Low-Temperature Atomic Layer Deposition of Copper Films Using

Jun 4, 2014 - Components Research, Intel Corporation, Hillsboro, Oregon 97124, United States. Chem. Mater. , 2014, 26 (12), pp 3731–3738. DOI: 10.10...
2 downloads 0 Views 7MB Size
Article pubs.acs.org/cm

Low-Temperature Atomic Layer Deposition of Copper Films Using Borane Dimethylamine as the Reducing Co-reagent Lakmal C. Kalutarage,† Scott B. Clendenning,‡ and Charles H. Winter*,† †

Department of Chemistry, Wayne State University, Detroit, Michigan 48202, United States Components Research, Intel Corporation, Hillsboro, Oregon 97124, United States



S Supporting Information *

ABSTRACT: The atomic layer deposition (ALD) of Cu metal films was carried out by a two-step process with Cu(OCHMeCH2NMe2)2 and BH3(NHMe2) on Ru substrates and by a three-step process employing Cu(OCHMeCH2NMe2)2, formic acid, and BH3(NHMe2) on Pd and Pt substrates. The two-step process demonstrated self-limited ALD growth at 150 °C with Cu(OCHMeCH2NMe2)2 and BH3(NHMe2) pulse lengths of ≥3.0 and ≥1.0 s, respectively. An ALD window was observed between 130 and 160 °C, with a growth rate of about 0.13 Å/cycle. Atomic force microscopy (AFM) and scanning electron microscopy (SEM) revealed rough Cu films that likely originate from the Cu nanoparticle seed layer. The Cu films exhibited poor electrical conductivity because of their nanoparticulate natures. The three-step process showed self-limited ALD growth on Pd and Pt at 150 °C with Cu(OCHMeCH2NMe2)2, formic acid, and BH3(NHMe2) pulse lengths of ≥3.0, ≥ 0.3, and ≥1.0 s, respectively. ALD windows were observed between 135 and 165 °C on both Pd and Pt, with growth rates of 0.20 Å/cycle on both substrates. Plots of film thickness versus number of cycles showed linear growth behavior on Pd with a growth rate of 0.20 Å/cycle up to 2000 cycles. By contrast, a similar plot for growth on Pt revealed nonlinear growth behavior, with a growth rate of about 0.4 Å/cycle up to 500 cycles, and then a growth rate of about 0.03 Å/cycle between 500 and 2000 cycles. The large difference in growth behavior between Pd and Pt substrates is proposed to occur by formation of a Cu/Pd alloy film and continuous catalytic decomposition of the BH3(NHMe2) by the surface Pd sites. By contrast, there is much less surface Pt in the growing Cu film, and catalytic decomposition of BH3(NHMe2) by the diminishing surface Pt as the Cu film grows leads to a decreased growth rate beyond 500 cycles. X-ray photoelectron spectroscopy reveals the formation of high purity Cu metal for all depositions, with low levels of C, N, O, and B. The Cu films on Pd and Pt showed smooth, continuous films at all thicknesses and had low electrical resistivities.



INTRODUCTION Copper (Cu) is the primary interconnect material in microelectronics devices, because of its low resistivity and resistance to electromigration.1 Cu is currently deposited in trenches and vias on top of a thin diffusion barrier by a two-step process that employs formation of a conformal Cu seed layer by physical vapor deposition, followed by Cu fill using electrodeposition.1 The smallest features in microelectronics devices are scheduled to reach 12 nm by 2016,2 which places severe demands upon the film growth techniques used in device fabrication. The atomic layer deposition (ALD) film growth method is ideally suited for future microelectronics manufacturing, since it inherently provides highly conformal thin films, even in high aspect ratio nanoscale features, and allows subnanometer control over film thicknesses.3 The Cu seed layer needs to be deposited by ALD to meet future conformality and film thickness uniformity requirements in microelectronics devices. In addition, the Cu metal should be deposited ideally at ≤160 °C to afford the smallest surface roughnesses, promote facile nucleation, and give continuous films even at thicknesses of a few nanometers.4 Thermal ALD is preferred over plasma ALD, © 2014 American Chemical Society

