for the thicker layers. As the c–a interface moves through the sample, peaks in the TRR reflectivity trace occur every λ/2n. By combining the measured TRR traces and a theoretical reflectivity versus c–a interface position curve the velocity of the interface was determined. These data were collected while the samples were held on a resistively heated vacuum chuck and annealed in air over the temperature range of 460–660 °C for Si and 300–540 °C for Ge. The temperature of the samples during the anneals was calibrated by comparing the reading of a type-K thermocouple embedded in the sample stage with the melting points of various suitably encapsulated metal films evaporated onto Si wafers. The error associated with the temperature reading was found to be ±1 °C. Measurements were performed in air to match the experimental conditions of other studies and so that the effects of H infiltration could
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
be examined and quantified. Further details on the experimental apparatus are presented elsewhere [19]. 3.3. Secondary ion mass spectrometry Secondary ion mass spectroscopy (SIMS) was performed on selected samples to measure the H concentration profile in the amorphous layer after a partial anneal. A Cameca IMS-5F secondary ion mass spectrometer equipped with a Cs ion source. A primary ion beam of Cs+ with a net impact energy of 14.5 keV was applied for the depth profiling. The Cs+ beam was rastered over an area of 250 μm2. Negative secondary ions were collected from the central part of the rastered region (55 μm in diameter) in order to avoid crater side wall effects. The SIMS detection limits were 2 × 1017 cm–3 and 5 × 1015 cm–3 for H and B, respectively. The depth scale was quantified by a stylus profilometer with an accuracy of within 2%. 4. Results and discussion 4.1. Hydrogen retardation of SPE Fig. 1a) and b) are schematics of the expected H concentration ([H]) at two different times during an anneal of a surface a-Si layer at 660 °C. In Fig. 1a) the c–a interface has moved 0.25 μm from its original position. During this time H has infiltrated into the amorphous layer from the surface oxide. In Fig. 1b), the H has reached the c–a interface and causes the SPE rate to decrease while being strongly segregated on the amorphous side of the interface. During segregation a fraction of the H passes over to the crystalline side of the interface where it rapidly diffuses until trapped at defect sites as denoted by the Gaussian distribution around the original c–a interface. Fig. 1c) shows the corresponding SPE rate as a function of the interface depth as determined by TRR. Between depths of 1.2 and 2.2 μm the SPE rate is the intrinsic value which can be determined with Eq. (1) above. Once the H reaches the c–a interface, the rate decreases. To study this decrease beyond the H concentrations previously considered, H was implanted into buried a-Si layers. The buried layers
Fig. 1. Schematic of an a-Si surface layer with the expected [H] profiles after a partial anneal for a) 50 s and b) 140 s at 660 °C. The interface velocities, va and vb are the intrinsic and H-retarded values, respectively. The dashed line shows the position of the original a–c interface. c) Corresponding SPE rate with the velocities, va and vb indicated.
2319
were formed by two Si implants at 0.6 and 2 MeV while the substrates were held at −10 °C. This implant temperature ensures that a surface crystalline capping remains after implantation which inhibits H infiltration from the surface into the amorphous layer to a concentration below at least 1017 cm− 3[19]. Fig. 2a shows the evolution of the implanted H profile after partial anneals at 580 °C for 400, 2550 and 5400 s. The original c–a interfaces of the buried layer are indicated by the dashed lines at 0.2 and 2.2 μm. After the first partial anneal the H profile decreases in concentration and broadens. As expected, H then segregates at the c–a interfaces which refines the profile as SPE proceeds. A fraction of the implanted H diffuses out of the amorphous layer and is gettered by defects which have formed around the original position of both the front and back interfaces. The H at these positions increases with annealing time with [H] in the near-surface region increasing at a greater rate. The original front interface position acts as a sink for both the implanted H escaping from the amorphous region and H diffusing in from the surface. This is especially clear for the 400 s anneal case. The implanted H has not yet reached the interface as shown by the lack of segregation but a relatively high [H] value is present in the crystalline capping layer. This could be due to the presence of implantation induced damage in the capping layer and could aid in inhibiting H diffusion into the amorphous layer. The refined [H] profile grows to a concentration similar to that of the original implanted concentration. Initially, it was hoped that the refinement of the [H] profile by the c–a interface would lead to a more efficient way to form a high density of H-related defects as used in the SmartCut™ technique where H is implanted into c-Si [20]. In this technique, H clusters together to form bubbles with an anneal at around 800 °C. However, too much H is lost
Fig. 2. a) SIMS of the implanted [H] profile in buried a-Si layers after partial anneals of 400, 2550 and 5400 s at 580 °C. b) The dependence of the SPE rate on the [H] determined from a). Data from Roth et al. is also shown [13].
