Area-selective atomic layer deposition (AS-ALD) has attracted a great deal of attention in recent years for self-aligned accurate pattern placement with subnanometer thickness control. The authors demonstrate a methodology to achieve AS-ALD by using undecanethiol and octadecanethiol masking layers deposited selectively in vapor phase on copper versus low-κ. Their approach has been demonstrated in combination with an in situ Hf3N4 ALD. In situ spectroscopic ellipsometer was performed to investigate the blocking ability of the thiols on the copper surface against ALD nucleation. A considerable growth inhibition up to 480 cycles for Hf3N4 deposited at 170 °C has been observed on the copper surface, while the same functionalization did not inhibit the nucleation on the dielectric surfaces.
Atomic layer deposition (ALD) is a well-established method to deposit a variety of thin film materials with excellent thickness control, uniformity, and conformality. These properties are inherited from the self-limited nature of ALD surface reactions where two gas precursors (occasionally more) are sequentially introduced to chemisorb in a saturative fashion, leading to a monolayer-by-monolayer growth. While ALD is primarily used to deposit conformal blanket films, its chemically driven growth can be intentionally inhibited if chemical sites suitable to the chemisorption of ALD reactants are not initially present. This inherent chemical specificity has been successfully exploited in the context of selective-area deposition (ASD) by locally passivating a surface so that ALD films are only added only where needed. This additive process, referred to as bottom-up approach, could lead to a paradigm shift from today's top-down VLSI fabrication by not only reducing the number of processing steps, but also by alleviating key challenges (and costs) associated with lithography and layer alignment at the sub-10 nm node.
The use of self-assembled monolayers (SAMs) as a chemically selective mask layer has been widely reported.1,2 SAMs indeed present several key benefits as effective barriers: by terminating these organic molecules with a nonreactive terminal group, typically methyl -CH3, these ultrathin (1–3 nm) organic layers can effectively lower the surface energy and block the potential chemisorption of ALD reactants. By selecting longer alkyl chains (C11–C18), high packing density can be maintained despite the presence of surface defects which would otherwise critically limit the SAMs performance as an ALD barrier.3 In addition, SAMs reactivity to specific surface ligands, e.g., -OH on oxides versus -H terminated on metals, can be effectively tuned. For example, silane headgroups exhibit a strong affinity to chemically bind with oxides over metals, while the reversed effect is observed with thiol-based groups.
As a result, ASD of a variety of oxides and metal ALD films have been reported, including TiO2 ALD on alkylsiloxane-passivated oxides4,5 and on alkanethiolate/gold,6–8 Al2O3 on hexadecanethiol/gold6,9,10 and octadecylphosphonic acid/copper,11 HfO2 on alkylchlorosilane/oxide,3 SnO2 on dodecanethiol/gold,6 Ni and Co metals on octadecyltrichlorosilane (ODTS)/oxide,12,13 Ir on octadecanethiol (ODT)/copper,14 and Pt on oxides.15 It should however be noted that the reported ALD retardation is relatively low, typically less than 50 cycles for oxide ALD (with exception of ZnO on C12SH-passivated gold where 100 cycle retardation was reported6) and about 100 cycles for metal ALD such as Pt or Ru (though metal ALD is inherently prone to significant nucleation delays in the first place).14 Additionally, the vast majority of reported works relate to SAMs deposited either via sample immersion into a diluted solution or via microcontact printing where the liquid SAMs is patterned on the surface by a polydimethylsiloxane (PDMS) stamp.5 Both approaches present significant limitations, particularly in the context of process integration and scaling. Solution-based functionalization tends to require long immersion times from 24 to 48 h, are prone to copolymerization reactions leading to defect formation, and are ill-suited for conformal coating in high aspect ratio nanostructures. Microcontact printing allows to directly transfer a patterned SAMs from a PDMS stamp on a planar surface but is also inadequate for three-dimensional passivation.