to avoid the loss of conformality arising from the recombination of atomic hydrogen on the walls of high aspect ratio features.4,5 Recent Cu ALD processes are beginning to address these demands. Examples of low temperature thermal Cu ALD processes include Cu(OCHMeCH2NMe2)2 (1)/ZnEt2 at 100− 150 °C,6 CuL2/ZnEt2 at 120−150 °C (L = N-ethyl-2pyrrolylaldiminate),7 and 1/formic acid/hydrazine at 100− 170 °C.8 However, the processes using ZnEt2 lead to Cu/Zn alloys through competitive thermal decomposition of ZnEt2, perhaps even at the lowest growth temperatures.7 The 1/formic acid/hydrazine process affords high purity Cu films,8 but requires the safe handling of hydrazine. To date, none of these low temperature Cu ALD processes has demonstrated the growth of smooth, conformal films at thicknesses of a few nanometers. Examples of Cu ALD processes at higher temperatures include Cu(thd)2/H2 at 190−260 °C (thd = Received: March 31, 2014 Revised: May 29, 2014 Published: June 4, 2014 3731

dx.doi.org/10.1021/cm501109r | Chem. Mater. 2014, 26, 3731−3738

Chemistry of Materials

Article

2,2,6,6-tetramethyl-3,5-heptanedionate),9 [Cu-(sBuNCMeNsBu)]2/H2 at 150−250 °C,10−12 Cu(hfac)2/alcohol at 300 °C (hfac = 1,1,1,5,5,5-hexafluoro-3,5-pentanedionate),13 CuCl/H2 at 360−410 °C,14 and CuCl/Zn at 440−500 °C.15 Cu ALD growth was reported using a Cu(I) β-diketiminate precursor and diethylsilane,16 but a later study showed that this process proceeds by a pulsed chemical vapor deposition (CVD) mechanism.17 Indirect routes to Cu films include reduction of ALD CuO by isopropanol,18 treatment of ALD Cu3N with H2,19 and reduction of ALD Cu2O by formic acid in conjunction with a ruthenium seed layer.20 Finally, plasmabased ALD processes include Cu(acac)2/hydrogen plasma (acac = 2,4-pentanedionate),21a Cu(thd)2/hydrogen plasma,21b 1/hydrogen plasma,21c a Cu(I) N-heterocyclic carbene precursor with hydrogen plasma,21d Cu(OCMeCHCMeNEt)2 and a structurally related precursor with hydrogen plasma,21e and the precursors Cu(OCMeCHCMeNEt)2 and the bis(aminoalkoxide)copper(II) precursor CTA-1 (supplied by Adeka) with hydrogen plasma.21f The syntheses of several promising classes of Cu(I) precursors have been reported recently, but ALD growth has not yet been described.22 We recently reported the ALD growth of metallic Ni, Co, Fe, Cr, and possibly Mn films using metal(II) α-imino alkoxide precursors and BH3(NHMe2).23 There is significant interest in Cu/Mn alloy films, since annealing of 150 nm thick 90/10 Cu/ Mn alloy films on SiO2 substrates at temperatures between 250 and 450 °C was reported to afford migration of the Mn atoms to the SiO2 interface to form a segregated 2−8 nm MnSixOy layer between the SiO2 and Cu layers.24 Most significantly, the MnSixOy layer served as a Cu diffusion barrier for up to 100 h at 450 °C.24c This work suggests that ultrathin Mn-based films can replace current nitride-based barriers (TaN, WNx) in future microelectronics devices. Growth of the Cu/Mn alloy films relied upon sputtering,24 which affords poor conformal coverage in the narrow and deep features of future microelectronics devices. To date, a thermal Cu/Mn alloy ALD process has not been disclosed. A plasma-based ALD process for Cu/Mn metal alloy films with up to 10% Mn was reported, using Cu(OCMeEtCH2NMe2)2, Mn(tmhd)3, and H2 plasma with substrate temperatures between 100 and 180 °C,25 but plasma processes may not be able to meet conformal coverage requirements in high aspect ratio features. A key breakthrough would be identification of a highly reactive reducing coreagent that can reduce the Cu(II) ion (E° = 0.3419 V26) and Mn(II) ion (E° = −1.185 V26) equally well. Within the above considerations, we report new Cu ALD processes that employ the highly reactive reducing coreagent BH3(NHMe2). A two-step ALD process employing 1 and BH3(NHMe2) affords Cu films within an ALD window of 130− 160 °C on Ru substrates, but the films have rough surfaces because of the agglomeration of the Cu atoms during the initial stages of growth. However, three-step processes using 1, formic acid, and BH3(NHMe2) give high purity, low resistivity Cu films within ALD windows of 135−160 °C on Pd and Pt substrates. Importantly, these three-step processes afford dispersed, smooth Cu films at film thicknesses as low as 2 nm. Hence, BH3(NHMe2) is a general reducing agent for first row transition metal precursors, which should facilitate the ALD growth of alloy films.

