MOCVD Copper Metallization for High Aspect Ratios TSV 3D Integration. Open Access

3D integration enables stacking of heterogeneous functions by vertically connecting devices using “Through-Silicon Via” (TSV) interconnects. This architecture offers considerable footprint reduction on the circuit board but also performances improvements: from reduction of signal delay and power consumption thanks to interconnect length reduction, to increased data bandwidth through the high density interconnects. TSV technology is considered as a lead element to deliver high performances compact devices [1]–[2]. To ensure the communication between chips, TSV have to be filled with a conductive material. Copper is the most common material used for via fill because of its low resistivity (1.67 μΩcm), excellent electromigration performance and stress resistance [3]. The fig.1, illustrates 3D-IC TSV integration schemes. Depending on the targeted application, “TSV metallization” can be performed at different step of the process flow. In the “Via First” integration, TSV are performed before the FEOL, in this case the thermal budget in not a limitation (~1000°C). In the “Via Middle” approach, via are performed between FEOL and BEOL, the critical TSV parameter is the geometry since current designs require narrow and very high aspect ratio vias [4]. In the ‘Via Last’ approach, wafers are thinned and bonded to a glass or silicon carrier before to perform the vias. The bonding glue has a maximum temperature limit of 225°C. Beyond this temperature the glue is modified and generates wafer-level thermo-mechanical issues.

These kinds of geometries and thermal budget considerations are producing an evolution of standard depositing techniques.

We developed a MOCVD (Metal-Organic Chemical Vapor Deposition) copper seed layer process compatible with architectures such as the TSV Last approach (thermal budget restrictions, typically 200°C - Ø: 30μm AR3:1), TSV Middle process (Ø: 10μm AR more than 10:1) and TSV Last high density (AR10 with Ø < 1μm) [5].

To perform TSV metallization, firstly the silicon substrate is etched using Bosch® process to form the vias. Next, an oxide liner is deposited for electric isolation, then a barrier film is deposited to prevent copper diffusion in the silicon. Finally copper seed layer deposition is required to initiate copper via fill by electrochemical deposition (ECD). In this study, the under layer used is a TiN barrier film deposited by MOCVD at 200°C, covering all thermal budget scenarii. Compared to Ta/TaN material, TiN is cost effective and less stressed. TaN films show relatively rough morphology and poor adhesion [6].

The copper seed layer deposition has to match with every 3D integration scheme restriction and E-plating criteria. A thin, continuous and conformal copper seed layer with low resistivity is necessary to produce bottom-up fill without voids. The high conformality requested, eliminates physical vapor deposition techniques (PVD) as a solution. The main advantage of MOCVD is to overcome the step coverage limitations of ionized-PVD (i-PVD) approaches used with lower aspect ratio designs [7], typically (Ø: 10μm AR8:1).

Since MOCVD is mainly based on chemical reactions, this is the ideal technique regarding structure complexity ensuring molecular transport with an equal treatment of whole surface. Considering that the deposition mechanisms and films properties depend on the precursor chemistry.

Other alternatives for metallization are being explored to replace copper to fill the TSV such as MOCVD cobalt [8], or CVD/ALD ruthenium [9]. The compatibility with the chemistries used during electro-plating, cost of the material and the difficulty to planarize by subsequent chemical mechanical polishing (CMP) have to be taken into account.

In front of challenging and complex devices, the aim is to define the best technique using the most suitable material and in the case of chemical deposition; the relevant precursor associated to the good reagent. MOCVD offers high conformality similar to Atomic Layer Deposition (ALD), but provides significantly higher deposition rates [10], and is therefore more suitable for industry use. Since a strong reactivity between Ti and F is well-known to form Ti-F and Ti-N-F bonds, and to avoid generating non-volatile compounds linked to potential adhesion issues from subsequent reactions between the TiN and the copper film [11], a fluorine-free organometallic Cu precursor, the bis (dimethylamino-2-propoxy)copper(II), Cu(dmap)₂, was chosen. The molecule structure is illustrated on fig.2.

