Quasiatomic layer etching of silicon nitride enhanced by low temperature

Atomic layer etching (ALE) is a cyclic, self-limiting etching process, involving alternating steps of surface modification and removal, which can allow for the atomic level precision of etch depth.¹ This process allows for advanced lithography techniques, such as self-aligned multiple patterning, and the reduction in feature sizes in nanofabrication, including transistor size.^2–4 ALE is often categorized as either thermal ALE involving cyclic temperature-driven chemical reaction steps or plasma ALE involving alternating plasma chemistries in a reactive ion chamber. The primary focus of this paper is on a plasma ALE process. Due to the self-limiting nature of the process, ALE will often have reduced surface roughness of the etched material.^5,6 Such precision is necessary for future technology nodes and reduced loss in optical waveguides.

Silicon nitride (SiN) is a common dielectric material with a wide array of uses in nanofabrication and nanotechnology. It can be used as a dielectric spacer layer, a sacrificial mandrel, or a sidewall layer in the self-aligned multiple patterning process⁷ or as a high index core material in waveguides and integrated photonic devices.⁸ Silicon nitride with various ratios of silicon, nitrogen, and hydrogen can have advantages over one another depending on the application.⁹ Low stress silicon nitride is advantageous over stoichiometric silicon nitride for the growth of thicker films, as the reduced stress prevents cracking,^10,11 and can provide nonlinear effects for advanced photonic applications.¹² The processing method used to fabricate these waveguides can have significant effects on their propagation loss. For tightly confined waveguide modes in high index contrast platforms, scattering off of surface roughness created in the etch process can be a dominant source of loss.^13–15 In this paper, we develop a cryogenic atomic layer etching recipe that reduces spontaneous fluorine etching of silicon in Si-rich SiN, improving sidewall homogeneity.

Atomic layer etching of silicon nitride is well studied, using both fluorocarbon plasma chemistry,^16–18 hydrogen plasma surface modification and hydrofluoric acid wet etching,¹⁹ thermal atomic layer etching,²⁰ and hydrogen plasma surface modification followed by fluorine plasma surface removal, which is the focus of this paper.^21–24 In hydrogen/fluorine plasma ALE, the silicon nitride is first exposed to a hydrogen plasma, where the surface is modified by hydrogen implantation in the nitride, creating a chemically modified-damaged layer of thickness ranging from 3 to 9 nm.²¹ This chemically modified layer is then removed by the fluorinated plasma, which ideally stops upon reaching the unmodified, pristine silicon nitride. The high hydrogen-implantation depth leads to a significantly larger etch per cycle than conventional atomic layer etching recipes, opening practical applications to quickly etch thicker devices, such as optical waveguides. This established SiN ALE recipe works well in conventional reactive ion etch (RIE) plasma etchers; however, the fluorinated plasma step runs at 500 mT pressure,²¹ which is unachievable by some modern inductively coupled plasma-reactive ion etch (ICP RIE) plasma systems. In this paper, we develop a recipe using a 20 mT SF₆ etch step, which can be achieved by most modern ICP etchers. We find that in our modified etch recipe, the chemically modified layer can reach up to 20 nm in thickness, leading to an etch per cycle rate that is doubled compared to previous reports. We observed that at this lower pressure and at 10 °C wafer temperature, the SF₆ modified surface removal step loses its self-limiting aspect and etches material beyond the surface modified layer. However, we find that the self-limiting aspect of the SF₆ etch step can be recovered by reducing the wafer temperature during the etch, with the added benefit of improved sidewall homogeneity and decreased surface roughness.

In this experiment, we study the etching characteristics of two different films of silicon-rich silicon nitride films, referred to as film 1 and film 2, grown to thicknesses between 1 and 3 μm by low pressure chemical vapor deposition on silicon substrates at NASA Jet Propulsion Laboratory. The two recipes use different proportions of dichlorosilane and ammonia gas to produce different atomic ratios of silicon, nitrogen, and hydrogen in the films, which, in turn, have different levels of stress and slightly different etch characteristics. Figures 1(a) and 1(b) show x-ray photoelectron spectroscopy (XPS) data, showing the atomic ratio to be roughly Si_0.55N_0.45, with <3 at. % oxygen, and show that film 1 has a slightly higher nitrogen content, while film 2 has a slightly higher Si content. Hydrogen forward scattering (HFS) measurements show <1 at. % of hydrogen in the films, and secondary ion mass spectroscopy (SIMS) measurements in Fig. 1(c) show that film 1 has three times as much hydrogen as film 2.

