Background

In the past 40 years, metal oxide semiconductor field effect transistors (MOSFET) are used as basic component in integrated circuits (IC) where the transistor size was continuously scaled down [1–4]. As a result, main transistor characteristics, e.g., power consumption, and electric performances were improved by every new generation.

During this technological evolution, one of a central issue has been to improve transistor performance by using different strain engineering methods to enhance channel mobility [4]. SiGe alloys have been used in source/drain regions already in 90 nm node by Intel in 2003. In such transistors, selective epitaxial growth (SEG) was used to fill the source/drain-recessed regions to create uniaxial strain in the channel region. To further enhance the channel mobility, the Ge content in SiGe (or strain) has been continuously increased from lower to remarkably higher by every node [2, 5–7]. The main issue with selective epitaxy growth is that the SiGe film strain is dependent on variation of growth parameters. These parameters were optimized for growth of highly strained SiGe film and integration in pMOS source/drain areas of 22 nm node. In such transistors, the strain in the channel region is generated from SiGe which uniaxially exerts from source/drain. Another important issue to enhance the channel control and improve performance is introducing the high dielectric material (high-k) and metal gate (MG) into the traditional MOSFET [8, 9]. One of the main issue for process integration of high-k and metal gate is conformal film filling in the gate with the small trench. Atomic layer deposition (ALD) is a technology based on sequential self-saturated surface treatment and reactions, which lead to the controlled cycle-by-cycle period growth of very thin films. ALD technology is applied to deposit the high-k materials and metal gate due to its excellent trench filling and process flexibility, which is widely applied in the gate-last process integration scheme [10–12].

This article mainly presents how to grow highly strained SiGe film for source and drain application for 22 nm pMOSFETs with high-k and metal gate. The high-k material is HfO2 thin film and filling metal in the trench was B-doped W layer, both of these films are deposited by ALD technology [13, 14]. This study provides the knowledge of how to grow and apply high-quality selective epitaxy SiGe film in the transistor structures for advanced technology nodes. Finally, the transistor characteristics were measured and discussed.

Methods

The SiGe layers were grown on 8-in. Si (100) wafers at 650–750 °C and total pressure of 20–40 Torr by using reduced pressure chemical vapor deposition (RPCVD). The Si, Ge, and B precursors were dichlorosilane (SiH2Cl2), 10% germane (GeH4), and 1% diborane (B2H6) in H2, respectively. During epitaxy, HCl gas was introduced to obtain selectivity against the oxide and nitride layers on the wafers.

The growth parameters, e.g., total pressure, growth temperature, and HCl partial pressure were tuned to grow highly strained SiGe layers with a certain layer thickness. The Ge content in SiGe layers was measured directly on the patterned substrates by ω-2θ rocking curves (RCs) using high-resolution x-ray diffraction (HRXRD). High-resolution reciprocal lattice mapping (HRRLM) was performed to measure the misfit parameters in-parallel and perpendicular to the growth direction (f // and f ┴, respectively) and layer quality during the optimization of the different growth parameters [15, 16].

Cross-sectional high-resolution transmission electron microscope (HRTEM) was employed to evaluate layer quality of the grown SiGe layers in source/drain areas. Energy dispersive spectroscopy (EDS) was also performed to find out the layer profile and to examine the contamination in epi-films. The layer thickness was also measured by Tencor profilometer over different parts of the chip.

For 22 nm pMOSFETs production wafers, conformal SiO2 and SiN were deposited as gate side-wall materials. The Si recess in source/drain regions was formed by a dry etching process. All the wafers were chemically cleaned using standard procedure (SPM followed by APM with DHF at last) and placed immediately inside the load-locks of RPCVD reactor. Later, the load-locks were pumped down in order to avoid any surface contaminations (oxygen and carbon) on the wafers. Prebaking was performed by annealing in the temperature range of 800 to 825 °C for 7 min to remove the native oxide.

The Ni silicidation was performed on SiGe layers in order to reduce the contact resistance. A low resistivity NiSiGe phase was formed by two steps of annealing treatment at 300 and 450 °C for 30 s in N2 ambient [17–19].

The key process module of HK and MG contained a gate stack. At first, the dummy gate (Poly Si) and the oxide was removed, then 20-Å HfO2 layer was deposited by ALD upon formation of ~8-Å-thick interfacial layer (IL) of Si oxide by chemical method (O3-DI water). Afterwards, four layers of ALD TiN/PVD Ti/ CVD TiN/ALD W were subsequently deposited on the HK layer.

