Self-forming Al oxide barrier for nanoscale Cu interconnects created by hybrid atomic layer deposition of Cu–Al alloy

Advanced technology nodes require a uniform barrier that is less than 5 nm thick, has good diffusion barrier properties, and adheres well to Cu. However, use of conventional physical vapor deposition (PVD) processes to synthesize barrier layers has not been completely successful, mainly because of poor step coverage.^1–3 Moreover, there is an unwanted increase in the resistivity of Cu wires occurs due to the size effect in metal films.^4,5 To overcome these problems, self-forming diffusion barrier layers have been proposed and studied extensively. Self-forming barriers, when fabricated by directly depositing a Cu–M (Mn, Ti) alloy on dense SiO₂ substrate followed by heat treatment, have been demonstrated to function as effective Cu diffusion barriers.^6,7 The self-forming barrier approach requires just one Cu–M alloy layer instead of two (Ta/TaN barrier and Cu seed layer). The thinner overall film thickness helps to minimize the overhang when a PVD process is used and also reduces the aspect ratio, which is of benefit in the Cu plating process as it promotes void-free filling. Achieving step coverage, accurate control of layer thickness, and uniformity of the alloy films are challenges that have to be met for commercial application of Cu alloy films as Cu interconnects. In previous studies, however, Cu alloy films were deposited by a PVD method. To obtain high step coverage, it is common sense that atomic layer deposition (ALD) is a suitable technique, because ALD uses a self-limiting film growth mode by surface-saturated reaction, which enables atomic-scale control of the film thickness with excellent step coverage.^8–10 We, therefore, investigated Cu–Al alloys created by alternating ALD as materials for a self-forming barrier.

Aluminum has been used for IC metallization in the past. The oxidation resistance, morphology, and thermal stability contributed by addition of small amounts of Al into bulk copper were well studied in the 1990 s.^11–14 According to the Al–Cu phase diagram, there is no Cu–Al binary phase when the Cu content is greater than ∼85%.¹⁵ A low concentration of Al is completely soluble in Cu. Al tends to segregate to film interfaces, including grain boundaries.¹⁶ Furthermore, Cu interconnects based on a Cu–Al alloy seed layer have been shown to have improved electromigration (EM) and stress migration (SM) performances.^16,17

Al has a more negative standard free energy of oxidation than Cu, Si, and C.¹⁸ Thus, Al is the strongest oxide former among these elements. To minimize the Gibbs free energy during annealing of a Cu–Al alloy on SiO₂ substrate, reduction of the oxide material occurs. This frees up oxygen, and aluminum oxide is expected to form at the interface of the Cu–Al alloy and SiO₂ substrate.¹⁹ A small amount of Al in Cu can also be directly plated by copper plating without the presence of a Cu seed layer. Those characteristics suggest that aluminum is a viable candidate for a self-forming Cu diffusion barrier that also meets the requirements for advanced Cu interconnects.

In this paper, we introduce an alternative ALD system for Cu–Al alloy deposition that can be used to generate Cu–Al alloy films that also contain oxygen atoms from the Cu precursor at a high deposition rate. The deposition process involves alternate deposition of Cu and Al layers using different ALD cycles in sequence to avoid gas phase chemical reactions.