The silicon nitride films are etched in an Oxford PlasmaPro 100 Cobra ICP RIE, which has both capacitively and inductively coupled plasmas, as well as a liquid nitrogen cooled stage to characterize the etch recipe at various sample temperatures. The wafer is mechanically clamped, and heat transfer is achieved by backside cooling. The temperature is measured during etch on the substrate pedestal. The ALE process and details of the recipe are shown in Fig. 2. Throughout the etch process, the plasma is maintained by 500 W ICP power and a high flow of Ar gas at 20 mT. During the hydrogen bombardment step, there is an additional flow of 50 sccm of H₂, and the capacitively coupled vertical RF field is applied for 5 s at 75 W, (except for Fig. 6 where these parameters are varied). The RF power supply measures the DC bias, or the electrical potential energy between the bulk plasma and SiN wafer, in real time by taking the voltage potential between the substrate chuck and ground. During the SF₆ plasma step, there is no applied vertical bias field, only the addition of 25 sccm SF₆ gas for 30 s unless otherwise noted. Each step is followed by a 30 s purge step with only Ar gas, with plasma maintained by the ICP.

The silicon nitride wafers are coated by an AZ 5214 photoresist, spun to 1.6 μm thickness, and patterned by photolithography to create plateaus with a width of 1 mm, such that sidewalls are placed relatively far from one another. The wafers are then cleaved into chips, fixed on a silicon carrier wafer using Fomblin oil, and etched. The chips are etched by at least 30 cycles of the ALE process to obtain an average etch depth per cycle over many cycles. After etching, the resist is stripped by sonication in acetone. Before each process, the plasma chamber is cleaned by an Ar plasma clean recipe on a blank silicon wafer with 1500 W ICP and 100 W HF bias for 60 min to ensure that the chamber condition is consistent between runs.

The etch depth of the silicon nitride films is characterized by a DektakXT profilometer with 0.4 nm resolution, comparing the difference in the height between the etched and protected areas of the film. For samples with less than 50 nm etch depth, etch depth is measured by noncontact mode atomic force microscopy (AFM), with surface height measurement noise less than 50 pm. The surface roughness of the etched material is measured by noncontact mode AFM, and the sidewalls of the etch are characterized by AFM and scanning electron microscopy (SEM). The Si, C, and O compositions of the films are determined by XPS, and the relative H content between the films is determined by SIMS, using OH⁻ as a proxy for the hydrogen content, and normalizing by the O⁻ content of the films. HFS was performed by Eurofins EAG Laboratories as a calibration for the overall at. % of H.

where $E H 2$ is the etch depth of the H₂ step alone, $E S F 6$ is the etch depth of the pristine silicon nitride without H₂ plasma modification, and $E c y c l e$ is the etch depth of the full ALE cycle.

Figure 3 shows the resulting etch depths from the individual steps compared to the full cycle. There is a small amount of etching done in the H₂ step, which is likely due to the physical sputtering of the material in the presence of the plasma with the vertical bias field applied. There is a significantly larger etch depth during the SF₆ step alone, which could be attributed to the spontaneous SF₆ etching of the silicon-rich film. Due to the absence of the vertical bias field in this step, this etch is purely chemical and isotropic, which may be nonideal for certain applications. However, we observe that when the wafer temperature is lowered from 10 to −50 °C, the spontaneous etching by the fluorinated plasma is significantly reduced. At 10 °C, the synergy value, S%, is 52% and 53% for the two films of silicon nitride and improves to 66% and 70% at −50 C, respectively.

