Atmospheric-Pressure Plasma-Enhanced Spatial Atomic Layer Deposition of Silicon Nitride at Low Temperature

Atmospheric-pressure plasma-enhanced spatial atomic layer deposition (PE-spatial-ALD) of SiN x is demonstrated for the first time. Using bis(diethylamino)silane (BDEAS) and N 2 plasma from a dielectric barrier discharge source, a process was developed at low deposition temperatures (≤ 250 o C). The effect of N 2 plasma exposure time and overall cycle time on layer composition was investigated. In particular, the oxygen content was found to decrease with decreasing both above-mentioned parameters. As measured by depth profile X-ray photoelectron spectroscopy, 4.7 at.% was the lowest oxygen content obtained, whilst 13.7 at.% carbon was still present at a deposition temperature of 200 o C. At the same time, deposition rates up to 1.5 nm/min were obtained, approaching those of plasma enhanced chemical vapor deposition and thus opening new opportunities for high-throughput atomic-level processing of nitride materials.


Introduction
Silicon nitride is an essential material in device fabrication, with applications ranging from microelectronics, thin-film transistors, optoelectronics, encapsulation and patterning.[1]- [3] For many of these applications SiNx should be deposited at relatively low temperatures (< 350 o C) with excellent material properties and uniformity over large areas as well as 3D topographies.
Atomic layer deposition (ALD) and in particular plasma-enhanced ALD (PE-ALD) can fulfil all of the above listed requirements.In conventional, temporal ALD, a substrate is cyclically exposed to a precursor and a co-reactant that are separated in time and undergo self-limiting surface reactions, thus allowing for Ångström-level thickness control and unparalleled uniformity PREPRINT OF PEER REVIEWED AND ACCEPTED MANUSCRIPT 3 ALDJ 2023-27-02; DOI: 10.15212/ALDJ-2023-1000.Available online at www.atomiclayerdeposition.com constraints on the currently employed DBD plasma source.In such spatial ALD setup, the substrate is rotated under continuous gas flows of BDEAS and N2 plasma, which are separated by N2 gas curtains.Each rotation of the substrate underneath the injector head corresponds to one spatial ALD cycle.Because of the fixed geometry of the injector and the reactor design, the exposure times of the precursor and of the co-reactant, as well as their purge times, are coupled.The exposure time of the substrate to the precursors/co-reactant, texp, can be calculated using Equation 1: where W is the width of the deposition zone, r the radial distance from the center of the wafer, and f the rotation frequency in rotations per minute (RPM).The rotation speed was varied between 10 and 80 RPM, corresponding to exposure times ranging from 30 ms to 400 ms.

Decoupling the effect of each half-reaction.
As mentioned above, due to the fixed geometry of the reactor all exposure and purge times are coupled and defined by the rotation speed.The total dose (partial pressure times exposure time, p•texp) can be changed in order to be able to decouple the effect of each half-reaction.Hence, for a fixed rotation speed, by varying the partial pressure of the precursor, one can obtain the saturation curve for the precursor half-reaction while keeping the plasma exposure time constant.From the precursor saturation behavior, the total dose required for precursor saturation can be calculated at any given rotation speed.By ensuring precursor saturation while changing the RPM, the effect of the plasma half-reaction exposure time can also be investigated, provided the 'purge times' are long enough to avoid CVD-like reactions.

Analytical methods.
The growth of SiNx was investigated by depositing on 150-mm doublepolished Si (100) wafers.SiNx film thickness and optical properties were measured by ex-situ spectroscopic ellipsometry (SE) using a Horiba UVISEL II.Film thicknesses were extracted using a Tauc-Lorentz model in the spectral range of 1.5-5 eV.The chemical composition and stoichiometry of the SiNx films were obtained by X-ray photoelectron spectroscopy (XPS), using a Thermo Scientific K-Alpha spectrometer with a monochromatic Al Kα X-ray source.Depth profiles were measured by sputtering with Ar + ions using 500 eV, with steps of 15 s.The relative hydrogen content was estimated by Fourier transform infrared spectroscopy (FTIR) of the SiNx films, using a VERTEX 70 spectrometer from Bruker.Absorbance spectra were collected within the wavenumber range of 650-4000 cm −1 with a resolution of 8 cm −1 using a room temperature deuterated tri-glycine sulfate (DTGS) detector.The H-concentration from FTIR spectra was evaluated based on the Lanford and Rand method, [24] using the absorption cross-sections, σ, σSi-H = 7.4 x 10 -18 cm 2 and σN-H = 5.3 x 10 -18 cm 2 .

