B and Ga Co-Doped Si_1−xGe_x for p-Type Source/Drain Contacts

The SIMS composition profiles, extracted as a function of depth, are presented in Fig. 1 for samples R, A, B1, and D. Unstable signals recorded in the first few nm are attributed to near-surface (depth < 5 nm) transient distortions. ²⁴ This region of the layer will therefore be disregarded in the upcoming analysis.

Zoom In Zoom Out Reset image size

The Ge concentrations (x_Ge) are reported using a linear scale (right axis) and are subject to error bars whose magnitude are difficult to evaluate. The Ge concentration in sample B1 was extracted twice using a different set of Si_1−xGe_x standards in each run. This resulted in a difference between x_Ge values of 2 at.%. In the following discussions, this value will be adopted as an indicative error bar for the SIMS-measured Ge concentrations. In samples R and A, the Ge concentrations are constant throughout the layers depth and are close to the 45% target value. The Ge content is lower in sample B1 and slightly decreases when moving towards the surface (from 43.5% at the layer/substrate interface to 42.5%, just below the surface peak). In sample D, a more pronounced decrease in Ge concentration is observed (from 50%, before the overshoot near the substrate, down to 42%, near the surface). Ge profiles with similar gradients have been observed before in Ga-doped and B + Ga co-doped Si_1−xGe_x. ²⁵ The formation of a Ge gradient might be associated with the co-flow of Ga-precursors.

Boron concentration profiles ([B]_chem) are flat at ∼ 1 × 10²¹ cm⁻³ in all the measured samples, except for sample D, where the B level gradually decreases from a maximum of 6 × 10²⁰ cm⁻³, close to the interface with the substrate, down to 4.5 × 10²⁰ cm⁻³ in the near-surface region. However, due to the gradient in x_Ge reported for this layer, the sputtered matrix gradually changes during the analysis, possibly leading to variations in the B quantification.

The Ga chemical concentration ([Ga]_chem) profiles, displayed for samples A, B1, and D in Fig. 1, appear graded, with increasing level when moving towards the surface. Abrupt concentration peaks are observed in the first 2 nm from the surface, yet transient effects obstacle the correct elemental quantification at the surface proximity. To verify the actual incorporation of Ga in the epilayer, a piece of sample B1 was cleaned with the basic solution described in the experimental section and remeasured. The comparison of the SIMS profiles between the layer in its as-grown form and after the surface cleaning is shown in Fig. 2. A variation in thickness of ∼ 2 nm due to the cleaning treatment can be estimated by overlapping the two B profiles. The B bulk levels remain unchanged, while the extracted Ga profile is dramatically impacted, with a significant reduction in Ga concentration after the cleaning. This decrease is visible (i) near the surface, and (ii), also in the bulk part of the layer. The surface decrease is explained by the removal of the Ga-rich top layer formed on the as-grown surface. This result confirms that during epitaxy, Ga accumulates at the growth front and the large gradient in Ga concentration towards the surface is not real. The decrease in bulk Ga concentration evidences the occurrence of a SIMS artefact caused by Ga knock-on during sputtering (recoil mixing). ^26,27 The Ga dose which is detected when no surface clean is applied, therefore originates mainly from the sample surface, with only a small fraction of it being incorporated in the bulk region, with a saturation level below 1 × 10¹⁸ at. cm⁻³. The real Ga profile therefore presents a steep surface peak (removed with the surface clean), followed by a lowly doped region and a zone fully depleted of dopants adjacent to the interface with the substrate. As described by Greene et al., this profile is characteristic of layers doped with elements exhibiting a strong segregation tendency. ²⁸ For sample B1, the fractions of dopants present at the surface and within the bulk region, respectively, can be estimated by integrating the two recorded profiles as measured before and after the clean. Following such an approach, a total Ga dose of 2.7 × 10¹³ at. cm⁻² is obtained before clean, of which only 7 × 10¹¹ at. cm⁻² is incorporated within the layer (after clean, i.e. ∼ 3% of the total). The estimated total Ga doses for the other samples are reported in Table I. The quantification of the superficial Ga may be incorrect since Ga is not incorporated in a Si_1−xGe_x matrix but is likely present over the surface in a different phase. The calculated doses can still be used for relative comparison among the samples. Finally, the Ga profile measured in sample D presents an inflection point at around 25 nm from the surface (Fig. 1d). A probable explanation is that, from that point on, the bulk Ga signal emerges from the decaying tail originated at the surface. This would result in a bulk concentration of ∼5 × 10¹⁸ at. cm⁻³, significantly higher than the bulk Ga level measured in sample B1 (<1 × 10¹⁸ at. cm⁻³).