In order to test the etch saturation of the SF₆ step, we varied the length of the SF₆ plasma exposure as at different temperatures. We performed this experiment both including the H₂ surface modification step and without etching the pristine silicon nitride. Figure 4 shows the etch rate as a function of SF₆ step time at various temperatures. At all temperatures, we find that the H₂ modified silicon nitride etches significantly faster than the unmodified surface, in agreement with the data from the synergy test in Fig. 3. However, at 10 °C, we observe that the SF₆ step continues to etch SiN as this step time is increased, and we do not observe saturation behavior of the etch. This indicates that in the full atomic layer etch recipe, the silicon nitride continues to be etched by SF₆, even past surface modified level. As we reduce the temperature of the wafer, we find that the etch per cycle of the modified SiN reduces slightly, and the etch per cycle of the pristine SiN reduces significantly. At 10 °C, the SiN etch per cycle of SF₆ only step is 2.8 nm/cycle, which reduces to 1.4 nm/cycle at −10 and −30 °C, and to 1.1 nm/cycle at −50 °C, indicating a reduction of the spontaneous etching of pristine SiN by SF₆. Additionally, at −50 °C, we observe that the etch per cycle of the full ALE recipe saturates, indicating that the reduced temperature reduces the spontaneous etching of the unmodified SiN by the SF₆ plasma. This self-limiting aspect of the etch process greatly increases the ability to precisely control the etch depth.

In order to test the hydrogen-implantation depth of the surface modification layer, we performed the etch with varying RF powers during the H₂ plasma step, shown in Fig. 5(a). As the vertical electric field is increased, hydrogen ions can penetrate further into the silicon nitride, allowing for the increased thickness of the surface modified layer. We observe a roughly linear relationship between the DC bias in the surface modification step and the overall etch rate of an ALE cycle, indicating that the bias power in the H₂ step can be a valuable control in applications needing a small, precise etch depth. Figure 5(b) shows the etch depth as a function of time of the H₂ plasma step, where we observe an increase in the etch depth per cycle with increasing H₂ dose time. This increase in the etch depth comes from two sources. The first is the increased amount of the material that is physically sputtered off the film in this H₂ plasma step, which is determined by measuring the etch rate of the H₂ dose step without the SF₆ step. The remaining increase in the etch depth comes from the increased depth of the surface modified layer, shown by the “Diff.” dashed line in Fig. 5(b). We find that the thickness of the surface modified layer saturates at 20 nm, twice as thick as previous reports, which both use lower vertical RF power and do not have an inductively coupled plasma.^21,22 These results show that the H₂ step conditions can be tuned to control etch depth per cycle, which are important for precise control of total etch depth.

Sidewall profile is often important for waveguide design. We find that the sidewall profile is significantly improved by reduced etching temperature in terms of sidewall angle homogeneity or the relative etch angle along the height of the sidewall. Figures 6(a) and 6(b) show the AFM topography measurements of the sidewall of the etch, etched for 60 cycles at 10 and −50 °C, for different SF₆ step times. For the 10 °C samples, we observe a nonhomogenous etch profile, where the top of the sidewall is straight, and the base of the sidewall is rounded. The rounded etch profile is indicative of the presence of isotropic etching, which we attribute to spontaneous etching done by the SF₆ plasma, as there is no vertical bias field during this step. Specifically, we attribute this rounded sidewall profile to the isotropic etching of pristine, unmodified SiN, as the modified surface layer is created anisotropically due to the vertical bias field applied during the H₂ modification step. Meanwhile, the sample etched at −50 °C features a homogenous etch profile, where the etch angle is consistent from the top to bottom of the sidewall. Additionally, we observe very little difference in the etch depth between the sample with 15 and 30 s SF₆ step length, indicating a higher degree of etch saturation, confirming the results from Fig. 4. The sidewall angle of the −50 °C etch is 38° from the surface flat, which may be nonideal for applications where the benefits of a vertical etch outweigh the importance of sidewall homogeneity. Figures 6(c) and 6(d) show the SEM images of the silicon nitride etched at 10 and −50 °C, for 30 s SF₆ time. For the sample etched at 10 °C, the image of the sidewall shows a rounded edge at the bottom and ridges along the length of the etch, which may lead to a source of scattering for waveguide applications. The SEM image of the sample etched at −50 °C shows a sharper bottom edge, without rounding or ridges. Such control of the sidewall profile is highly valuable for optical waveguides fabricated in SiN. The correlation between the reduction of etching done by the SF₆ step alone at reduced wafer temperature and the disappearance of the rounded sidewall base provides further evidence that the undesirable rounded sidewall profile occurs due to excess spontaneous SF₆ etching.