Cu–Al alloys with various Al contents were deposited by plasma enhanced atomic layer deposition (PEALD) using bis(1-dimethylamino-2-methyl-2-butoxy)copper (MABOC) and trimethylaluminum (TMA) as copper and aluminum precursors, respectively. Cu–Al alloy films were deposited on 100-nm-thick SiO₂/Si plane wafers. To prevent precursor condensation in the delivery line, the line was heated from 85 to 95 °C. Figure 1 shows the pulse sequence used to deposit the Cu–Al alloy films. A subcycle consisted of step 1 (Cu cycle) and step 2 (Al cycle). One of the two deposition cycle (step 1) was comprised of four pulses: 5 s of a pulse of MABOC vapors (and N₂ gas), 10 s of a purge pulse with N₂ gas, 8 s of a pulse for H₂ plasma exposure, and 10 s of second purge pulse with N₂ gas. The other deposition cycle (step 2) was comprised of four pulses: 0.5 s of a pulse of TMA vapors, 5 s of a purge pulse with N₂ gas, 5 s of a pulse for H₂ plasma exposure, and 10 s of second purge pulse with N₂ gas. The Al concentration in the Cu–Al alloy films was controlled by the Cu and Al cycle ratio (Al cycles/Cu cycles). The Cu and Al cycle ratio was varied from 0.02 to 0.2 (Table I). The atomic concentration of the films was investigated by x-ray photoelectron spectroscopy (XPS, ThermoVG, SIGMA PROBE). Film resistivity was calculated from the sheet resistance measured with a four-point probe (JANDEL). After deposition of Cu–Al alloy films, annealing to promote the formation of AlO_x was performed in a vacuum at a pressure of 1.33 × 10⁻³ Pa. The annealing temperature ranged from 200 to 500 °C for 1 h. High-resolution x-ray diffraction (XRD; Rigaku D/max-2500/pc, Cu-Kα = 1.54062 nm, Japan) analysis was applied to examine the crystal structure of the deposited thin film. Standard θ-2θ XRD measurements were performed at 40 kV and 30 mA. A transmission electron microscope (TEM, JEOL JEM-2100 F, operating voltage = 200 kV) was used to investigate the film thickness and self-formed interlayer thickness. The Al content across the interfacial region was examined by energy-dispersive x-ray spectroscopy (EDS) line scanning on the cross-sectional TEM sample. The chemical state of the exposed interface after the Cu reverse-plating process was analyzed by XPS. To measure the leakage current density of Cu-alloy/SiO₂/Si, multilayer samples were formed by a lift-off process and the thermal stability of the copper diffusion barrier was tested. MOS capacitor structures, Cu, Cu–Al alloy, and reference Cu–Mn alloy films were used as gate electrodes. The electrode dot was 0.04 cm² and was formed by a lift-off process. Current–voltage (I–V) curves were measured using a HP4145 picoammeter/DC voltage source with the electric field swept from zero to 1.0 MV/cm. To evaluate adhesion before and after annealing of Cu–Al alloy films and Cu–Mn alloy films deposited on the SiO₂/Si substrate, a peel-off test using 3 M Scotch tape was performed.

Figure 2 shows the resistivity of the Cu–Al alloy films and atomic concentration of Al in the Cu–Al alloy films as a function of the Al cycle/Cu cycle ratio. Cu–Al alloy films were deposited with Al and Cu deposition cycle ratios of 0.02, 0.05, 0.1, and 0.2. The composition of Al in the Cu–Al alloy films was determined by XPS, and the relationship between the measured electrical resistivity and the number of Al cycles/Cu cycles was investigated by the four-point probe method. As expected, the amount of Al in the films increased with increasing Al/Cu cycle ratio. The Al atomic percent in the Cu–Al alloy films varied from 0 to 15.6 at. %. This means that the Al content in Cu can be controlled by the ratio of the number of Al cycles to Cu cycles. The electrical resistivity of the Cu film without Al incorporation was 28 μΩ cm. As the Al/Cu cycle ratio was increased, electrical resistivity of the films increased. In general, resistivity is affected by electron scattering caused by grain boundary scattering, surface scattering, impurity scattering, phonon scattering, and surface roughness-induced scattering. Cu–Al alloy films deposited using a higher Al/Cu cycle ratio showed increased electrical resistivity due to a severe increase in impurity scattering caused by the addition of a higher concentration of Al. Figure 2 shows that the resistivity of Cu–Al alloy films, which is one of the key properties that should be tunable for application of these alloy films as copper interconnects, could be controlled by varying the ratio of the number of Cu to Al cycles.

The XRD diffraction patterns of the postannealed Cu–Al (4.6 at. %)/SiO₂/Si samples are shown in Fig. 3. There was no other detectable phase besides Cu (111) and Cu (200). Also, no binary Cu intermetallic compound was detected for thermal annealing up to 400 °C. The similar result was reported recently by Perng et al.²⁰ They observed formation of (111)-oriented texture in a higher Cu–Al (3 at. %) film after annealing at various annealing temperatures, while no texture formation was formed in the Cu–Al (3 at. %) films. Meanwhile, the Cu signals become sharper as the annealing temperature increases, indicating Cu grain growth at an elevated temperature. In addition, Wong et al. had reported that a strong Cu(111) can enhance the adhesion between the Cu seed layer and diffusion barrier.²¹ For these point of view, enhancing the Cu (111) peak in the Cu–Al (4.6 at. %) alloy/SiO₂/Si samples after annealing can be beneficial for the performance of Cu interconnect.