Zoom In Zoom Out Reset image size

Carbon might be unintentionally incorporated within the Ga-doped layers due to the C-containing groups present in the Ga metal-organic precursor. The presence of adventitious C on the sample surfaces, however, has a significant impact on the C profiles shown in Figs. 1a–1d. This is confirmed by C being detected in the first 20 nm of the reference layer, which was grown without any C-containing precursor. Furthermore, beyond 20 nm in depth, the measured C concentrations in all layers are lower than those in the underlying substrates. This indicates that the C concentration in the reported samples is negligible and likely lower than the C SIMS detection limit (∼1 × 10¹⁸ at. cm⁻³).

Figure 3 shows top-view SEM images and XRR spectra acquired on the samples at wafer center. Arbitrary vertical offsets have been applied to the XRR curves to enable a sample-to-sample comparison. Samples R, A, and B1 exhibit smooth surfaces (Fig. 3a). This is also confirmed by XRR which shows clear Pendellösung fringes for the full spectra. For these three samples, a similar root mean square roughness (r_RMS) of around 0.5 nm is obtained (Table I). Thus, despite the high B concentration in the Si_1−xGe_x layers, an enhancement of surface roughening does not occur. The surfaces of samples B2 and C, however, are rougher, with estimated r_RMS values of 1.4 and 3.7 nm, respectively. Although samples B1 and B2 were grown with identical process parameters (except for the deposition time), B2 exhibits a significantly rougher surface. This can be explained by the higher Ga surface concentration in this sample, due to longer Ga accumulation during the epitaxy. The epilayer in sample C was grown with the highest Ga precursor flow, resulting in a high Ga surface concentration and a consequent surface roughness. Sample D has the highest Ga dose among the samples measured by SIMS, but it still has a smooth surface morphology (r_RMS = 0.4 nm), indicating that the amount of surface-segregated Ga in this sample is limited. In samples B2 and C, we expect higher Ga doses. Sample thicknesses as extracted from XRR spectra fitting are reported in Table I.

Zoom In Zoom Out Reset image size

HR-XRD ω−2θ scans acquired around the (004) Bragg reflection are shown in Fig. 4a. Again, arbitrary offsets were applied to the curves to ease sample-to-sample comparison. Pendellösung fringes located on both sides of the Si_1−xGe_x diffraction peaks are observed for all the samples. Their presence provides a first indication that the layers are strained. The thicknesses extracted using the side fringes are in good agreement with the values obtained from XRR (Table I). The out-of-plane lattice parameters (a_⊥) of the different samples were derived from the Si_1−xGe_x peak positions and used to determine the Ge content in the layers as described in the following.

Zoom In Zoom Out Reset image size

In undoped Si_1−xGe_x layers, the relaxed lattice constant is determined by the modified Vegard's law: ^29,30

$$\begin{eqnarray}&&{a}_{Si\left(1-x\right)Gex}={a}_{Ge}x\,+{a}_{Si}\left(1-x\right)-bx\left(1-x\right)\end{eqnarray} \tag{ 1 }$$

where a_Si(1−x)Gex, a_Ge and a_Si are the lattice constants of relaxed Si_1−xGe_x, Ge, and Si, respectively. The last term of the expression corresponds to the deviation from the pure Vegard's law and b = 0.0272 is the bowing parameter. The samples reported herein were targeted to be fully strained (pseudomorphic). In this case, the Si_1−xGe_x has a tetragonally distorted unit cell, with the in-plane (a_∣∣) and out-of-plane (a_⊥) lattice constants expressed as follows: ¹⁷

$$\begin{eqnarray}&&{a}_{\parallel }={a}_{Si},{a}_{\perp }={a}_{Si}\left(1+k\,\displaystyle \frac{{a}_{Si\left(1-x\right)Gex}-{a}_{Si}}{{a}_{Si}}\,\right)\end{eqnarray} \tag{ 2 }$$