Vapor-phase deposited SAMs (v-SAMs) provide key benefits in terms of cycle times, conformality and ease of integration (to be further detailed in Sec. III G). However, its use has been primarily limited to shorter and more volatile precursors (C8-C12 SAMs) for hydrophobic and antistiction control in MEMS.16 In the context of ASD, v-SAMs reports are scarce. Hong et al. deposited ODTS and tridecafluoro-tetrahydrooctyltrichlorosilane (FOTS) on silicon oxide in vapor phase.17 Following 48 h exposures to ODTS and 6 h to FOTS, tetrakis(dimethylamido)hafnium [Hf(N(CH3)2)4 or TDMAHf] and water were pulsed-cycled at 250 °C with 3 min purge steps. Under these intentionally slow ALD cycle times, HfO2 growth was inhibited for up to 50 cycles. ASD of iridium metal and polyimide on copper substrates using vapor-phase deposited C12 alkyl thiol was reported by Farm et al.14 Dodecanethiol vapor [CH3(CH2)11SH] heated at 55 °C was pulsed 360 times for 5 s (per pulse). The corresponding 30 min SAMs processing time was the shortest exposure required to inhibit 500 cycles of iridium ALD at 250 °C (deposited in same reactor) and 20 cycles of polyimide at 160 °C with 20 s purge times. Thiol exposure for inhibiting the polyimide growth had to be extended to 20 h.
Avila et al.6 deposited dodecanethiol on gold. Using the same reactor, several oxides (Al2O3, TiO2, SnO2, and ZnO) were deposited by ALD from 100 to 250 °C, with 20–30 s purge times. Although the ALD growth retardation was consistent with previous works done in liquid phase (<50 cycles with exception of ZnO at 100 cycles), the work is noticeable as mass measurements by in situ quartz crystal microbalance during the ODT exposure showed that the packing density of the SAMs films achieved in vapor-phase was as good or better than in the liquid phase. Moreover, ODT saturation was achieved in a relatively short time (20 min). The same group followed up on Avila's work more recently.9 They showed that in situ H2-plasma pretreatment of the gold substrate lead to a significant improvement of the ALD growth inhibition using a long chain dodecanethiol as a masking layer and could even block Al2O3 from trimethylaluminum (TMA) and water for up to 20 cycles. Hashemi et al. also demonstrated that densely packed thiol SAMs layers could be deposited on copper upon short vapor exposures, effectively blocking the growth of ZnO and TiO2 ALD for several hundred cycles with the added benefits of regenerating the thiol SAMs from the vapor in between ALD cycles.8
In the current work, we report our results for the growth of thiols deposited in vapor-phase on copper and low-κ in order to selectively inhibit the deposition of hafnium nitride by ALD. Such bottom-up approach aims at self-aligned deposition of dielectrics on dielectrics in order to address the critical challenges associated with lithographic steps in future nodes. Using in situ spectroscopic ellipsometry (iSE), process dynamics were characterized in real-time at different critical stages of the ASD process flow, including in situ reduction of the native copper oxide, deposition of thiol mask layers with C11 and C18 SAMs and its impact on following deposition of Hf3N4 by ALD. The benefits (and limitations) of vapor-phase deposited SAMs, particularly in the context of its fast deposition kinetics, and the implications for its potential integration into a single-tool platform, will also be discussed.
Depositions of C11 undecanethiol (UDT) and C18 ODT SAMs, as well as Hf3N4 ALD were carried out in a single commercial ALD platform, a Veeco CNT Savannah S200 reactor. In its single wafer configuration (up to 200 mm diameter), the reactor can either be used in continuous flow mode, thus enabling minimum residence times suitable for ALD processing, or in static mode, referred to as Exposure Mode™, ideal for SAMs deposition in vapor phase. Under Expo conditions, long residence times can be achieved by isolating the reactor from the vacuum pump.
Ten milliliters of Sigma Aldrich 98% 1-undecanethiol (C11H23SH) and n-octadecanethiol (C18H37SH) were loaded in 50 ml stainless steel containers and heated, respectively, to 80 and 130 °C. C11 and C18 vapors were introduced into the reactor via a SAMs delivery kit. SAMs doses were accurately quantified by feeding the SAMs vapor into a 150 ml accumulator up to a recipe-defined pressure setpoint. The accumulator and coupled capacitance gauge were temperature controlled to 100 °C for UDT and 160 °C for ODT. The actual dose (in micromolars) was readily determined from the pressure change inside the accumulator during the SAMs pulse, the known accumulator volume and its temperature.