In previous studies, various co-reagents, were evaluated with Cu(dmap)₂, such as diethylzinc (Et₂Zn), enables atomic layer deposition at low temperature (100–150°C) by ligands exchange reactions. But it offered a very narrow process window, since above 130°C the decomposition of Zn(dmap)₂ led to the formation of ZnO compounds, and Cu-Zn alloy. This surface contamination generated a modification of the resistivity and a significant change in adhesion [12].

The choice of the good co-reagent is as critical as the choice of precursor chemistry. It needs to offer good reactivity with the molecule and to give a reaction as clean as possible.

In our study, argon was used as a carrier gas to transport the Cu(dmap)₂ to the reactor. Hydrogen is firstly used for pre-treating the TiN film to promote nucleation sites density. H₂ is also used as a co-reagent, to help the adsorption of Cu(dmap)₂ molecule and ligand dissociation by β-hydrogen elimination and reductive elimination at low temperature [17–18].

We aimed to synthesize thinner copper film at 175°C, with lower stress, without any additional integration steps, since materials (TiN and Cu) are identical to those used by PVD.

This study was performed on 300 mm silicon oxide wafers. The TiN barrier and the Cu seed layer films were deposited on an SPTS Sigma fxP system equipped with two separate 300mm MOCVD reactors dedicated for each material. All studied TiN/Cu stacks were deposited without any vacuum break. In each case, the gaseous precursor was introduced into the reactor using a dual showerhead system to prevent any preliminary upstream reactions. The MOCVD TiN film deposited using TDEAT as the precursor, offers good barrier properties [13].

The first part of this study, is focused on characterization of copper films deposited on blanket wafers to determinate the process window related to the selected organometallic precursor to control nucleation and growth mechanisms. We studied the influence of deposition parameters described in Table 1 below with the investigated parameter ranges:

In the second part, the copper seed layer was integrated on high aspect ratio structures to confirm molecular transport and chemical reactions down to the TSV base, in order to obtain continuous coverage of Cu required for subsequent ECD.

Copper film conformality was measured using SEM (Hitachi S-5500) cross section along the 10 μm × 120 μm TSV sidewalls. Via fill using Cu electroplating was observed using FIB (HELIOS) on 10μm×100μm TSV structures. MOCVD copper microstructure was analyzed by X-ray diffraction (XRD X'pert Pro), in configuration θ-2θ with copper Kα using a 0.05° acquisition step. Compositional analysis was performed using X-ray photo-electron spectroscopy (XPS), using a PHI 5000 VersaProbe II spectrometer. The stack studied was: Cu 100nm deposited at 175°C / TiN 60nm deposited at 200°C / Thermal SiO₂ 100nm. The stress was evaluated by measurement on blanket wafers (FRT Microprofilometer). The electrical resistivity was measured using the four point probe method, 49pts EE 6mm (Napson).

The interface TiN/Cu was observed by STEM (FEI Titan Themis). Adhesion was checked by a scribe test, post anneal (400°C, 30 min under N₂/N₂H₂ flow, ramp 0.5 °C.min⁻¹).

XRD characterization was done on TSV sidewalls using Empyrean Panalytical, to compare the stress inside the copper film for each deposition technique, CVD and i-PVD. To assess the impact of the crystalline copper seed layer texture on ECD growth, we characterized MOCVD and i-PVD copper films on TSV sidewalls.

At this end, a characterization protocol adapted to a 3-dimensional structure has been developed. It consists of measuring the XRD patterns of the TSV as a function of the depth from the top surface down to the TSV base. As illustrated in fig. 3, thin and hard X-ray beam, is directed perpendicular to the to the copper-filled TSV length and the XRD diagram is recorded in a transmission configuration.

The sample is then scanned by step of 10μm along the TSV axis (z-axis) and a 2θ scan is performed at every step. From these 2θ scans, each diffraction peak can be analyzed in terms of peak position, width and intensity. Experiments have been performed on an Empyrean Panalytical diffractometer, using a focusing primary optics and divergent slit of 50μm. A parallel plate collimator (0.18° aperture) and a scintillator are used for the secondary optics.

Sample preparation: For this XRD analysis, a Kα Ag X-ray source was used. With this hard X-ray source, the penetration depth into silicon is a few hundred of microns. A significant absorption in the silicon has to be taken into account, assuming that the sample must have a maximum thickness of 500 μm, and length of 15mm.