where k ∼ 1.75. Once a_⊥ is measured, the x_Ge concentration in the alloy is obtained through:

$$\begin{eqnarray}&&\begin{array}{l}{x}_{Ge}({a}_{\perp })=\\ \displaystyle \frac{-\left({a}_{Ge}-{a}_{Si}-b\right)+\,\sqrt{{\left({a}_{Ge}-{a}_{Si}-b\right)}^{2}+4b\left({a}_{Si}+\displaystyle \frac{{a}_{\perp }-{a}_{Si}}{k}\right)\,}}{2b}\end{array}\end{eqnarray} \tag{ 3 }$$

The extracted apparent Ge concentrations are reported in Fig. 4b and Table I. The apparent Ge concentrations (x_Ge,app) extracted in this way do not account for the presence of substitutional boron in the layers. Substitutional B, in fact, induces a shrinkage of the Si_1−xGe_x lattice parameter causing a shift of the diffraction peak towards the Si substrate peak. The measured apparent Ge concentrations will thus result in lower values compared to the real value. A decrease in x_Ge,app is noticed by comparing samples R and A with samples B1, B2, and C. More specifically, for B1, B2, and C, x_Ge,app decreases with the dose of Ga-precursor flown during the epitaxial growth. This decrease corresponds to a reduction in the out-of-plane lattice parameters for these samples, that might be caused by several factors: (i) an increase in substitutional B concentration, (ii) a decrease in Ge concentration, or (iii) the onset of plastic strain-relaxation. However, the available SIMS data does not show differences in [B]_chem for samples R, A, and B1, and the electrical characterization of the layers (discussed in the next section) rules out the possibility of different activation levels, that should otherwise be visible from the resistivity and the Hall carrier concentration data. Our observations suggest that the decrease in Ge concentration (ii) with the increase in Ga-precursor flow is a more likely explanation. Sample D was grown with a different set of process conditions and shows a higher x_Ge,app. According to the SIMS spectra (Fig. 1d), [B]_chem is lower in sample D. In addition, the gradient in the x_Ge profile is larger than in the other samples and it starts from a higher concentration near the substrate. This explains the higher x_Ge,app. Finally, the absence of strain-relaxation—proposed cause (iii)—has been verified with reciprocal space maps, acquired on both samples B2 and C around the asymmetric (113) Bragg reflection (Fig. 4c). These two samples were chosen because of their non-ideal surface morphologies. In both maps, the Si_1−xGe_x:B:Ga and the Si substrate peaks exhibit the same h(110) value, meaning that the in-plane lattice parameters of the epitaxial layers and of the underlying substrate are equal. Both layers are therefore fully strained, thus confirming the interpretation of the thickness-fringes visible in the ω−2θ scan spectra. However, in Fig. 4c halos are noticed around the Si_1−XGe_X:B:Ga diffraction peaks. Their presence may be explained by a certain degree of defectivity introduced by the doping with Ga.

In addition, the main Si_1−xGe_x diffraction peaks acquired on samples B2 and D present some variations in shape if compared to the spectra corresponding to samples R, A, B1. This difference is explained by the graded Ge concentration within these epilayers—as observed in the SIMS spectrum of sample D (Fig. 1d). x_Ge progressively decreases by moving from the interface with the Si substrate towards the surface. If a graded Ge concentration profile is considered within the XRD simulation model, it results in an excellent match between the simulated curve and experimental data (Fig. 5). With this method the gradients in Ge concentration can be estimated. The extracted differences between surface and bottom x_Ge values are ∼ 6 and 7.5 at.% for sample B2 (Fig. 5a) and D (Fig. 5b), respectively. The latter is in nice agreement with the variation of Ge extracted from SIMS (8 at.%). In addition, sample D presents an asymmetric peak shape. This is well fitted if the Ge overshoot near the interface with the substrate (Fig. 1d) is also taken into consideration within the XRD fitting model. Finally, an even broader diffraction peak as seen for sample C (Fig. 4a), which can be explained by the higher defectivity. This renders the interpretation of the peak shape too ambiguous to extract any meaningful information from it.