Hf3N4 ALD films were deposited at 170 °C with ammonia and TDMAHf, sequentially pulsing NH3 for 0.05 s and TDMAHf for 0.2 s pulses with a 5–15 s purge step. In situ thickness measurements were continuously acquired with a Woollam M2000V spectroscopic ellipsometer over 390–1000 nm spectral range at a scan rate of 3 s. Direct optical path to the center of the wafer was provided through two N2-purged silica windows integrated in the S200 lid. The SAMs films were optically modeled with a basic Cauchy model, while a Tauc–Lorentz model was used to model the native copper oxide.
Different types of substrates were used for this study: 100 nm thermal SiO2 on Si, 90 nm plasma enhanced chemical vapor deposited dense low-κ (κ value ∼ 3.0) on Si (provided by IMEC), copper deposited by electrochemical deposition (ECD) and planarized by chemical mechanical planarization, and PVD deposited copper. It is to be noted that unless specified otherwise, these samples were processed as-is with no precleaning and loaded into the hot-wall S200 at 170 °C where they were pumped down within seconds to limit oxidation of the samples. Once at 0.1 Torr with 20 sccm N2 flowing, the native copper oxide thickness, as measured by iSE, appeared stable over extended period.
The wetting properties of the films were evaluated ex situ via static sessile drop measurements of water contact angle (WCA) using a Rame-Hart 200 goniometer.
A. Copper oxidation kinetics
One of the main difficulties with the use of copper substrates resides in its rapid kinetics of oxidation which can significantly complicate process flows and impede the run-to-run process reproducibility, particularly when working in a noninert environment. It is well established that metallic copper Cu(0) readily oxidizes to form cuprous oxide Cu2O (Cul) and cupric oxide CuO (CuII).18 Exposure to air also readily leads to the formation of copper hydroxide Cu(OH)2 and copper carboxylate. The formation of these-passivating oxides has been a known challenge for copper-based IC manufacturing as they reduce the adhesion of barrier films and impact the electro-migration behavior. In the context of SAMs, it is fair to assume they will play a critical role in the chemisorption of the SAMs head group with the copper ligands.
To characterize the kinetics of oxidation formation at various temperatures, ECD copper samples were dipped for 120 s in glacial acetic acid, a mild organic CuOx etchant,19 and rinsed in ethanol. The samples were then rapidly loaded (within 10 s) onto the heated reactor and remained exposed to air for up to 40 min. The iSE data acquired continuously during the oxidation are presented in Fig. 1 for temperatures ranging from 30 to 170 °C. In the inset, the time scale is extended from 60 to 2400 s to illustrate the passivating effect of the oxide layer. It is noticeable that while the copper remained mainly oxide free at 30 °C for 30 s or so, its oxidation at 50 °C and higher was almost instantaneous. At 170 °C, the temperature to be used in the upcoming steps for thiol SAMs and Hf3N4 ALD, 5 Å or so were formed within 10 s. It is to be noticed that once under vacuum with 20 sccm of nitrogen flow, no further oxidation of the copper was observed even over extended period.
Corresponding variations of the water contact angles for as-is versus ex situ cleaned copper are also presented (Fig. 2) after 2, 45, and 1000 min (∼17 h) in air at room temperature, showing the progression from hydrophilic copper surface (32° WCA 2 min after cleaning) to hydrophobic copper oxide (83° at 17 h). This is consistent with prior results showing that low surface energy Cu2O (80° WCA) grows more abundant over the reduced copper/CuO, which forms after a short air exposure. Over time, the surface might adsorb carbon resulting in a further decrease in the surface energy.20
B. Monolayer deposition of UDT and ODT on wet-cleaned copper
Alkanethiols were selected as masking layers on copper due to: (1) the strong affinity of mercapto groups (-HS) and long-established ability to form self-assembled layers on copper;21 (2) their reasonably high vapor pressure above 0.1 Torr (at 80 °C and above). C11-UDT benefits from a noticeably higher vapor pressure over bulkier C18-ODT, enabling faster replenishing of the SAMs accumulator and facilitating run-to-run dose reproducibility. Therefore, UDT deposition in vapor phase was first investigated before studying ODT.