The diffracted beam is thus recorded by a mobile scintillation meter, in a 2θ angle. Since the wavelength is fixed during the measurement, inter reticular distances d can be deduced from the θ angular position of diffraction peaks, according to the Bragg's Law. This study was performed using the sample referenced as AR 10 TSV, with the following stack:

• Thermal oxide : 500 nm.

• MOCVD TiN barrier : 40 nm.

• Copper seed layer; MOCVD: 100 nm vs. i-PVD: 1500 nm.

• Copper ECD via fill.

• With and without anneal: 400 °C, 30 min, N₂H₂ flow.

In this study, we focused on identifying the best process conditions. We observed that H₂ pre-treatment of TiN has a major impact on deposition morphology. Without H₂ flow pre-treatment, the copper film consisted on agglomerated copper islands randomly dispersed. Adding H₂ flow pre-treatment, improved the copper deposition rate by increasing significantly the nucleation sites density, but the copper film was consisted of poor connected grains. By maintaining H₂ pre-treatment and using H₂ flow as a co-reagent, the morphology was improved by promoting lateral diffusion of copper grains and ligands release.

The fig.4a, illustrates a MOCVD stack TiN/Cu with respective thicknesses of 50nm and 130–150nm obtained in optimized conditions; MOCVD Cu deposited with H₂/H₂O flow at 175 °C, 2.10³ Pa. Adding H₂O as a second co-reagent, increased significantly the deposition rate, by creating hydroxyl groups on TiN surface which helped this adsorption of Cu(dmap)₂ molecule on Ti-N-O-H, Ti-N-H bonds and the ligands dissociation by a proton-transfer reaction such as H₂. H₂O promoted the well-connected grains, the copper film was continuous, and no void was observed as shown in fig. 4b.

Diffraction spectra that consists of several peaks of varying intensities indicates that the copper film has a polycrystalline texture, as shown on fig.5. XRD peaks identified at 43.30°, 50.43° and 74.13°, respectively correspond to Miller indices <111>, <200> and <220> (ICDD 00-004-0836 [14]), with a preferential orientation in the direction <111>, the one targeted to prevent electromigration [15]. These peaks are consistent with the copper metallic state, no other peak was detected, indicating that the copper film was pure without CuO or Cu₂O phases. The calculation of the crystallite size by the Scherer method indicates an average value of 270 nm for the direction <111> and 85 nm for the direction <200>.

The composition of the copper film was analyzed by XPS fig.6. MOCVD Cu was deposited with H₂/H₂O flow at 175 °C, 2.10³ Pa. The deconvolution of Cu2p bond energy spectra provides peaks Cu 2p_1/2 and Cu 2p_3/2 with respective binding energies of 952.6 eV, specific for metallic state and 932,7 eV. The abrasion depth is around 20nm, to ensure a composition analysis inside the copper film. Cu2p_3/2 binding energies of Cu₂O and CuO are respectively 932.6 eV and 933.6 eV [16]. In our study, at low deposition temperature (175°C), the XPS depth profile highlights that oxygen is only presents at the extreme surface, due to the vacuum break after deposition and not inside the bulk of the copper layer. Below 225 °C, ligands of the Cu(dmap)₂ molecule, with carbon and oxygen content are eliminated without any fragmentation [17]. Even with oxygen atoms present in the molecule structure, and H₂O used as co-reagent, no oxygen was found in the deposited material, suggesting that H₂ promotes Cu^II molecule dissociation and clean reduction to Cu⁰.

At constant pressure, 400 Pa, we defined kinetic/diffusion phases process windows. To control surface reactions, it is preferable to target a deposition temperature lower than 225 °C. Below 150 °C, no deposition is observed, the Cu(dmap)₂ molecule is not dissociated [18]. We estimated the activation energy around; Ea = 40 kJ.mol⁻¹. For comparison with another fluorine-free Copper(I) precursor, (mhd)Cu(BTMSA) was used for copper deposition at 133 Pa on PVD Ta/TaN, Ea is equal to 58 kJ.mol⁻¹ [19]. The lower activation energy we obtained with the Cu(dmap)2 on TiN substrates, indicates that the reaction occurred between the gaseous precursor, reagents, and the surface required less energy to take place.