Chart 1. Structure of 1

precursor, since it has been widely used as a CVD and ALD precursor for Cu-containing films.6,8,27 We previously reported that 1 sublimes on a preparative scale (∼1 g) at 90 °C/0.05 Torr within ∼4 h with 89% sublimed recovery and 3% nonvolatile residue, and undergoes solid state decomposition between 185 and 188 °C.8 We have also described the ALD growth of Cu films from 1, formic acid, and hydrazine within an ALD window of 100−170 °C.8 We sought to identify reducing agents that can reduce 1 to Cu metal, and which should be sufficiently reactive to reduce other first row transition metal ions to the metals. Our search was restricted to Lewis base adducts of boranes, since many of these compounds are available commercially, most are volatile, and first row transition metal hydrides MH2 are known to undergo rapid reductive elimination of H2 to afford the metals.28 Solution reactions7,8,29 were used initially to screen borane adducts. Treatment of 1 with 5 equiv of BH3(NHMe2), BH3(NEt3), or BH3(NiPr2Et) at ambient temperature in tetrahydrofuran afforded immediate precipitation of Cu metal powders. Among these borane adducts, BH3(NHMe2) had the highest volatility (sublimes 70 °C/0.05 Torr) and could be delivered reproducibly in our ALD reactor. BH3(NHMe2) is reported to decompose at 130 °C to afford H2 and [BH2NMe2]2.30 Binary Process Growth Studies. Initial deposition studies were carried out using 1 as the Cu source and BH3(NHMe2) as the reducing agent. In these trials, 1000 cycles were used at a growth temperature of 150 °C, along with 3.0 s pulses of 1, a 5.0 s purge, 1.0 s pulses of BH3(NHMe2), and a 10.0 s purge. A range of substrates was employed in initial growth trials, including Si with native oxide, SiO2 (100 nm)/Si, Si−H/Si, Ru (5 nm)/SiO2 (100 nm)/Si, Pd (15 nm)/Ti (2 nm)/SiO2 (100 nm)/Si, Pt (15 nm)/Ti (2 nm)/ SiO2 (100 nm)/Si, Cu (10 nm)/Ta (3 nm)/Si, and TiN (20 nm)/Si. Initial deposition studies resulted in thin (2−3 nm), discontinuous Cu films on Ru (5 nm)/SiO2 (100 nm)/Si, TiN (20 nm)/Si, SiO2, Si with native oxide, and Si−H/Si. These results clearly indicated poor nucleation. To enhance nucleation, 50 cycles comprising a 20 s pulse of 1, a 5 s nitrogen purge, a 1 s pulse of BH3(NHMe2), and a 10 s purge were carried out. This nucleation step was developed by trial and error, and is similar to the one used in our previous report of Ni, Co, Fe, and Cr ALD.23 Figure 1 shows a scanning electron microscopy (SEM) top view of a Ru (5 nm)/SiO2 (100 nm)/Si substrate after the 50 cycle nucleation sequence at a growth temperature of 150 °C. The nucleation sequence leads to separated nanoparticles, with diameters ranging from about 10 to 90 nm. The height of the nanoparticles is about 13 nm, as determined by side view SEM. The areas between the nanoparticles are likely covered by a thin Cu layer. The Cu nanoparticle growth may be facilitated by slow β-hydride elimination from the dmap ligands during the 20 s pulses of 1. However, no Cu nanoparticles were observed