Process development
Figure 2 shows the SiNx film thickness as a function of the number of atmospheric pressure-PEspatial-ALD cycles.A rotation speed of 20 RPM was chosen for this set of experiments to ensure long enough precursor and plasma exposures.This corresponds to exposure times between 109 ms and 200 ms for both BDEAS and the N2 plasma and a total ALD cycle time of 3 seconds.A fairly linear thickness increase was observed at all deposition temperatures investigated, from which growth per cycle (GPC) values of 0.31 Å, 0.22 Å and 0.19 Å were obtained for depositions at 150 o C, 200 o C and 250 o C, respectively.
In agreement with earlier literature, the GPC was found to decrease with increasing the deposition temperature, and GPC values were comparable to those reported for low-pressure temporal ALD processes using similar chemistries.[8] These deposition settings, however, resulted in layers with high concentration of O-and C-impurities, tending towards SiOxCyNz type of materials.In particular, impurity levels as high as 14 at.%O and 13.7 at.%C were measured in the bulk of these films by depth profile XPS.Increasing the deposition temperature from 150 o C to 250 o C led to a decrease in carbon content from 13.7 at.% to 7.5 at.%, and to an opposite trend for O, i.e. an increase from 10.3 at.% to 14 at.%.
FTIR was employed to determine the relative hydrogen content in the deposited SiNx layers.Figure 3a shows an example of an infrared spectrum of a SiNx layer deposited at 250 o C at 20 RPM.Si-N (~930 cm -1 ), Si-H (~2190 cm -1 ), N-H (~3353 cm -1 ) and NH2 (~1600 cm -1 ) modes are visible.[6] From these spectra the H-bond densities (Si-H plus N-H) were calculated using the method described by Lanford and Rand.[24] As shown in Figure 3b a linear decrease in the number of H-bond density was found upon increasing the deposition temperature from 150 o C to 250 o C, with the lowest value being ~4.7 x 10 22 cm -3 .As a reference typical H-bond densities reported for PE-CVD in the same temperature range are between 6 x 10 21 (low hydrogen content, <10 at.%) [25]