The stress of the full MOCVD barrier/seed layer has to be limited to minimize the keep out zone around the TSV [20]. The stress was measured on 300mm silicon blanket wafers and reached −100 MPa (compressive). This MOCVD TiN/Cu stack deposited at a temperature ≤ 200 °C clearly conforms to 3D applications. The electrical resistivity was measured around 7 μΩ·cm, slightly higher than the bulk value but enough to ensure the next plating.

The MOCVD copper seed layer was deposited at 175 °C, on 300 mm structure wafer, with TSV AR12:1, corresponding to 10 μm diameter and 120 μm depth. We deposited 15 nm of copper in the field. As shown in Table 2, the film thickness increased along the sidewalls to reach a minimum of 76 nm and a maximum of 130 nm at the bottom corner, thanks to H₂/H₂O co-reagents, to increase molecule transport on the deep structure. The seed layer is continuous and very conformal along the sidewalls even over scallops. At the bottom corner of the TSV, the coverage exceeds 100% as shown in fig.7. The TSV bottom is a blind area: the precursor and co-reactant concentration, the deposition mechanism and by-products desorption are different from the TSV top.

Because of the high copper conformality without overhang, a successful electro-chemical deposition fill on AR 10 TSV was performed. FIB SEM X-section fig.8, performed after anneal, confirms the continuity of the copper film, produced a void-free TSV fill. These results highlight the transport and the nucleation/reaction of the gaseous precursor molecule down to the TSV base. The copper overburden post-plating is around 2μm compared with using an i-PVD seed layer, where a 1,5μm Cu seed thickness is required to ensure good plating of a 10μm×80μm TSV and the overburden is around 4μm.

Adhesion was evaluated using a tape test after anneal 400 °C for 30 min under N₂/N₂H₂ atmosphere. The test revealed strong adhesion, with no delamination observed, thanks to a 5nm thick intermetallic interface composed of randomly oriented crystallites TiN/Cu, as shown in, fig.9.

During the thermal treatment, the MOCVD TiN barrier deposited at 200°C, has effectively stopped the copper diffusion. The interface layer is formed by chemical reaction and diffusion between Cu and TiN leading to strong interfacial bonding. The modification of the surface energy caused by the presence of impurities as carbon, fluorine or oxygen, impact the wettability and the adhesion strength [21]. Since Ti has high oxygen affinity, native oxides on the barrier surface could hinder the interaction between Cu and TiN underlayer causing poor adhesion [22].

In our study, an anneal performed on the pure copper film deposited at low temperature using a fluorine-free precursor without vacuum break between Cu and TiN led to strong adhesion.

The exploitation of all peaks, allow to obtain the relative intensity of the surface of each peak according to the TSV depth. The “0 position” was arbitrary. The beam has a diameter of 50 μm, relatively important in comparison with the dimensions of TSV 10×100μm. The beam gradually runs through the sample thickness. Therefore, diffraction signal spread over 120–160μm.

After the electrochemical deposition, relative intensities of XRD reflections show a in-depth texture variation (fig. 10a). It should be noted that with this transmission configuration, the measured diffraction peaks correspond to lattice planes which are almost parallels to the sample surface.

The MOCVD copper seed layer deposited on TSV 10 × 100 μm is a polycristalline film on the TSV sidewalls. Intensities are normalized to the most intense peak. For copper powder randomly textured, according to the ICCD database, relative intensities values are 100 % for <111>, 46 % for <200>, 20 % for <220> and 17 % for <311> [23]. Here the more intense peak at 100 %, corresponds to the <111> reflection. Relative intensity values of <200>, <311> and <220> peaks are respectively; 30 %, 20 % and 10 % at the TSV top and increases progressively along the depth profile to 50 %, 30 % and 65 % respectively. This indicates that the texture is close to a randomly textured polycristalline film at the top surface while it turns towards a <220> prefered orientation at the base.