ECD-copper chips were cleaned via 60 s dip in 1% HCl followed by ethanol rinse and nitrogen blown dry. Samples were loaded directly at 120 °C and put under vacuum within seconds to minimize copper reoxidation. The resulting CuO thickness was measured by in situ SE prior to the SAM exposure and determined to be around 4–5 Å. The UDT vapor was then built up to 0.25 Torr inside the SAMs accumulator (typically in less than 10 s), pulsed for 2 s into the reactor where it was maintained under Exposure Mode condition for 900 s. The calculated doses based on the pressure change in the accumulator during the pulse were close to 1 μmol.
Figure 3 shows the thickness as measured by iSE during the 900 s SAMs exposure as well as the following 100 s during the following UDT pump-out sequence. UDT on copper was introduced in two different ways, i.e., a single 900 s expo (1 × 0.9 μmol dose) or three consecutive 300 s expos (3 × 0.9 μmol doses). The respective depositions led to 15 and 16 Å thickness plateaus and the minimal additional uptake during the second and third exposures were consistent with a SAM self-limited growth. Respective WCAs at 105° and 107° were also indicative of the copper surface being successfully functionalized with hydrophobic UDT. As expected, UDT exposure on SiO2 showed a minimal uptake of less than 1 Å during the exposure. The immediate film desorption upon the subsequent pump-out suggests a weakly physisorbed layer on SiO2. Postprocess WCA measurements showed that SiO2 remained hydrophilic with a contact angle below 20° further indicating that the thiol molecules did not effectively bind to the silicon oxide.
Deposition of ODT on HCl-cleaned copper is also shown in Fig. 3. Due to the slow evaporation rate of ODT, the precursor temperature was raised to 130 °C for ODT. To prevent any upstream condensation, the SAMs accumulator temperature was accordingly raised from 100 to 160 °C and the reactor temperature from 120 to 170 °C. Similar to C11 UDT, the C18 thiol exposure also resulted in a self-limited growth with a 25 Å film thickness consistent with the ratio of chain length between C18 and C11 thiols. WCA for ODT treated copper ranged from 110° to 112° indicative of proper surface functionalization.
C. Hf3N4 ALD on self-limited UDT and ODT
Immediately following the SAMs depositions, a Hf3N4 film was deposited by ALD in the same chamber at 170 °C using TDMAHf and NH3. Figure 4 shows the Hf3N4 thickness obtained from iSE on as-is and UDT-treated SiO2 as well as on SAMs-free copper. Hf3N4 on SiO2 grows linearly with no apparent nucleation delay. UDT-treated SiO2 presents a minor nucleation delay of less than 5 cycles indicating that the UDT treatment had a minimal effect on SiO2 (consistent with Fig. 3 results). Interestingly, iSE data on copper exhibited an initial “dip” during the first 6 ALD cycles before rapidly transitioning to a linear growth similar to the one observed on SiO2. The origin of this dip will be further discussed in Sec. III D, but the key insight is that Hf3N4 ALD presented either no or minimal nucleation retardation on thiol-coated SiO2 and thiol-free copper.
In contrast, depositions of Hf3N4 on HCl-cleaned copper immediately after the growth of UDT or ODT SAMs are shown in Fig. 5. The nucleation delay on 1.6 nm UDT was about 40 cycles followed by a slow growth regime (∼1 Å/cycle) from 40 to 60 cycles and full growth (2 Å/cycle) after that. The growth inhibition was extended from 40 to 75 cycles on the 2.5 nm ODT. Based on Figs. 4 and 5, the selectivity of Hf3N4 on copper versus SiO2 treated with UDT and ODT correspond, respectively, to thicknesses of 8 and 15 nm.
D. In situ reduction of copper with metalorganic precursors and ethanol
The observation of the initial dip during the deposition of Hf3N4 on as-is copper led to further investigation of its origin. Copper samples with about 23 Å native copper oxide CuO were therefore exposed at 170 °C to 25 consecutive TDMAHf pulses, 0.3 s long and 20 s apart. Similar exposures using common ALD metalorganic precursors, i.e., TMA and tetrakis(dimethylamido)zirconium (TDMAZr) were also used. Psi and delta (Ψ,Δ) measurements by the iSE during the TDMAHf pulses are shown in Fig. 6 over the 400–700 nm spectral range.