RESULTS AND DISCUSSION Precursor Selection and Properties. Cu(OCHMeCH2NMe2)2 (1) (Chart 1) was chosen as the Cu 3732

dx.doi.org/10.1021/cm501109r | Chem. Mater. 2014, 26, 3731−3738

Chemistry of Materials

Article

saturation with 1 occurs at ≥3.0 s pulse lengths. A saturative growth rate of 0.26 Å per cycle was obtained, although this value includes the ∼13 nm of Cu nanoparticles documented in the nucleation step. Pulse lengths below 3.0 s afforded subsaturative growth. A similar plot of growth rate versus BH3(NHMe2) pulse length showed saturative growth with ≥1.0 s pulse lengths, and also afforded an apparent growth rate of 0.26 Å per cycle, which equates to 0.13 Å cycle when the Cu seed layer thickness is removed. Pulse lengths of 3.0 and 1.0 s for 1 and BH3(NHMe2), respectively, were used in the growth studies described below. The purge lengths were 5.0 and 10.0 s after the pulses of 1 and BH3(NHMe2), respectively. The long purge time after the BH3(NHMe2) pulse was used to minimize B incorporation in the films. Figure 3 shows the dependence of growth rate on substrate temperature. A constant apparent growth rate of about 0.26 Å

Figure 1. SEM top view of the Ru substrate after the 50 cycle nucleation step at a growth temperature of 150 °C.

in the absence of a BH3(NHMe2) pulse, so this situation remains unclear. The 50 cycle sequence led to improved nucleation on Ru (5 nm)/SiO2 (100 nm)/Si, TiN/Si, SiO2, Si with native oxide, and Si−H/Si substrates, but the deposits on Ru (5 nm)/SiO2 (100 nm)/Si were the densest. Accordingly, the deposition study described below was carried out on these Ru-coated substrates. Cu film growth studies were carried out on Ru (5 nm)/SiO2 (100 nm)/Si substrates that were subjected to the 50 nucleation cycles described above prior to regular ALD growth. The growth behavior was investigated by changing the precursor pulse lengths, substrate temperatures, and number of deposition cycles. A plot of growth rate versus pulse length of 1 at 150 °C is shown in Figure 2. In these experiments, the number of deposition cycles, length of BH3(NHMe2) pulse, and lengths of the purge cycles after the pulses of 1 and BH3(NHMe2) were kept constant at 1000 cycles, 1.0, 5.0, and 10 s, respectively. In ALD, a constant growth rate is observed once sufficient precursor has been passed over the substrate to use up the surface reactive sites.3 According to Figure 2,

Figure 3. Plot of growth rate versus deposition temperature on a Ru substrate. Each point corresponds to 50 nucleation cycles and 1000 regular growth cycles.

per cycle was observed over the temperature range of 130−160 °C. This temperature range is known as “ALD Window”, where the growth rate does not vary with the substrate temperature.3 Lower growth rates were observed below 130 °C, most likely due to diminished precursor reactivity. Higher growth rates were observed above 160 °C, presumably because of increasingly rapid gas phase thermal decomposition of 1. A plot of thickness versus number of cycles is shown in Figure 4. The substrate temperature for these depositions was 150 °C. There is a linear relationship between the film thickness and number of cycles. The slope of the line indicates a growth rate of about 0.12 Å/cycle, which fits the growth rate described in Figure 2 when the film thickness from the nucleation step is subtracted. The y-intercept in Figure 4 is about 12 nm, which is close to the 13 nm film thickness that was measured for the nucleation step by SEM. Characterization of Cu Films from the Binary Process. The compositions of films deposited at 120, 140, and 160 °C were assessed by X-ray photoelectron spectroscopy (XPS). These films were grown on Ru substrates with 50 nucleation cycles, followed by 1000 regular growth cycles. Similar compositions were obtained at all temperatures, and are summarized in Table 1. O contents of 2.5−7.9% were observed, even after sputtering. Low B contents of 1.2−3.1% were obtained after sputtering. High resolution scans of the Cu 2p

Figure 2. Growth rate versus pulse length of 1 at a substrate temperature of 150 °C on a Ru substrate. Each point corresponds to 50 nucleation cycles and 1000 regular growth cycles. 3733

dx.doi.org/10.1021/cm501109r | Chem. Mater. 2014, 26, 3731−3738

Chemistry of Materials

Article

Figure 4. Film thickness versus number of cycles at a substrate temperature of 150 °C on a Ru substrate. Each point corresponds to 50 nucleation cycles and the indicated number of regular growth cycles.