After the anneal (fig. 10b), the sample is randomly textured without any texture variation along the TSV sidewalls. The weak preferential orientation remains <111> direction, but the MOCVD copper film structure is more homogeneous and we observed a rearrangement of grain orientation, especially at the TSV base.

We performed a comparison using an i-PVD deposition technique for the seed layer on 10×80μm TSV, as shown in fig.11. The copper film is also polycrystalline and the most intense peak is also the <111> with the other reflections below 20% at the TSV top indicating a preferred <111> texture. The i-PVD seed layer seems to be more homogenous along the sidewalls, with more grains oriented in <200> at the base, as illustrated in fig. 11 a.

After anneal, grain rearrangement is clearly different from the MOCVD copper seed layer. Here the i-PVD film has a more heterogeneous texture composition and for approximately all the depth profile as shown in fig. 11b, the higher relative intensity is not the <111> direction anymore but the <200>, specific of the influence of thermal stress on texture.

Relative intensity values of <220> and <311> peaks are respectively 45% and 30%. The size of the crystallites is related to the peak brodening but it is also due to other parameters: defects, dislocations or impurities and microstrains.

The peak width provides information on the microstructure crystalline quality of copper. The larger the peak, the worse the film quality. The macrostrain is deduced from the peak position. Assuming the copper is pure, as shown in fig.6, peak position enable to calculate a strain (ε = (d-d₀)/d₀) by measuring an inter-reticular distance d and compare this value to the reference value for pure copper d₀. As shown in fig.12, CVD copper film strain decreased significantly along the TSV sidewalls. Strain range is from 0.6% TSV top to 0.1% at the TSV base, while the constant peak width along the TSV length indicates a homogeneous film quality.

In the case of the i-PVD seed layer, the strain distribution profile is linked to the deposition technique used. Using physical deposition, a thicker film is required to ensure film continuity. Since the deposition is directionnal, more material are present on TSV top and base compared with lower sidewalls.

Ionised-PVD helps to improve the poor conformality on lower sidewalls, applying RF bias on substrates.

Accelerated particles are energetics enough to lead to a re-sputtering effect on TSV base and increase the coverage by sputtering material from the base to the lower sidewalls.

The energy provided by ion bombardment increases the surface temperature, their impacts on the growing films could alter the microstructure [24], and could explain the higher stressed observed on TSV top and base.

The copper seed layer texture, grain size, and orientation is influenced by the deposition technique used, and impact the electrochemical deposition mechanisms growth. This will be described in a next paper.

MOCVD pure copper, oxygen and carbon free, was deposited at 175 °C on 300 mm wafers, using bis(dimethylamino-2-propoxy)copper(II), with H₂/H₂O as co-reagents.

The deposition technique used with this organometallic precursor offers very high conformality for high aspect ratio TSV. A strong adhesion was observed between the Cu and the TiN barrier film. The copper seed layer was integrated in 10 μm × 100 μm TSV.

By using XRD protocol adapted to a 3-dimensional structure, we observed higher homogeneous crystalline copper film deposited by MOCVD along TSV sidewalls compared with the same from i-PVD copper. Copper film orientation <111> along the sidewalls as deposition and post E-plating and anneal offers a good criteria for electromigration. The thinnest film and the deposition technique used could influence electroplating mechanisms and the microstructure quality of the copper layer.

Reducing the copper seed layer thickness for TSV metallization, presents other advantages for this MOCVD approach; decreasing significantly film constraints and cost with regard to the deposition step and subsequent chemical mechanical polishing step. The precursor used is a good candidate for copper seed layer required for TSV-middle and TSV-last applications.

Authors would like to thank CEA-Léti and particularly: AUDOIT Guillaume, FABBRI Jean-Marc for XRD sidewalls analysis sample preparation, BERNIER Nicolas for TEM analysis, BOROWIK Lukasz for XPS analysis, FAVIER Sylvie and RICHY Jérôme for XRD analysis, GOTTARDI Mathilde, RIBIERE Céline, ROMERO Gilles, and PHILIP Pierre-Emile for electro-deposition TSV filling, CHAUSSE Pascal for relevant advises and CHERAMY Séverine for the support, this work has been performed within the frame of the IRT Nanoelec Program on Léti's 3D300mmPiloteLine.