In addition to the improvement in sidewall homogeneity, we also observe a decrease in the surface roughness of the etched silicon nitride at −50 °C. Figure 7(a) shows the AFM scans of the surface of the silicon nitride protected during the etch by the photoresist, giving the initial surface roughness. Figures 7(b) and 7(c) show the AFM of the surface after 60 cycles of ALE at 10 and −50 °C. In this figure, we observe that etching at reduced temperature decreases the RMS surface roughness of the material from 0.42 to 0.36 nm, while etching at 10 °C increases the surface roughness to 0.81 nm. This reduction in the surface roughness indicates a reduction in spontaneous etch processes, which can have varied etch rates at the microscopic level due to local differences in the plasma or surface composition. Decreased surface roughness is useful for waveguides with tight optical confinement, where scattering off surface defects can lead to a significant source of loss.^14,15

Optical waveguides in silicon nitride are often greater than 40 nm thickness, and, thus, in order to obtain a useful etch depth for this application, many cycles of the ALE process must be performed. However, other applications may require much lower etch depth, and, thus, information on the etch depth of the first few cycles is valuable. Figure 8 shows the SiN etch depth end etch per cycle for the first few cycles of the ALE recipe. We find that the initial condition of the plasma chamber, determined by the previous recipe(s) run in the chamber, significantly affects the etch depth of the first ALE cycle. When the chamber is in the condition after the Ar clean run, where silicon is sputtered onto the chamber walls in the cleaning process, the etch depth of the first etch cycle is significantly lower than the steady state etch per cycle. This is likely due to the interaction between the chamber sidewall condition and the plasma chemistry.^25,26 Alternatively, when the plasma chamber is in the condition where the ALE recipe has been previously run for many cycles prior to the etch test, the etch per cycle of the first etch cycle is larger than the steady state etch per cycle. This phenomenon is consistent with previous reports on the ALE of silicon nitride.²¹ This etch per cycle dependence on the chamber condition was repeated for two test samples, removing the possibility of an erroneous etching run. After 6 cycles, the etch per cycle for the two chamber conditions approach one another at a value slightly higher than the etch per cycle averaged over 60 cycles with a clean chamber condition.

In this work, we have found a quasiatomic layer etch recipe for silicon nitride that is compatible with low pressure ICP RIEs and explored the effect of wafer temperature on the etch characteristics. We explore the etch saturation of both the H₂ and SF₆ steps of this recipe and find that both steps show etch saturation, in agreement with previous studies on the atomic layer etching of silicon nitride.^21,22 Furthermore, we find that reducing the temperature of the wafer during etch decreases the surface roughness of the etched material and significantly enhances the self-limiting aspect of the SF₆ modified surface removal step, which, in turn, enhances the synergy of the etch. Finally, we show that the sidewall profile becomes more homogenous at reduced temperature, due to the reduction of isotropic etching done by the SF₆ plasma.

Atomic layer etching has numerous advantages over reactive ion etching, including precise etch depth control,^2,27 reduced surface roughness,^6,17 and reduced surface damage due to the often-lower bias power used. Silicon nitride films are very commonly used as a dielectric of waveguide materials, with silicon-rich SiN being increasingly used for optical waveguides due to its increased index of refraction and reduced stress compared to stoichiometric SiN.⁸ For tightly confined waveguide modes in silicon nitride, sidewall roughness plays an important role in loss by scattering.^13,14 The enhanced sidewall homogeneity using cryogenic atomic layer etch will likely enable lower loss in these waveguides.

This research was primarily carried out at the Jet Propulsion Laboratory (JPL), California Institute of Technology, under contract with the National Aeronautics and Space Administration (NASA). This research was, in part, carried out at the Molecular Materials Research Center in the Beckman Institute and Microanalysis Center of the California Institute of Technology and, in part, by Eurofins EAG Laboratories.

The authors have no conflicts to disclose.

Daniel N. Shanks: Investigation (equal); Methodology (equal); Validation (equal); Visualization (equal); Writing – original draft (equal). Rania K. Ahmed: Investigation (equal). John D. Femi-Oyetoro: Investigation (supporting); Writing – review & editing (equal). Matthew R. Dickie: Investigation (supporting). Andrew D. Beyer: Methodology (supporting); Writing – review & editing (equal). Frank Greer: Conceptualization (equal); Methodology (equal); Supervision (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.