2320
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
through the interfaces during the SPE anneal to successfully form bubbles using this method [21]. Such a study, however, did allow the [H] dependence of the SPE rate to be studied to [H] values up to 2.3 × 1020 cm− 3. Fig. 2b) shows the retardation of the SPE rate as a function of [H] determined from Fig. 2a). This data compares well with previous studies by Roth et al. [4]. Below [H]=3 × 1019 cm− 3 the normalized SPE rate has a linear dependence on the [H] as predicted by Eq. (5). Above 3 × 1019 cm− 3 the SPE rate has a weak dependence on [H] and continues to decrease to 20% of the intrinsic value at a concentration of 2.3 × 1020 cm− 3. 4.2. Hydrogen segregation As shown in Fig. 2, H segregates at the interface implying a segregation coefficient, k, of less than one. SPE rate profiles such as that shown in Fig. 1 can be used to study the segregation of H further since the dependence of SPE retardation on [H] is known. Fig. 3 shows the normalized SPE rate for surface a-Si layers for anneal temperatures between 500–640 °C. The retardation is linearly proportional to [H] as determined in Fig. 2b) and is indicated on the right axis. The SPE retardation has a strong temperature dependence with the greatest retardation occurring for the lowest anneal temperatures. This can be explained by a temperature dependent segregation coefficient. For example, during the 640 °C anneal, H more readily crosses the c–a interface resulting in less segregation and therefore less SPE retardation. Such a parameter is extremely important if the extent of retardation for a particular SPE anneal is to be determined during device fabrication. A numerical model is currently being developed to study this effect. 4.3. Hydrogen-dopant interactions It is well-known that dopants enhance the SPE rate in both a-Si and a-Ge [6,7]. The behaviour and effect of H during the SPE anneals required to activate dopants is not so well-known. Fig. 4 summarizes a recent experiment involving the implantation of B or P into the nearsurface region of a 2.2 μm thick surface a-Si layer [22]. Fig. 4a) shows the B concentration profile measured by SIMS. The B profile was formed by multiple energy B implantations in order to broaden the profile. Overlaid are two calculated H profiles for a 560 °C anneal for 90 min. To schematically illustrate the in-diffusion of H, simulated profiles are also included. The dashed line is the H profile in an intrinsic sample whereas the solid line is the H profile in the B implanted sample. Segregation was not considered in the calculation and it was assumed that B enhances the H diffusion coefficient by a factor of 20. This factor was determined from Eq. (4) and g and E−
Fig. 3. Normalized SPE rate for surface a-Si layers showing H retardation. [H] at the c–a interface on the right axis is calculated using Eq. (5) with [X0]/a = 6 × 1019 H/cm3.
Fig. 4. a) Implanted B profile in a-Si determined by SIMS and numerically calculated H concentration profiles for H diffusion from the surface. The B was assumed to enhance the H diffusion coefficient by a factor of 20. b) SIMS concentration profiles of H in intrinsic (circles), B implanted (triangles) and P implanted (squares) a-Si layers after an anneal of about 90 min at 560 °C. The resulting c–a interface depth is 1.2 μm.
values discussed in Section 4.4. Such values could not be extracted from this data due to insufficient knowledge of segregation. The H is transported through the B implanted region at a much faster rate than in the intrinsic case leading to an increased [H] value in the amorphous layer beyond the B implanted region. Fig. 4b shows the SIMS profiles for partially annealed intrinsic, B implanted and P implanted samples and shows that dopant-enhanced H diffusion is indeed occurring. The peak P concentration was greater than the peak B concentration at 5 × 1020/cm3 however the H in-diffusion is not as significant. This behaviour is similar to that found in SPE data where B enhancement is much greater than P enhancement [6]. The [H] profile around the dopant implanted region has a complex shape which is attributed to a matrix type effect of the SIMS measurement but beyond the dopant region [H] is clearly enhanced. The normalized SPE rates at a depth of 1.2 μm for these samples are shown in Fig. 5. The SPE rates corresponding to the SIMS profiles in Fig. 4b are also plotted in Fig. 2b and agree well with Roth et al. [13] The retarded rates from each sample decrease linearly as a function of 1/kT due to the temperature dependence of the segregation coefficient, k, discussed above and observed in Fig. 3. In intrinsic a-Si, we see a 20% reduction of the normalized SPE rate over the temperature range studied. The SPE retardation for the dopant implanted layers is much greater than in the intrinsic case due to the enhancement of [H]. The B implanted layers show the greatest retardation. The H-retarded SPE rates for a layer containing both B and P are also shown. These rates are much less than the B implanted sample suggesting that P compensates the doping. This results in an H diffusion coefficient that is closer to the intrinsic value.
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
Fig. 5. Normalized SPE rates determined at a depth of 1.2 μm in intrinsic (circles), B (triangles), P (squares) and compensated implanted Si layers. The solid lines are linear fits. Errors are shown on the intrinsic data only for clarity.