Resistivity values of pure Cu and Cu–Al (4.6 at. %) films, as a function of the annealing temperature, are shown in Fig. 4. After annealing at 100–500 °C for 1 h, the resistivity of all films decreased; however, the reduction was only significant for the Cu–Al (4.6 at. %) film. We attributed the decrease in resistivity of the films to out-diffusion of Al atoms, annihilation of defects, and grain growth of the films. The small decrease in normalized resistivity of pure Cu film suggests that defect annihilation and grain growth only had a relatively minor effect on the decrease in resistivity. Therefore, the major reason for the decrease in resistivity of the Cu–Al alloy films was out-diffusion of Al atoms. Out-diffused Al atoms are expected to segregate at the surface of the Cu–Al alloy film and the Cu–Al alloy/SiO₂ interface to form AlO_x.

TEM cross-sectional images of Cu–Al (4.6 at. %) alloy/SiO₂/Si samples before and after annealing at 400 °C for 1 h are shown in Figs. 5(a) and 5(b), respectively. At the interface between the Cu–Al alloy film and SiO₂ film, an ultrathin layer started to form at 400 °C, while no such layer was seen in the as-deposited sample that did not undergo heat treatment. This self-formed layer was uniform with a thickness of approximately 2 nm. An EDS line scan of the region near this interface is shown in Fig. 5(c). The EDS line scan showed obvious accumulation of Al at the interface between the Cu–Al (4.6 at. %) alloy and SiO₂ film. Furthermore, it indicated migration of Al to the top surface. Al atoms on the top surface reacted with residual oxygen during annealing.

To examine the Cu–Al (4.6 at. %) alloy/SiO₂ interfaces, we have stropped the Cu–Al (4.6 at. %) alloy layer of the as-deposited and 400 °C annealed samples for 1 h by the Cu reverse-plating process. The Cu reverse-plating process ensures that the self-formed interfacial layer is intact. Figure 6 shows the XPS spectra of the Al 2p_1/2 and the O 1 s spectra of the bare self-formed layer. The XPS spectra of the samples with the 400 °C, 1 h heat treatment and without heat treatment have an Al 2p_1/2 peak at 74.2 eV, which is attributed to Al³⁺.²⁰ However, Al signals intensity at the sample with heat treatment is much greater than the sample without heat treatment. The oxygen binding energy from O 1 s in Fig. 6(b) is also slightly shifted from the Si–O bond (532.4 eV) to the Al₂O₃ bond (531.8 eV).¹⁹ These data indicate that stoichiometric Al₂O₃ is the self-formed layer. Formation of Al₂O₃ is due to the reaction of Al with SiO₂ and the reaction is driven by the reduction of overall free energy.¹⁸ Therefore, the reaction of Al and SiO₂ is spontaneous and Al additives in Cu–Al alloy films will segregate to the interface to form Al₂O_3.