The whole device fabrication was accomplished by metallization and alloy at 425 °C in forming gas annealing (FGA). The electrical characterization (I–V) was performed with HP4156C precision semiconductor parameter analyzer.

Results and Discussions

One of the most important issues for performance of MOSFETs is integrity of SiGe SEG in terms of layer quality, selectivity, surface roughness, and strain amount and pattern dependency [2]. Although these parameters are dependent to each other but still there are ways to deal with these problems individually. For example, SiGe layers are grown in metastable region in the crystal growth and any strain relaxation results in poor layer quality and surface roughness.

The pattern dependency of SEG is referred to the situation when the layer profile (composition and layer thickness) is dependent to the pattern layout (density and size of oxide openings) and architecture (oxide or nitride) of Si wafer [20–23].

The layer quality is directly related to the cleanness of Si surface prior to the epitaxy as well as the optimization of growth parameters. Figure 1a–e shows the micrographs of the samples prior and after epitaxy. Carbon residuals from the polymer after plasma dry etch is a typical problem for epitaxy. The epitaxial layer can be deposited only on the Si clean areas, and the growth occurs through nucleation as shown in Fig. 1b, d. The EDS analysis from the cross section of the S/D areas in Fig. 1e confirms the carbon and oxygen contamination on the initial Si surface. Meanwhile, a standard chemical cleaning will remove all undesired impurities, and a two-dimensional SiGe layer could be grown successfully as shown in Fig. 1c.

Fig. 1
figure 1

HRSEM micrographs showing cross section of samples (a) prior to SiGe SEG growth, (b) poor Si surface clean, (c) good Si surface clean and SiGe growth. (d) TEM cross section of sample b and (e) EDX mapping of sample b

Full size image

Optimizing the growth parameters was performed to deposit highly strained SiGe layers with high quality, and the growth rate is high for production line. Figure 2a–c shows HRRLMs at (115) reflection of SiGe layers grown in range of 650–750 °C. The layer thickness for these samples was kept below the critical thickness for strained SiGe layers. The SiGe peaks are still aligned with Si peak along k ┴ direction showing minor strain relaxation. However, the position of SiGe layers moves closer to Si, and the broadening of contour features are ncreased with increasing the growth temperature. One reason for such behavior is the increase of the growth rate which decreases the Ge content (28.2, 25.4, and 20.1% for 650, 700, and 750 °C, respectively). The broadening of SiGe peak in Fig. 3c is an indicator for defect density in the epi-layers. One may conclude 650 and 700 °C are most suitable temperature for SiGe layers. The next step for SEG SiGe was to study the effect of HCl partial pressure on the growth kinetics. The purpose of the experiments was to obtain a working range for HCl partial pressures where the growth is selective with decent growth rate and SiGe layers have high Ge content. The Ge content was increased, and the growth rate was decreased by increasing HCl partial pressure. For example, when HCl partial pressure was 60, 80 (good selectivity), and 100 mTorr then the growth rate became 9.4, 8.4, and 4.8 nm/min, respectively. This is due to the decrease of growth rate where more Si atoms were etched by HCl molecules. Meanwhile, the Ge content was monitored 28.6, 32, and 32.6% for the above samples. The saturation of Ge content for higher HCl partial pressures occurs when the Cl aroms will not only etch Si atoms but also the Ge atoms as well. A good outcome of high HCl partial pressure during epitaxy is a better control of pattern dependency of the growth [24, 25]. At the same time, the higher amount of HCl is helpful to obtain good selectivity at the top of dummy gate and the surface of SiN spacers [3].

Fig. 2
figure 2

HRRLMs around (115) reflection of SiGe selective growth with different growth temperature (a) 650, (b) 700, and (c) 750 °C

Full size image
Fig. 3
figure 3

(a) A HRSEM of a multilayer structure with eight periods where the boron partial pressure varied and (b) its SIMS profile