The dopant-enhanced SPE rates are also affected by the H. The Benhanced SPE rates were 63–47% lower in the temperature range 500–640 °C than the expected enhanced rates in H-free material [6,22]. This is a dramatic effect and fabrication codes will need to take this into account. Eq. (5) can be used in this situation to predict the SPE rate but only if the value of [H] at the interface and the passivation behaviour of H at the interface is known. This then becomes quite a complex problem with many interdependences. However, Eq. (5) in the context of the so-called asymmetric effect offers some insight into this interdependence as discussed below [3]. In previous studies an asymmetry between the dopant-enhanced SPE rate and the arsenic concentration profile has been reported. This can also be understood in terms of the infiltration of H from the surface into the amorphous layer [23,24]. By a comparison between surface and buried amorphous layers the effect of H infiltration on the SPE kinetics can be discerned. It has been found that for the nearsurface side of the As-enhanced SPE profile in surface a-Si layers SPE is retarded by the infiltrating H causing an apparent asymmetry between the enhanced SPE rate and the As profile. The SPE rate profile in buried a-Si layers does not show any asymmetry. This asymmetry was also modeled using simple H in-diffusion calculations and Eq. (5) [24]. The model showed good agreement with the observed rates if it was assumed that both donors and SPE defects (DBs) were passivated by H. The As is expected to enhance the H diffusion coefficient but this was not considered in this earlier work. Similar asymmetries have been observed in both B and P implanted surface a-Si layers [25]. We suggest that the detailed kinetics of dopantenhanced H diffusion need to be considered in order to fully account for the observed effects.
2321
Fig. 6. [H] profiles in a-Si after an anneal of 455 °C for 5 min in intrinsic (circles) and in samples implanted with 1 × 1020 B/cm3 (crosses) and 1.5 × 1020 B/cm3 (diamonds). The B profile determined by SIMS is also shown.
normalized H diffusion coefficients are fit well with Eq. (2) when E+ − Ev = (0.44 ± 0.06) eV and g+ = (0.1 ± 0.1) for a [B] = 1 × 1020 B/cm3 is used. We note that the value of g is much smaller than expected and a value of zero also lies within its error bounds. We believe this to arise from the difficulty in accurately calculating the Fermi level in a-Si. However, Fermi level shifting in a-Si is plausible once the dopant concentration exceeds the midgap defect concentration [26]. The ability of the model to fit H diffusion data well supports the proposition that both SPE and H diffusion are governed by the same defect. Furthermore, a single model can be used to describe both processes during thermal anneals. 4.5. Hydrogen in a-Ge during SPE In a-Ge, H has also been shown to affect the SPE rates but the behaviour is quite different to that observed in a-Si [7]. For ~3.25 μm thick surface a-Ge layers no large scale velocity reductions are observed until the final ~ 0.4 μm of regrowth. This had previously been reported by a number of workers who speculated that the retardation was either due to surface impurities driven into the sample during amorphization implants or H in-diffusion [4,10]. Fig. 7 shows [H] in partially annealed a-Ge layers. In the asimplanted state, H was found only in the near-surface region. With an anneal of 121 s at 420 °C the H is shown to diffuse into the sample but has not reached the c–a interface which is at a depth of 0.79 μm. In a-
4.4. Dopant-enhanced H diffusion To determine the value of [H] at the c–a interface during SPE the dependence of H diffusion on dopant concentration needs to be determined. To study this, multiple B implants were used to form a constant concentration profile over which a single H profile was implanted. Fig. 6 shows this [H] profile after an anneal for 5 min at 455 °C in intrinsic and B implanted samples. At the surface, the in-diffusion of H is similar for each sample. When B is present, the H profiles are much broader signifying B-enhanced H diffusion. The solid curves were determined by solving the diffusion equation numerically using an implicit finite difference algorithm (to be published elsewhere). The
Fig. 7. SIMS profiles of the hydrogen content in thick a-Ge samples: in the as-implanted state (○), and for partial anneals at 420 °C for durations of 121 s (△) and 227 s (◇). The expected c–a interface depths for each of these partial anneals based on the TRR data were 0.79 μm and 0.24 μm, respectively.