Effect of plasma exposure on the material properties
With the aim of optimizing the deposition process towards high-quality SiNx, we investigated the two separate half-reactions of the spatial ALD process with particular focus on the N2 plasma exposure since it is has been reported that the plasma exposure can have dramatic influence on the impurity levels of SiNx prepared by PE-ALD.[8], [9] Figures 4a and 4b show the saturation curves for SiNx grown from BDEAS and N2 plasma at 200 o C and 250 o C. The GPC as a function of the BDEAS partial pressure, p, shows saturating behavior for SiNx, indicating self-limiting surface reactions during the precursor step, consistent with ALD behavior and with previous literature on temporal ALD.[8], [10] The rotation speed was kept fixed at 20 RPM to fix the plasma exposure time, while the Ar bubbling flow through the BDEAS precursor was varied from 25 sccm to 150 sccm.For details see the Experimental section.In terms of BDEAS dose, p•texp, saturation behavior was obtained for precursor doses above 0.015 Torr•s, as shown in Figure S1, which is in good agreement with an earlier report on SiO2 spatial ALD using the same precursor.[4] Conversely, the GPC as a function of the plasma exposure time displays deviation from ideal behavior.A similar effect has been previously observed for temporal PE-ALD of SiNx using similar precursor chemistries and described as soft-saturating behavior.[7], [8] For example, in the case of bis(tert-butyl)aminosilane, the N2 plasma exposure was observed to directly affect the Ccontent, through a re-deposition effect, [26] while an Si-H terminated surface upon precursor adsorption was speculated to be responsible for such soft-saturating behavior in the case of di(secbutyl)aminosilane and neopentasilane as precursors.[7]  The refractive index as calculated by optical modelling of the ex-situ SE data shows an increasing trend with decreasing N2 plasma exposure time (see Table S1 for details), pointing towards denser layers and/or lower impurities (H, C, O).In order to shine light on the effect of N2 plasma exposure observed in our case, XPS depth profiles and FTIR measurements were carried out for different plasma exposures.
Figure 5 shows the SiNx composition in the bulk of each film as measured by XPS depth profile for different N2 plasma exposure times and at a deposition temperature of 200 o C (see also Figure S2).While the Si at.% remains fairly constant within the measurement error, the O at % shows an almost linear increase with the N2 plasma exposure time and the C at.% decreases slightly with increasing N2 plasma exposure.An optimum thus exists in trying to minimize the C-and the O-content corresponding to a N2 plasma exposure of ~75 ms, i.e. a rotation speed of 40 RPM.The lowest O-content was instead obtained for a plasma exposure time of ~37 ms, i.e. a rotation speed of 80 RPM.The reasons for such trends are not fully understood at this moment.It can be speculated, however, that a combination of O-impurities present at ppm-levels in the N2 gas and in the background ambient of the reactor are responsible for the incorporation of oxygen, or additionally, that plasma-induced outgassing of the DBD ceramic elements is responsible for O incorporation.Hence, a shorter plasma exposure and an overall shorter cycle time are expected to reduce the oxygen content.On the other hand, longer plasma exposure times might be needed to completely remove the carbonaceous ligands.With this in mind, also a mixture of N2/H2 was tested since H-radicals are known to be effective in removing carbon.However, GPC values as low as 0.06 Å were obtained, similar to earlier reports.[8] Ande et al. have used density functional theory to show that such low GPC can be attributed to the formation of an H-terminated surface that is short of suitable adsorption sites for the subsequent precursor step.[27] By optimizing the plasma exposure time to 37 ms, together with an overall ALD cycle time of 750 ms, SiNx layers with only 4.7 at.%O could be obtained, yet with 13.9 at.% C. Figure 6a shows that the absolute oxygen content drastically decreases within the bulk of the SiNx layers when going from 10 RPM (red lines), 40 RPM (black lines) to 80 RPM (blue lines).Conversely, the absolute nitrogen and carbon contents increase.Furthermore, a decrease in hydrogen bond density down to 3.5 x 10 22 cm -3 with increasing rotation speed was derived from FTIR measurements, as shown in Figure 6b.Concurrently, the GPC decreased from 0.27 Å at 10 RPM to 0.19 Å at 80 RPM, which corresponds to an increase in deposition rate from 0.27 nm/min to 1.5 nm/min.See Table S1 for more details.

Comparison to PE-ALD and PE-CVD
Table I provides an overview of chemistries, deposition temperatures and rates, and resulting material properties such as film refractive index and composition for PE-CVD, atmospheric pressure PE-CVD, low-pressure temporal ALD, and atmospheric-pressure spatial ALD, at deposition temperatures of 200 o C or 250 o C.
By comparing the results obtained in this work with those reported in literature a few conclusions can be drawn.With atmospheric-pressure spatial ALD one can grow films with oxygen contents comparable to low-pressure temporal ALD.The carbon content is even lower than that reported for films grown from BDEAS and high-frequency Capacitively Coupled N2 plasma (CCP) in low-pressure temporal PE-ALD, [10] and comparable to the films grown from bis(tert-butyl)aminosilane (BTBAS) in combination with N2 plasma, yet much higher than those obtained by di(sec-butyl)aminosilane (DSBAS) and TSA in combination with N2 plasma.It is likely that the oxygen contaminations in the case of spatial ALD are originated by O2 and H2O traces that are present in the background ambient of the reactor, similarly to what reported for the atmospheric pressure PE-CVD process using SiH4/N2/He [12].
Since carbon incorporation within the bulk of the SiNx films originates either from incomplete removal of the precursor's ligands or from redeposition effects, one can expect that atmospheric-pressure spatial ALD using precursors with better exchangeable ligand groups such as BTBAS, or carbon-free precursors such as trisilylamine or neopentasilane will result in much lower carbon contaminations thus further improving the SiNx quality.Alternatively, a dual-plasma approach with an H2/N2 plasma exposure followed by an N2 plasma can be developed in order to drive out even more carbon.
Regarding the hydrogen content, a direct comparison could not be made because of the different measurement methods present in the literature: infrared spectroscopy and elastic recoil detection.
When comparing the different vapor phase deposition methods listed in Table I, a clear difference can be observed between the deposition rates at atmospheric-pressure and those at lowpressure.For atmospheric pressure PE-CVD much higher deposition rates have been reported compared to low-pressure PE-CVD, likewise atmospheric-pressure spatial ALD can afford between ~8 to 42 times higher deposition rates than temporal ALD, which almost approach the ones obtained by PE-CVD.Finally, it should be noticed that by optimizing the spatial ALD reactor design even higher deposition rates may become accessible.