Table 1. Atomic Concentrations of Cu, C, N, O, and B Obtained by XPS after 120 s of Ar Ion Sputtering on the Indicated Substratesa 120 140 160 140 150 160 140 150 160 a

T (°C)

% Cu

%C

%N

%O

%B

(Ru, 2-step) (Ru, 2-step) (Ru, 2-step) (Pd, 3-step) (Pd, 3-step) (Pd, 3-step) (Pt, 3-step) (Pt, 3-step) (Pt, 3-step)

85.0 87.1 89.2 96.1 91.3 93.5 97.5 95.9 95.4

6.8 0.2 0.4 0.0 2.5 0.1 0.1 1.5 0.9

2.7 2.4 0.4 1.7 1.4 0.1 0.1 1.9 0.4

2.5 7.9 6.6 1.8 4.7 6.3 1.3 0.5 3.3

2.4 1.2 3.1 0.4 0.1 0.0 0.9 0.2 0.0

Figure 5. AFM images of Cu Film surfaces deposited at 120 °C (top, thickness = 17 nm, rms roughness = 5.8 nm) and 160 °C (bottom, thickness = 25 nm, rms roughness = 4.2 nm). The films were grown on Ru substrates with 50 nucleation cycles and 1000 regular deposition cycles.

The substrate signal is not included in the compositions.

The sheet resistivities of the films deposited between 120 and 160 °C ranged from 400 to 500 Ω/sq. These sheet resistivity values are close to those of the blank Ru substrates obtained upon annealing 120−160 °C under vacuum, and suggest poorly conductive Cu films. Additionally, the sheet resistivities of these films increased with time upon storage in air, suggesting slow oxidation of the films by ambient atmosphere. Cu nanoparticles are well-known to oxidize to Cu2O and CuO upon exposure to air.32 As outlined above, AFM and SEM images revealed that the films consist of densely packed nanoparticles, and XPS showed higher than expected levels of oxygen, even in sputtered samples. Accordingly, it is likely that the Cu particles on Ru substrates oxidize upon exposure to air to form Cu2O or CuO coatings on the surfaces of the constituent nanoparticles, leading to poorly interconnected and poorly conductive films. Formic Acid as a Nucleating Agent on Pd and Pt Substrates. As documented above, a successful nucleation sequence was developed that allowed ALD growth of Cu films on Ru substrates, but the films were rough due to the isolated nanoparticles in the seed layer. We have recently reported the growth of Cu films from 1, formic acid, and hydrazine, and were able to grow films on substrates such as thermal SiO2 and Si with native oxide without the need for any nucleation steps.8 Based upon this precedent, we postulated that formic acid might serve as a nucleation agent. Accordingly, the growth of

region revealed ionizations at 952.0 and 932.2 eV, which are very similar to the reference ionization values for Cu metal of 952.2 and 932.4 eV.31 However, Cu2O also shows Cu 2p ionizations that are very similar to those of Cu metal. Therefore, further analysis was required. A recent report showed that Cu metal shows an LMM Auger ionization at 567.7−567.9 eV, whereas the corresponding ionization for Cu2O appears at 570.0 eV.20 The Cu LMM Auger ionization for all films from the binary process appears at 568.0 eV, which confirms Cu metal. Powder X-ray diffraction experiments of the as-deposited films revealed that they were amorphous at all growth temperatures. A film deposited at 150 °C was annealed for 3 h at 300 °C under Ar. The annealed film revealed reflections for the (111), (200), and (220) planes of Cu metal (JCPDS file no. 04−0836). The surface morphologies of the as-deposited films on Ru substrates were examined by AFM and SEM. AFM images of 2 μm × 2 μm area of the films deposited at 120 and 160 °C are shown in Figure 5. The rms surface roughnesses of films deposited at 120 and 160 °C were 5.8 and 4.2 nm, respectively. SEM micrographs of the films deposited at these conditions also show rough surfaces. The SEM micrographs of the films after the nucleation step (Figure 1) showed the formation of separated nanoparticles, which results in rough surfaces after the regular ALD cycles. 3734