2322
B.C. Johnson et al. / Thin Solid Films 518 (2010) 2317–2322
Si the H would have segregated at the interface by this stage of regrowth [15]. After a longer anneal of 227 s, the H has come into contact with the interface at 0.24 μm and shows strong segregation (again k b 1 as for the Si system). The segregated [H] value is of a similar order to that observed in a-Si [15]. The in-diffusion of H in a-Ge is about one order of magnitude less than that observed in a-Si. This may be due to the inability of a-Ge to form a stable native oxide. Implanted H profiles have also been studied in a-Ge layers [7]. A saturation of the SPE rate retardation was observed for concentrations between 6 × 1018 and 6 × 1019 H/cm3. This agrees with the value of 3 × 1019 H/cm3 observed for a-Si. The SPE rate around this saturation level is a factor of ~3.7 times slower than the intrinsic rate, as compared to the factor of ~ 2 for a-Si. 5. Conclusions The effect of H on the SPE rate is complex. The H infiltrates from the surface into the amorphous surface layer during thermal processing. The H diffusion coefficient is dependent not only on the temperature but also the dopant concentration in the amorphous layer. Once H meets the c–a interface it strongly segregates on the amorphous side of the interface. The segregation coefficient is found to have a temperature dependence. The in-diffusion of H and its dopant-enhancement can be used to explain the so-called asymmetric effects of the SPE velocity profiles with the implanted dopant profiles. The infiltration of H into a-Ge is also observed but to a much smaller extent than in a-Si. For H implanted a-Ge layers the Hretarded SPE rate decreases by up to a factor of 3.7. The generalized Fermi level shifting model provides a convenient and accurate way of describing these effects under a single framework. In the model, both H diffusion and SPE occur via DBs. H affects SPE by passivating the DBs at the c–a interface. When dopants are also present H tends to passivate both the DB and the dopant atom which would otherwise take part in Fermi level shifting. Dopant and H effects are important parameters to consider in device modeling.
Acknowledgment The Department of Electronic Materials Engineering at the Australian National University is acknowledged for their support by providing access to SIMS and ion implanting facilities. This work was supported by a grant from the Australian Research Council. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26]
International Technology Roadmap for Semiconductors (2003 Edition). J.W. Mayer, L. Eriksson, S.T. Picraux, J.A. Davies, Can. J. Phys. 46 (1968) 663. G.L. Olson, J.A. Roth, Mat. Sci. Rep. 3 (1988) 1. G. Olson, J. Roth, in: D. Hurle (Ed.), Handbook of Crystal Growth, vol. 3, Elsevier Science B.V, 1994, p. 255, chap. 7. O. Hellman, Mater. Sci. Eng. R16 (1996) 1. B.C. Johnson, J.C. McCallum, Phys. Rev. B. 76 (2007) 045216. B.C. Johnson, P. Gortmaker, J.C. McCallum, Phys. Rev. B 77 (2008) 214109. N.G. Rudawski, K.S. Jones, S. Morarka, M.E. Law, R.G. Elliman, J. Appl. Phys. 105 (2009) 081101. L. Csepregi, E.F. Kennedy, J.W. Mayer, T.W. Sigmon, J. Appl. Phys. 49 (1978) 3906. G.-Q. Lu, E. Nygren, M.J. Aziz, J. Appl. Phys. 70 (1991) 5323. J. Matsuo, D. Takeuchi, T. Aoki, I. Yamada, Ion Implantation Technology. Proceedings of the 11th International Conference, 1996, p. 768. J.C. Oberlin, A.C. Chami, E. Ligeon, C. Prunier, Nucl. Instr. Meth. B 19/20 (1987) 462. J.A. Roth, G.L. Olson, D.C. Jacobson, J.M. Poate, Mat. Res. Soc. Symp. Proc. 297 (1993) 291. D. D'Angelo, L. Romano, I. Crupi, E. Carria, V. Privitera, M.G. Grimaldi, Appl. Phys. Lett. 93 (2008) 231901. J.A. Roth, G.L. Olson, D.C. Jacobson, J.M. Poate, Kinetics of Phase Transformations, MRS Symposia Proceedings, vol. 205, Materials Research Society, Pittsburgh, 1992. A. Armigliato, R. Nipoti, G. Bentini, M. Mazzone, M. Bianconi, A.N. Larsen, A. Gasparotto, Mater. Sci. Eng. B2 (1989) 63. B.R. Appleton, O.W. Holland, J. Narayan, O.E.S. III, J.S. Williams, K.T. Short, E. Lawson, Appl. Phys. Lett. 41 (1982) 711. O. Holland, B. Appleton, J. Narayan, J. Appl. Phys. 54 (1983) 2295. J.C. McCallum, Appl. Phys. Lett. 69 (1996) 925. M. Bruel, Electron. Lett. 31 (1995) 1201. D. J. Pyke, Master's thesis, University of Melbourne (2006). B.C. Johnson, P. Caradonna, J.C. McCallum, Mater. Sci. Eng. B 157 (2009) 6. B.C. Johnson, J.C. McCallum, J. Appl. Phys. 95 (2004) 4427. B.C. Johnson, J.C. McCallum, J. Appl. Phys. 96 (2004) 2381. R.M. Walser, Y.-J. Jeon, Mater. Res. Soc. Symp. Proc. 205 (1992) 27. S. Coffa, J.M. Poate, Appl. Phys. Lett. 63 (1993) 1080.