Both Ψ and Δ measurements vary during the first 15 pulses but remain stable beyond 20 pulses. This behavior could be caused by either a reduction reaction across the entire oxide thickness or a self-limited, likely diffusion-limited, reaction over the top oxide layer. Insights on the reaction mechanisms can be found in the works of Abdulagatov et al.22 and Gharachorlou et al.23 Using a quartz crystal microbalance, Abdulagatov et al. investigated the growth of Al2O3 by ALD on copper and copper oxide using TMA and water. Gharachorlou et al. followed up by studying specifically how TMA interacts on copper oxide using high-resolution electron loss spectroscopy (HREELS), scanning tunneling microscopy (STM), and x-ray photoelectron spectroscopy (XPS). HREELS data backed up by XPS show that TMA adsorbed and dissociated on the copper oxide Cu2O/Cu(111), leading to a reduction from Cul to Cu0 and the formation of copper aluminate, likely CuAlO2. Interestingly, STM analysis revealed that reduced copper formed islands surrounded by CuAlO2. This reaction was shown to be limited by the amount of oxygen present in the copper oxide.
Based on Gharachorlou's results, it is assumed that the (Ψ,Δ) changes observed during exposures of the native copper oxide to metalorganic pulses are linked to the same reduction mechanism. From an ellipsometry point of view, modeling a set of optical layers where the oxide is reduced into a mixture of copper metal and metal cuprate proved challenging and beyond the scope of this work. A reasonable “goodness-of-fit” was obtained using a Tauc–Lorentz model fitted over the top CuO layer. Figure 7 shows the decrease in that oxide layer over 25 metalorganic pulses (corresponding to 500 s).
Accordingly to (Ψ,Δ) data, the copper reduction was self-limited in all cases though faster for TMA than TDMAZr or TDMAHf. The cause for the differences between the remaining oxide thicknesses after 25 pulses for the three treatments was not clearly established but is likely related to model inadequacy since the metalorganic insertion (Al, Zr, or Hf) is not considered. From these data, we could not fully collaborate if all the copper oxide was reduced (as inferred by Gharachorlou) or if the reduction reaction was simply self-limited. Water contact angle measurements performed after the in situ reduction step were also consistent with a reduction of the copper oxide as the WCA decreased from 92° for the initial Cu/CuO to 31°. For reference, WCA on 1% HCl-cleaned copper was 32°.
In situ gas phase reduction presents key benefits when working with copper, and though the reduction of copper oxide by exposure to standard ALD precursor is attractive, the introduction of metal atoms into the oxide layers are not acceptable during device fabrication. Therefore, other in situ reduction methods were investigated. Removal of the copper oxide with vapor-phase ethyl alcohol (C2H5OH) was reported by Satta et al.24 Ten milliliters of 200-proof ethyl alcohol was loaded into a 50 ml precursor container for gas-phase delivery into the S200. As-is PVD-copper samples were loaded at temperatures ranging from 175 to 250 °C. Ethanol vapor was pulsed for 2 s at 10 s intervals with a 50 sccm N2 purge for up to 150 pulses. The initial thickness of the copper oxide ranged from 4 to 5 nm with a thicker oxide at higher temperatures as the samples were loaded from atmosphere onto the hot reactor chuck before being pumped down within 5 s. The data in Fig. 8 clearly indicate the temperature-dependent kinetics of the oxide reduction. At 250 °C, copper oxide is rapidly reduced within 40 pulses. Consistent with data from Satta, the reduction is most effective above 225 and at 175 °C, the copper oxide thickness only decreases by 5 Å after 150 pulses. It also to be noted that the excellent fit of a Cu/CuO bilayer using a standard Tauc–Lorentz model suggests a reduction process without the introduction of a third metal-cuprate phase as previously observed. Despite those benefits, the higher temperature required for the reduction of copper oxide does complicate the process flow as the subsequent SAMs deposition cannot be done above 180 °C due to excessive thermal decomposition. Although solutions exist to circumvent this issue (two-step process at different temperatures, clustered system where etch/deposition are done separately), the remaining of the work was carried out with the metalorganic reduction step.