Full size image

In a transistor, low sheet resistance in source/drain region is a crucial matter. Therefore, high boron doping is sought in the epi-layers. Figure 3a, b shows SIMS analysis and cross-section micrograph from a multilayer structure of SiGe/Si with nine periods where the boron concentration has been successively increased in the SiGe layers. No extended defects were observed in the micrograph indicating a high epitaxial quality. The Ge signal in the SIMS spectra is constant and was not affected in presence of boron in the epi-layers. This fact can be used to estimate the boron concentration from strain compensation using HRRLMs. Figure 4a, b shows HRRLMs from an intrinsic and B-doped Si0.65Ge0.35 layer with thickness of 100 nm. The shift of SiGe peak due to B-doping is only along k // direction showing no strain relaxation in epi-layer. The boron concentration (CB) was calculated from misfit parameters (f // and f ┴) using the following equations:

Fig. 4
figure 4

HRRLMs around (113) reflection of SiGe in 22 nm transistor S/D areas with (a) an intrinsic layer and (b) B-doped layer

Full size image
$$ f=\frac{1-\nu}{1+\nu}\left({f}_{\perp }-{f}_{//}\right)+{f}_{//}{} $$
(1)
$$ {\mathrm{C}}_{\mathrm{B}}=\frac{f}{\beta} $$
(2)

where ν is Poisson ratio for SiGe (ν = 0.287) and β is the contraction coefficient of boron in Si (6.3 ± 0.1 × 10−24 cm3/atom) [14]. The extracted value shows a boron doping level of 1–3 × 1020 cm−3. It is worth mentioning here that this extracted value is concentration of substitutional (or active) boron atoms in the SiGe matrix [26, 27].

In transistor structure, the boron-doped SiGe layer consisted of two layers where the main stressor material was Si0.60Ge0.40 but a sacrificial Si0.80Ge0.20 layer was deposited for silicidation in S/D regions. This cap layer is consumed during the silicide formation, and no harm was imposed to the SiGe beneath. A cross-section image of a processed transistor is shown in Fig. 5. EDS analysis demonstrates the profile of different layers. The investigated elements were germanium, silicon, nickel, and oxygen. The oxygen signal was at the noise level which shows no contamination at the interface or within the SiGe layer. The profile of formed NiSiGe on top of S/D has resulted a push-out of Ge atoms to the beneath SiGe layer causing a pile up at the interface [19]. There is a discrepancy for Ge content from XRD and EDS analysis (Si0.60Ge0.40 and Si0.65Ge0.35, respectively). It is worth mentioning here that Ge content was calculated by XRD from the strain in the layer which is partially compensated by boron atoms, whereas EDS shows the atomic Ge concentration.

Fig. 5
figure 5

TEM cross-section image of a processed transistor with EDS analysis of different layer profiles

Full size image

At final stage, 22 nm PMOS transistors with integrated SiGe S/D and HK and MG-process modules are electrically characterized. Figure 6a shows the Id–Vg transfer characteristic curves and b shows the Id–Vd output characteristic curves. The results show that saturation drive current of SiGe S/D device increases from 488 to 639 μA/μm, while the Ioff changes from 0.83 to 1.32 nA/μm, mainly due to SiGe source and drain replacement processes and defects present in the film. The inserted table summarizes the device’s electrical performance comparisons between 22 nm bulk PMOS SiGe S/D and Si device. The results show that the PMOS device with SiGe S/D has a 30% performance improvement compared with traditional silicon device, and the other related performance parameters are not changed too much.

Fig. 6
figure 6

The (a) transfer and (b) output characteristic curves of 22 nm planar devices with SiGe S/D compared to Si S/D

Full size image

Conclusions

The integration of selective epitaxy of SiGe (0.35 ≤ × ≤ 0.40) in the source/drain areas and high-k and metal gate was demonstrated for 22 nm node PMOS device in this research. The quality of SiGe layers was directly dependent on Si surface prior to epitaxy. This was obvious when the source/drain opening was plasma-etched and carbon or oxygen residual were formed on Si surface. The growth parameters, e.g., growth temperature, total growth pressure, and HCl partial pressure had also impact on the epitaxial quality and they were optimized. The boron concentration in SiGe layers was estimated from strain compensation between the intrinsic and B-doped SiGe layers by using HRRLMs. The B-doped SiGe layer in S/D regions consisted of strained Si0.60Ge0.40 and Si0.80Ge0.20 cap layers in order to protect the highly SiGe layer during the Ni-silicidation process. The results showed that the strain in Si0.60Ge0.40 in S/D was not affected by formation of NiSiGe in the cap layer. The Ge profile in the transistor structure was measured by EDS and XRD technique. The manufactured PMOS transistor with SiGe S/D showed a remarkable better performance compared with traditional silicon device.