Zoom In Zoom Out Reset image size

As mentioned above, a high substitutional boron concentration ([B]_sub) influences the Si_1−xGe_x diffraction peak position in the XRD spectra. The magnitude of the recorded shift can be used to estimate the concentration of substitutional boron [B]_sub. ³¹ For a certain boron molar fraction (y), the relaxed lattice parameter for Si_1−x−yGe_xB_y is given by: ³²

$$\begin{eqnarray}&&{a}_{Si\left(1-x-y\right)GexBy}={a}_{Ge}x+3.806y+{a}_{Si}\left(1-x-y\right)-bx\left(1-x\right)\end{eqnarray} \tag{ 4 }$$

where the contraction factor of 3.806 accounts for the substitutional B present in the layer. Then, for pseudomorphic B-doped Si_1−xGe_x, expression (2) becomes:

$$\begin{eqnarray}&&{a}_{\perp }={a}_{Si}\left(1+k\,\displaystyle \frac{{a}_{Si\left(1-x-y\right)GexBy}-{a}_{Si}}{{a}_{Si}}\,\right)\end{eqnarray} \tag{ 5 }$$

with k again assumed to be equal to 1.75. If the x value is known from other characterization techniques, it is possible to estimate the substitutional boron concentration from the measured out of plane lattice parameter through:

$$\begin{eqnarray}&&y\left({a}_{\perp }\right)=\displaystyle \frac{1}{3.806-{a}_{Si}}\left[\,\displaystyle \frac{{a}_{\perp }-{a}_{Si}}{k}+({a}_{Si}-{a}_{Ge})x\,+bx\left(1-x\right)\right]\end{eqnarray} \tag{ 6 }$$

For Si_1−xGe_x this procedure is not straightforward as the assumed x_Ge values deeply impact the [B]_sub estimated in this way. To extract [B]_sub, the Ge concentrations as measured by SIMS were therefore used. An inaccuracy in [B]_sub of ±1.3E20 cm⁻³ is estimated considering the aforementioned uncertainty of 2 at.% in the x_Ge quantification. It is consequently impossible to draw an accurate trend with this approach, but for all samples a [B]_sub value within the range 6–7.5 × 10²⁰ at. cm⁻³ was extracted. This implies an almost full activation of B in sample D but a small fraction of inactive B in the remaining samples.

The resistivity (ρ) of the samples is calculated from the sheet resistance and the film thickness as measured at the center of the wafers. A 0.1% relative error in R_s is considered. ³³ Figure 6a shows the resistivity values of samples R, A, B1, B2 and C, as a function of the Ga precursor flow used during epitaxial growth. The resistivity is found to increase by increasing the Ga precursor flow. Samples B1 and B2 show very similar resistivity values, despite the more degraded morphology observed for B2 (Fig. 3a). A possible explanation for the observed increase in ρ with Ga precursor flow is that the material crystallinity is affected by the Ga segregation mechanism which introduces defects leading to a degradation in carrier mobility. In addition, a depletion region is formed at the p-n junction at the interface between the epilayer and the substrate, slightly narrowing the electrical thickness in the layer. This has the largest impact on the thinnest layers (A, B1), causing the measured ρ to be higher. Figure 6b shows the measured resistivity as a function of [B]_chem as measured by SIMS. For samples B2 and C, a [B]_chem of 1 × 10²¹ cm⁻³ has been assumed. Literature data for Si_1−xGe_x:B (0.3 ≤ × ≤ 0.5), ^32,34–36 and the resistivity of an in-house processed B-doped Si_0.50Ge_0.50 layer (labelled as "Lowly doped Si_0.50Ge_0.50:B") are added for comparison. The resistivity of sample R aligns with the trend defined by the literature data, while the data points for the Ga-containing samples A, B1, B2, and C are placed above the trendline. Sample D, which has a smooth surface morphology and is the thickest sample in this study, has its ρ value in line with the trend. The different process conditions used for the growth of this sample seem to mitigate the detrimental effect of Ga-doping on material resistivity.