To identify the barrier characteristics of the self-formed interlayer, leakage current was measured using a MOS structure. Leakage current densities versus electric field (I–E) data were measured using a probe station with a semiconductor analyzer (Agilent HP 4145B) at room temperature in the dark. Leakage current densities shown in Fig. 7 include those for Cu–Al (4.6 at. %) and Cu–Mn (4.7 at. %) alloys after annealing for 1 h at various temperatures. To better understand the ability of the Al self-formed interlayer to act as a barrier to Cu diffusion, we compared its performance to that of a Mn self-formed interlayer. Reported studies on the self-forming barrier properties of Cu–Mn alloy films have demonstrated that Mn is a suitable alloying element. Mn atoms segregate out easily of Cu–Mn alloys and react with SiO₂ to form a stable barrier layer.^6,22,23 Thermal stress conditions ranged from room temperature to 550 °C. Figure 7 shows leakage current versus applied electric field curves of the (a) Cu–Al (4.6 at. %) alloy/SiO₂/Si sample after 1 h of annealing at 300 °C; (b) Cu–Al (4.6 at. %) alloy/SiO₂/Si sample after 1 h of annealing at 400 °C; (c) Cu–Mn (4.7 at. %) alloy/SiO₂/Si sample after 1 h of annealing at 300 °C; and (d) Cu–Mn (4.7 at. %) alloy/SiO₂/Si sample after 1 h of annealing at 400 °C. Leakage current density of the Cu–Al (4.6 at. %) alloy/SiO₂/Si sample indicated that this sample had poorer thermal stability than the Cu–Mn (4.7 at. %) alloy/SiO₂/Si sample after preannealing for 1 h at 300 °C. We attribute this to the lower diffusivity and lower activity coefficient of the Cu–Al alloy than those of the Cu–Mn alloy. Because the diffusivity of Mn (4.0 × 10⁻¹⁶ cm²/s) in Cu is higher than that of Al (3.3 × 10⁻¹⁶ cm²/s) in Cu, Mn can more easily segregate at the SiO₂ interface.^24,25 Furthermore, Mn has a higher activity coefficient for Cu than Al. Therefore, a uniform Mn-based interlayer can form at a low annealing temperature. However, after preannealing for 1 h at 400 °C, the Cu–Al (4.6 at. %) alloy/SiO₂/Si sample showed comparable thermal stability to the Cu–Mn (4.7 at. %) alloy/SiO₂/Si sample. The main reason for this effect of preannealing temperature on thermal stability is that the diffusivity of Al depends on temperature. Al diffusivity increases exponentially according to temperature. Higher diffusivity means that more Al can reach the SiO₂ interface and react there with SiO₂ to form an Al-based interlayer.

Cu thin films fabricated using ALD techniques have poor interfacial adhesion to oxide dielectrics, such as SiO₂, causing failure of the copper interconnect system.^26–28 To evaluate the interfacial adhesion of Cu–Al (4.6 at. %) alloy films and reference Cu–Mn (4.7 at. %) alloy films to the SiO₂ substrate, a peel-off adhesion test using Scotch tape was performed before and after annealing at 300 °C for 1 h. After tape peel-off, as-deposited pure Cu, Cu–Al (4.6 at. %) alloy, and Cu–Mn (4.7 at. %) alloy films were completely removed; in contrast, almost none of the annealed Cu–Al (4.6 at. %) and Cu–Mn (4.7 at. %) alloy films was removed. Interlayer formation by annealing improved interfacial adhesion of the alloy film to the SiO₂ substrate without any failure in the peel-off adhesion test, as summarized in Table II. One of our previous studies indicated that a Cu–Mn (4.7 at. %) alloy annealed at the same temperature, 400 °C, for 30 min exhibited improvement of interfacial adhesion to oxide dielectric; similarly, the Cu–Al (4.6 at. %) alloy film described here possessed good interfacial adhesion to the SiO₂ substrate.²² Chemical interactions between contact layers improve interfacial adhesion. Therefore, adhesion of Cu alloy films to SiO₂ was improved by the interlayer formed by chemical interactions between the Cu alloy film and SiO₂.

We prepared Cu–Al alloy thin films by PEALD and repetition of subcycles consisting of Cu and Al ALD cycles at 150 °C. The Al content in the Cu–Al alloy films was controlled by changing the ratio of the number of Al cycles to the number of Cu cycles. To investigate the properties of the self-forming barrier layers formed by alloying, the resistivity, barrier, and adhesion characteristics of the Cu–Al alloys were investigated to evaluate their potential use as copper seed layers. TEM images revealed that an AlO_x layer formed at the interface between Cu–Al alloy films and SiO₂ film after annealing at 400 °C for 1 h. The thickness of the self-formed interlayer was about 2 nm. The self-formed interlayer provided superior thermal stability to that of Cu–Mn as a Cu diffusion barrier and acted as an adhesion layer between Cu and SiO₂. Thus, Cu–Al alloy films synthesized by PEALD can function as Cu seed layers and self-forming barriers, meeting the requirements of 2-nm nodes.

This research was performed with the support of SK Hynix Semiconductor, Inc. The authors would like to acknowledge the assistance of the TEM operators at the Industry-University Cooperation Foundation of Hanyang University with TEM analysis.