Conclusions
A world-first low-temperature (≤ 250 o C) atmospheric-pressure spatial ALD process for SiNx has been developed, which can reach comparably low oxygen atomic percentage as those obtained by low-pressure temporal ALD processes based on similar chemistries.At the same time, deposition rates up to 42 times higher than temporal ALD were measured, i.e. 1.5 nm/min, thus almost approaching the deposition rates of low-pressure PE-CVD processes.
With the current reactor configuration we show that the N2 plasma step has a strong impact on the resulting layer composition and that short plasma exposure and cycle times lead to low oxygen contents.We speculate that this may be due to O2 and H2O traces being present as impurities in the background ambient and in the N2 gas lines, or due to plasma-induced outgassing of the ceramic elements of the DBD plasma source.Whilst carbon contamination still remains a concern for the current process, we foresee that different precursor chemistries will lead to improved film composition and that especially N2 purifiers will allow for obtaining SiNx compositions that are less dependent on plasma exposure.Such independency will probably be a prerequisite prior to testing the atmospheric-pressure spatial ALD process on high-aspect ratio structures.
We believe that the results presented in this work will enrich the toolbox of spatial ALD processes and open pathways for high-throughput spatial ALD of silicon nitrides, and other relevant nitrides like titanium nitride, at atmospheric pressure, thus making it compatible with sheet-to-sheet, roll-to-roll as well as semi batch-type of reactors for high-throughput atomic-level processing.

Figure 1 .
Figure 1.Schematic views of (a) the 150-mm wafer lab-scale reactor and (b) the bottom-side of the spatial ALD injector head.The N2 bearing gas streams are controlled in such a way that the substrate is coated at close proximity.

Figure 2 .
Figure 2. SiNx layer thickness as a function of the number of spatial ALD cycles, for deposition at 150 o C, 200 o C and 250 o C.

Figure 3 .
Figure 3. a) Typical infrared spectrum for of a SiNx layer deposited at 250 o C at 20 RPM, with Si-H and Si-N stretching mode, and N-H stretching and NH2 scissoring mode indicated; b) total Hbond density (N-H plus Si-H) derived from FITR measurements.

Figure 4 .
Figure 4. a) shows the growth per cycle (GPC) at (■) 200 o C and (■) 250 o C as a function of BDEAS exposure for a fixed N2 plasma exposure time.b) GPC at 200 o C (■) and 250 o C (■) as a function of the plasma exposure time, for BDEAS exposures above saturation.While the BDEAS exposure results in saturating behavior, the GPC as a function of the plasma exposure time shows a non-ideal behavior.

Figure 5 .
Figure 5. Contents in atomic percentage of O (▲), C (■), Si (•), N (•) in the bulk of the spatial ALD grown SiNx layers as a function of the N2 plasma exposure time.

Figure 6 .
Figure 6.a) Raw XPS data for O1s, N1s and C1s within the bulk part of SiNx; b) baselinesubtracted Si-H and N-H modes as measured by FTIR.All SiNx layers were deposited at 200 o C by atmospheric-pressure spatial ALD at 10 RPM (-), 40 RPM (-), and 80 RPM (-).

I
N T E R NA T I O N A L J O U R N A L PREPRINT OF PEER REVIEWED AND ACCEPTED MANUSCRIPT 10 ALDJ 2023-27-02; DOI: 10.15212/ALDJ-2023-1000.Available online at www.atomiclayerdeposition.com

I
N T E R NA T I O N A L J O U R N A L PREPRINT OF PEER REVIEWED AND ACCEPTED MANUSCRIPT 15 ALDJ 2023-27-2; DOI: tbd.Available online at www.atomiclayerdeposition.com Supporting Material Atmospheric-Pressure Plasma-Enhanced Spatial Atomic Layer Deposition of Silicon Nitride at Low Temperature

Figure S1 .
Figure S1.Growth per cycle for atmospheric pressure spatial ALD of SiNx at a rotation speed of 20 RPM and a deposition temperature of 250 o C.

Figure S2 .
Figure S2.XPS depth-profiles for SiNx layers deposited by atmospheric pressure spatial ALD at 200 o C and rotation speed of: a) 80 RPM; b) 40 RPM; c) 10 RPM

Table S1 .
Material properties as a function of the rotation speed, at a deposition temperature of 200 o C.