E. Non self-limited ODT on in situ cleaned copper
As-is copper samples were loaded at 170 °C. The native copper oxide was partially reduced in situ with 80 consecutive TDMAHf pulses and immediately exposed to a single 0.5 μmol ODT dose under static expo conditions for 900 s. iSE thickness measurement are shown in Fig. 9. A rapid initial thickness intake of about 10 Å occurred within the first 5 s followed by a relatively flat thickness plateau over the next 200 s. However, contrary to the saturating growth observed on reduced copper via ex situ HCl clean (Fig. 5), the ODT thickness increased steadily at a rate close to 2 Å/min, indicating a multilayer film. WCA above 110° indicate that the uppermost layer of that film remained ordered with the nonpolar alkyl functional groups remaining closely packed and oriented outward.
The deposition of multilayer thiols on copper has been previously reported and associated with the presence of copper oxide CuO or copper hydroxide Cu(OH)2 on the surface.25,26 Using IR reflection absorption spectroscopy and XPS, Keller et al. showed that octadecylthiol deposition over copper and its native oxide CuO lead to a thick (20–90 nm) multilayer system formed by conversion of the copper oxide into copper thiolate complexes (CuSR). The proposed reaction mechanism was based on a reduction of CuII present in CuO cupric oxide and Cu(OH)2 copper hydroxide upon exposure to thiols, followed by an oxidation of thiols to the corresponding disulfide (RS-SR). The authors also showed that while the alkyl tails of chemisorbed thiols were randomly oriented in the film bulk, the uppermost layer remained ordered with alkyl chains perpendicular to the surface. As a result, that self-assembled multilayer films exhibited a high hydrophobicity. We later confirmed that relatively thick (40 nm) multilayer formation could be obtained by exposing as-is copper with its 4 nm native oxide to ODT (Fig. 10).
In our case, the in situ reduction of copper and its native oxide with TDMAHf lead to the mixed formation of reduced copper metal, conducive to self-limited monolayer growth when exposed to thiols, and copper hafniate CuOHf complexes, conducive to the formation of a multilayer copper thiolate film.
As a result, the partial TDMAHf reduction done in situ leads to a moderately thin (3–4 nm) ODT multilayer film with an ordered uppermost layer. This suggests that controlling the amount of copper oxide at the interface is therefore critical to control the thickness of the deposited thiol film.
F. Hf3N4 ALD on multilayer ODT films
Following the multilayer ODT deposition, Hf3N4 ALD was immediately carried out at 170 °C. The thermal stability of the self-assembled layers under the ALD process conditions is very important because of the potential loss of ordering at elevated temperatures. Therefore, the ALD process was done under conditions compatible with the thermal budget of the underlying thiol film. The sequence including in situ copper oxide reduction, ODT deposition, and Hf3N4 ALD was repeated over four consecutive runs while only varying the number of Hf3N4 ALD cycles from 100 to 545 cycles. Three of those runs were completed using a relatively long 15 s purge times after TDMAHf and NH3 pulses while for the fourth run a shorter 5 s purge was adopted. Additional Si and low-κ chips were also loaded for ex situ characterization. iSE thickness profiles of Hf3N4 ALD obtained on the ODT-coated copper are shown in Fig. 11. As seen from the 545-cycle run, Hf3N4 growth was inhibited for as many as 480 cycles. Interestingly runs with 15 s purge exhibited a consistent slow thickness decay over 4 h (480 cycles) at 170 °C, corresponding to a decay rate of -2.5 Å/h. It is to be noticed that during the iSE acquisition, only Hf3N4 thickness was fitted while ODT optical parameters were all kept constant. Due to the relatively small thickness change involved (<10 Å over 4 h), ODT and Hf3N4 thicknesses over time were not deconvoluted. Therefore, the apparent decrease in Hf3N4 thickness most likely resulted from the thermal decomposition of the ODT sublayer at 170 °C, the thermal degradation of thiol SAMs due to C-S bond cleavage at 177 °C having been reported.27
Hf3N4 final thicknesses for all 4 runs (100, 300, 400, and 545 cycles) on copper, low-κ, and Si are shown in Fig. 12. The results clearly demonstrate the ability of ODT to block the ALD growth on copper, with no apparent retardation on low-κ and Si. At the 480 cycles mark, nearly 70 nm of nitride was deposited on low-κ versus none on ODT-treated copper. This multilayer ODT lead to >6× improvement in terms of ALD cycle retardation over the self-limited ODT monolayers (Fig. 5).