dx.doi.org/10.1021/cm501109r | Chem. Mater. 2014, 26, 3731−3738

Chemistry of Materials

Article

Cu films by a three-step process comprising 1, formic acid, and BH3(NHMe2) was investigated to determine if higher quality films would result. Initial depositions were carried out at 150 °C on the same range of substrates listed above for the two-step process. Specular Cu-colored films were obtained on the Pd (15 nm)/Ti (2 nm)/SiO2 (100 nm)/Si and Pt (15 nm)/Ti (2 nm)/SiO2 (100 nm)/ Si substrates, whereas the other substrates did not show improved film growth. Complete deposition studies were carried out on the Pd and Pt substrates, and full data are contained in the Supporting Information. Unlike the two-step process described above, films grown on Pd and Pt substrates by the three-step process did not require a separate nucleation step. Initial studies demonstrated saturative growth at 150 °C on the Pd and Pt substrates with pulse lengths of ≥3.0 s for 1, ≥ 0.3 s for formic acid, and ≥1.0 s for BH3(NHMe2). The saturative pulse lengths for 1 and BH3(NHMe2) are identical to those of the two-step process on the Ru substrates. Purge lengths of 5.0 s (after 1 and formic acid pulses) and 10.0 s (after BH3(NHMe2) pulse) were used. With 1000 cycles, films grown at 150 °C showed a saturative growth rate of 0.20 Å/cycle on both Pd and Pt substrates. Figure 6 shows the dependence of Cu film growth rate on temperature with a Pd substrate, 1000 cycles, and the optimized

Figure 7. Film thickness versus the number of deposition cycles on a Pd substrate by the three-step process at 150 °C.

Figure 8. Film thickness versus the number of deposition cycles on a Pt substrate with the three-step process at 150 °C.

experimental error of zero and suggests good nucleation at early stages of the deposition. By contrast, a similar plot for growth on a Pt substrate revealed nonlinear growth behavior (Figure 8). Between 250 and 500 cycles, a growth rate of 0.4 Å/cycle was observed, but the growth rate decreased to about 0.03 Å/ cycle between 500 and 2000 cycles. Cu films deposited by three-step process were characterized in detail. XPS analyses of 20 nm thick films grown at 140, 150, and 160 °C after Ar ion sputtering revealed metallic Cu. Compositions are listed in Table 1. The Cu films grown on Pd and Pt substrates were of higher purity than those grown on Ru by the two-step process. The Cu atomic concentrations ranged from 91.3 to 97.5%, and the highest values were obtained at 140 °C on both Pd and Pt substrates. Atomic concentrations of C (0.0−2.5%), N (0.1−1.9%), and B (0.0−0.9%) were low and showed no clear trend with deposition temperature or substrate. The uncertainties in these concentrations are probably 0.5 to 1.0%, which suggests that there is little difference among these values. Atomic concentrations of O ranged from 0.5 to 6.3%. The O concentrations were slightly higher on Pd substrates, and the levels increased with deposition temperature. On Pt substrates, the O concentrations were lower, and there was not a clear trend with deposition temperature. Overall, Cu films grown by the three-step process on Pd and Pt substrates are high purity, and have similar compositions to our previously reported Cu films grown from

Figure 6. Plot of growth rate versus deposition temperature of a Cu film grown by the three-step process on a Pd substrate with 1000 growth cycles.