Zoom In Zoom Out Reset image size

Micro-Hall effect (MHE) measurements were performed on samples R and B1 to extract and compare the Hall carrier concentrations (p_H ) and the Hall mobilities (μ_H). Sample B1 was measured before and after the surface cleaning described in the experimental section to identify any possible impact of the surface-Ga removal on the measured carrier concentration. The Hall scattering factor (r_H), not precisely known, was considered equal to 1. This may lead to an overestimation of the measured carrier concentration. To determine the error bars on the p_H values, an uncertainty of 0.5 nm was considered for the XRR-extracted thicknesses and the 3-sigma variation (3σ_N) over a 5-point average for the Hall-dose (N_H). The relative standard deviation for N_HS and μ_H is similar in magnitude to the values reported in. ³⁷ The results are shown in Fig. 7 with the measured p_H values found to be similar for all samples. The surface cleaning was found to have no significant impact on the carrier concentration, meaning that the segregated Ga sits inactive on top of the layer surface. On the other hand, for the Ga + B co-doped samples, the mobility is lower than for the B-doped reference. Combining these results with the ρ measurements (Fig. 6a), SEM inspections (Fig. 3a), and the RSM acquired on samples B2 and C (Fig. 5), where higher Ga-doses were deposited, it can be suggested that the Ga add-on causes a degradation of the material crystalline quality, which influences carrier scattering and, in turn, the hole mobility.

Zoom In Zoom Out Reset image size

MR-CTLM structures were finally fabricated to evaluate the specific contact resistivity of Ti/Si_1−xGe_x:B(:Ga) contacts. A possible source of error in this measurement is the deviation of the dielectric spacing width from the nominal dimension that may significantly affect the extracted contact resistivity. Therefore, the width of the dielectric spacings between the concentric metal rings has been measured by inspecting one of the smallest MR-CTLM fabricated structures with X-TEM, as shown in Fig. 8b. A 3% inaccuracy in the spacing width measurement is considered and it is used to estimate error bars on the extracted ρ_c values.

Zoom In Zoom Out Reset image size

ρ_c is plotted as a function of [B]_chem in Fig. 8a, and extracted values are summarized in Table II. Figure 8a also includes internal data from Si_1−xGe_x:B epilayers with several [B]_chem values (range: 4.6 × 10¹⁸–1 × 10²¹ cm⁻³) measured using the same MR-CTLM scheme as this work. In addition, the ρ_c vs [B]_chem trendline for Si_1−xGe_x:B/Ti contacts reported in ³⁸ is plotted. Surprisingly, the value of ρ_c in the Si_1−xGe_x:B reference sample (R) is much higher than the expected value (measured to be ∼ 9 × 10^–9 Ω.cm², while expected to be ∼2 × 10^–9 Ω.cm²). We believe this data point is an outlier due to a non-optimal CTLM fabrication process for this specific sample. In fact, the anomalous ρ_c measured is not explained by characterization results from sample R.

Among the Ga co-doped samples, the best contact properties are achieved for the lowest Ga dose (sample A), with a measured ρ_c below 3 × 10^–9 Ω.cm². In samples B2 and D, the increase in Ga dose and consequent morphology degradation, are correlated with an increase in contact resistivity and a deviation from the trend. In sample D, the morphology is not affected by the Ga surface accumulation and the ρ_c formed at the Ti - semiconductor –interface lies on the trendline from. ³⁸

It is worth highlighting that the layer thickness is not the same for all the samples, as summarized in Table II. In a recent study, ^34,39 we demonstrated that the thickness of Si_1−xGe_x:B films is a parameter affecting the specific resistivity of the Ti/Si_1−xGe_x:B contacts. In particular, a decrease in B incorporation was observed when the B-doped Si_1−xGe_x layer relaxes during epitaxial growth. However, from the different characterizations performed on the current samples, no evidence of strain relaxation was found. Also, the B SIMS profiles do not show any significant decrease towards the surfaces, as oppositely observed for the relaxed Si_1−xGe_x:B samples in the aforementioned references.

None of the Ga + B co-doped samples presented here exhibits a ρ_c below the trendline of Fig. 8a. Based on the results of this work, we cannot claim beneficial effects on the p-Si_1−xGe_x/Ti contact properties linked to the Ga add-on. On the contrary, Ga surface segregation seems to cause a degradation of the material and contact properties, once a critical Ga superficial dose is exceeded. However, it must be reminded that the Ga incorporation is quite poor in the epilayers here reported. By finding ways to control the dopant segregation, improvements in electrical and contact properties may be achieved by means of Ga in situ doping.