In the context of ALD film growth retardation, the ODT multilayer formation appears therefore beneficial. However, one key benefit of a self-limited monolayer growth is the inherent run-to-run process reproducibility due to the constant SAM thickness. From Fig. 9, t is noticeable that the multilayer deposition leads to some run-to-run variations in the ODT final thickness. We believe these variations are inherent to the experimental setup and were caused by variations in the initial CuO thickness upon ODT exposure as a result of the copper samples being loaded in air at 170 °C. It is reasonable to assume that an experimental setup allowing to load the samples in an inert environment, e.g., glove-box, would therefore prevent the oxidation of the copper and therefore significantly improve the reproducibility of the process.
G. Benefits of SAMs deposition in vapor-phase and future work
The current results offer a positive prognostic in developing an ASD process scheme that could be applicable to not only in R&D settings but also in high-volume manufacturing. The current work illustrates some key benefits over other existing methods based on solution-deposited SAMs that are worth mentioning:
The vapor exposure time required to achieve an effective ODT masking layer was 15 min and could very likely be shortened if necessary. This is significantly faster than any other published results based on sample immersion in a liquid solution which typically require 24–48 h to achieve a closely packed self-assembled layer.
Although data presented in this paper were based on single point in situ thickness measurements, metrology measurements indicate that this vapor-based process is highly uniform at the wafer scale with within-wafer nonuniformity of less than 2% for up to 300 mm diameter wafers. The process therefore appears scalable in terms of wafer size and is expected to be compatible with multiwafer batch processing.
The UDT and ODT precursors has sufficiently high vapor pressures once heated in the 80–150 °C regime to enable delivery via a standard-vapor draw method. It should however be noted than C18-ODT vaporization rate is significantly lower that C11-UDT and therefore alternative delivery methods for ODT are being explored.
The vacuum-based process enables a tight control over key process metrics including reactant dose, exposure times, pressure, and temperature. It was also key in enabling integration of the in situ wafer-state diagnostic via ellipsometry.
Finally, the vapor-based approach permitted to deposit both SAMs and ALD films without air exposures using a single reactor. It should however be noted that even though a single reactor approach is suitable for R&D purposes, potential migration to production would likely benefit from decoupling the SAMs and ALDs into separate clustered reactors. The benefits of that approach would be to limit potential cross-contamination between the two processes, minimize onset of film delamination (from chamber walls) which can be accelerated when alternating organic/inorganic film depositions, as well as decouple the process temperatures so that the SAMs and ALD could be run at different temperatures.
These results, based on in situ ellipsometry measurements, provide a good prognostic for the implementation of a vapor-based methodology in order to achieve an efficient ALD masking layer. Future efforts will focus on characterizing how the selectivity reported at the wafer scale in this paper transfers down onto micro- and nanoscale features as well as establishing detailed compositional analysis via XPS, SIMS, and RBS to characterize the chemical passivation of the thiol layers on copper versus low-k.
Thiol masking layers were selectively deposited on copper versus low-κ to effectively block the ALD growth of dielectric hafnium nitride. Thiols and ALD films were deposited in the vapor phase in a single vacuum chamber at 170 °C. In situ spectroscopic ellipsometry was used to monitor in real-time the growth kinetics of the thiol and ALD processes. The native copper oxide was either chemically reduced in a 1% HCl solution or partially reduced in situ by exposing the substrate to pulses of common ALD precursors such as TDMAHf or TMA, resulting in a mixture of reduced copper island and copper metal oxides. A subsequent single 900 s exposure to ODT vapor over the reduced copper resulted either in a single 2.5 nm ODT monolayer over the HCl-clean copper consistent with a SAM growth, or in a multilayer self-assembled formation over the in situ clean copper. The latter non-self-limited growth was linked to the remaining presence of CuO at the interface as a result of copper oxide partial reduction by thiols to give copper sulfates. The ODT masking layers remained stable at 170 °C to effectively block the hafnium nitride ALD growth for about 75 cycles on the ODT SAM and about 480 cycles on the multilayer ODT film. The thiol deposition proved highly selective on copper versus low-κ as the latter did not exhibit any measurable chemical affinity to sulfhydryl exposure.