pulse and purge lengths outlined above. An ALD window was observed between 135 and 165 °C, with a constant growth rate of 0.20 Å/cycle in this range. The related plot on a Pt substrate was identical. The ALD window range observed with the threestep process on Pd and Pt substrates is identical to that of the two-step process. However, the growth rate decreased at >170 °C in the three-step process, but increased at >170 °C in the two-step process. It is not clear why the growth rate drops at >170 °C in the three-step process, but a similar drop in growth rate was observed in our Cu ALD process employing 1, formic acid, and hydrazine.8 Cu(OCHO)2 decomposes thermally only at >200 °C,8 so the lower growth rates at >170 °C are unlikely to occur via Cu(OCHO)2 self-decomposition. Plots of film thickness versus number of cycles for growth on Pd and Pt substrates are shown in Figures 7 and 8, respectively. These films were grown at 150 °C using the optimized pulse and purge lengths outlined above. The plot for growth on Pd (Figure 7) varies linearly with the number of cycles, with a slope of 0.20 Å/cycle. The y-intercept was 3.4, which is within 3735

dx.doi.org/10.1021/cm501109r | Chem. Mater. 2014, 26, 3731−3738

Chemistry of Materials

Article

Cu films from the three-step process directly with those of sputtered Cu films. The initial stages of Cu film growth on the Pd and Pt substrates were investigated by examining top-down SEM micrographs after 50 and 100 cycles, at a growth temperature of 150 °C. On both substrates, no islanding and no surface structure was observed. The thicknesses of the films after 100 cycles are about 2 nm, which is about the lower measurement limit of our electron microscope. These findings suggest efficient Cu nucleation and smooth films even at early stages of growth. A key challenge in the ALD growth of Cu films is to obtain smooth, dispersed films at all thicknesses that can serve as seed layers for the electrochemical deposition of smooth, void-free Cu deposits. As described herein and in other deposition studies,6−22 Cu atoms on many substrates have high surface mobilities and can agglomerate even at low temperatures to form nanoparticles and resultant rough films. In the present work, the seed layers obtained after the 50 nucleation cycles with the two-step process at growth temperatures of 130−160 °C on Ru substrates consisted of isolated Cu nanoparticles with 10−90 nm diameters, which apparently formed by agglomeration of the Cu atoms. These rough seed layers then afforded rough films in subsequent Cu ALD growth. By contrast, ALD growth with the three-step process on Pd and Pt substrates led to smooth, dispersed Cu films at thicknesses as low as 2 nm (the smallest size that can be reliably measured with our SEM). Several studies have documented the growth of continuous, thin Cu films by ALD and related techniques. ALD growth from Cu2(sBuNCMeNsBu)2 and H2 at 200 °C showed continuous, densely nucleated Cu films on Co substrates at thicknesses of 1.4 and 2.0 nm.11,12 Reduction of ALD Cu3N films on Ru/Ta/SiO2/Si substrates with H2 gave electrically continuous films at thicknesses as low as 0.8 nm.19 In the present study, Ru substrates gave agglomerated Cu nanoparticles, which led to rough films at increased thicknesses. By contrast, Pd and Pt substrates led to smooth, continuous films. The ALD growth of Cu films on Pd and Pt substrates was reported using Cu(thd)2 and H2 at 235 °C.9 Notably, the authors demonstrated by XPS and transmission electron microscopy (TEM) that growth of Cu on the Pd substrates at 235 °C led to the formation of Cu/Pd alloys, with no clear Cu−Pd film interface and an approximately constant surface Cu/Pd ratio of 70:30 up to 1000 cycles. Hence, a Cu−Pd alloy was present at all film thicknesses in that work.9 In the same study, a combination of XPS and TEM showed that Cu films grown on Pt using Cu(thd)2 and H2 at 235 °C had a clear interface, with a small amount of Pt diffusion into the Cu film at the interface and no Pt on the surface of thicker films.9 In our Cu films grown on Pd, traces of Pd (140 °C, < 1%; 160 °C, ∼1%)) were observed on the surface of the as-deposited films by XPS. The lower amounts of Pd on the surfaces of our films grown at 140 and 160 °C are consistent with less Pd diffusion into the growing Cu films, compared to those grown at 235 °C with Cu(thd)2 and H2.9 Cu films deposited on Pt at 140 and 160 °C showed undetectable (