Improving the Luminescent Properties of Atomic Layer Deposition Eu:Y2O3 Thin Films through Optimized Thermal Annealing

Crystalline rare‐earth (RE)‐doped Y2O3 films are an attractive system for a wide range of photonics applications including quantum technologies which aim at harnessing optical or spin transitions with long coherence times to achieve new functionalities such as quantum storage or information processing. Herein, atomic layer deposition (ALD) of Eu‐doped Y2O3 thin films with improved optical properties is presented. A crucial post‐treatment step to obtain high‐quality films is annealing at elevated temperatures (>900 °C). However, the main drawback of this approach is the formation of unwanted parasitic phases due to reaction at the interface with the substrate, especially with silicon. In this article, this issue is discussed for different kinds of substrates and buffer layers. The use of such modified substrates allows advantageously extending the maximum thermal treatment up to 1150 °C without being limited by interface reactions. It is demonstrated that the emission of the 5D0 → 7F2 transition for Eu3+ in Y2O3 film can be as narrow as that of bulk materials when optimized thermal treatments and a thin undoped Y2O3 buffer layer are used. Thus, a versatile method to reduce the impact of the substrate–film interface on the optical properties is proposed.

The ORCID identification number(s) for the author(s) of this article can be found under https://doi.org/10.1002/pssa.201900909.

DOI: 10.1002/pssa.201900909
Crystalline rare-earth (RE)-doped Y 2 O 3 films are an attractive system for a wide range of photonics applications including quantum technologies which aim at harnessing optical or spin transitions with long coherence times to achieve new functionalities such as quantum storage or information processing. Herein, atomic layer deposition (ALD) of Eu-doped Y 2 O 3 thin films with improved optical properties is presented. A crucial post-treatment step to obtain high-quality films is annealing at elevated temperatures (>900 C). However, the main drawback of this approach is the formation of unwanted parasitic phases due to reaction at the interface with the substrate, especially with silicon. In this article, this issue is discussed for different kinds of substrates and buffer layers. The use of such modified substrates allows advantageously extending the maximum thermal treatment up to 1150 C without being limited by interface reactions. It is demonstrated that the emission of the 5 D 0 ! 7 F 2 transition for Eu 3þ in Y 2 O 3 film can be as narrow as that of bulk materials when optimized thermal treatments and a thin undoped Y 2 O 3 buffer layer are used. Thus, a versatile method to reduce the impact of the substrate-film interface on the optical properties is proposed. deposition process is surface limited (temperature range of 250-375 C for Y(tmhd) 3 and O 3 precursors). On the contrary, the annealing post-treatment temperature can only be increased until about 950 C on silicon above which parasitic silicate phases are formed and modify the emission properties of the RE ions. [18] In this work, we present strategies to further improve the optical properties of thin Eu:Y 2 O 3 ALD films and go beyond those limits. First, we investigated the deposition on alternative refractory transparent oxide substrates such as sapphire (Al 2 O 3 ) and yttria (Y 2 O 3 ) transparent ceramics in an attempt to further extend the post-treatment temperature. Another approach is based on the use of buffer layers between the silicon and the luminescent layer. The main idea is to space the reactive interface of silicon from the active RE:Y 2 O 3 layer. Two different buffer layers were tested: Si 3 N 4 or in situ grown undoped Y 2 O 3 . This approach aims at developing devices compatible with silicon-integrated photonics. [18]

Results
RE sesquioxides are polymorphic materials with at least five different crystallographic structures. [19] In the case of Y 2 O 3 , two main crystal structures are found: a monoclinic (space group C2/m) and a cubic (Ia-3) one. [19,20] Unlike bulk crystals for which the cubic structure is predominant, some competition occurs for nanoscale materials and a monoclinic contribution may be detected. In a previous study, we indeed observed that both phases are deposited by ALD on silicon. [18] The selection during the nucleation between the cubic or the monoclinic structure is complex to predict because it is connected to deposition parameters (temperature and pressure) and to the interaction between the surface and the yttria nuclei. Indeed, a correlation between the surface energy and the polymorphism has been demonstrated for nanocrystals. [20,21] ALD process is expected to be particularly sensitive to the nature of the surface because it is based on self-limited adsorption. Modification of the orientation or the nature of the substrate could lead to texture or the crystallization of new phases in comparison with silicon. To probe the effect of the substrate on the crystalline environment, we used X-ray diffraction (XRD) and photoluminescence (PL) of europium ion transitions. In particular, the 5 D 0 ! 7 F 2 emission is used as a structural probe to distinguish the cubic from the monoclinic phase. Indeed, a narrow line centered at 611 nm is expected for the cubic one, whereas PL is broader and redshifted for the monoclinic structure (centered around 625 nm). [22][23][24] This large variation is observed because the forced electric dipole (ED) transition 5 D 0 ! 7 F 2 is hypersensitive (ΔJ ¼ 2). Those hypersensitive transitions are excellent environment probes because their intensities are much more influenced by the local symmetry around the RE 3þ ion than the intensities of the other ED transitions. [25,26]

Deposition on Sapphire Substrates
The main advantages of sapphire in comparison with silicon are the chemical compatibilities of oxide films (e.g., insensitive to oxidation), the transparency, and the lower refractive index (1.77 for sapphire against 1.89 for Y 2 O 3 at 600 nm) useful for waveguide technology. [27][28][29] Furthermore, the higher melting point and thermal stability of sapphire in comparison with silicon implies a lower reactivity. A lower thermal mismatch is also expected because the thermal expansion of Y 2 O 3 and Al 2 O 3 is close. [30] Among the different Al 2 O 3 orientations, we selected the a-plane (11)(12)(13)(14)(15)(16)(17)(18)(19)(20) because it shows rectangular symmetry and can be commercially obtained easily. Figure 1a shows the XRD pattern of 87 nm-thick films doped with 5% Eu 3þ and grown on a-plane sapphire for different postannealing temperatures. The position of the two main diffraction peaks (222) and (400) is indicated. From the position of the (222) peaks, we calculated a lattice parameter of 10.52 and 10.53Ǻ for annealing temperatures of 900 and 1000 C, respectively. This is slightly lower than the expected value for a pure Y 2 O 3 ceramic (10.61Ǻ). For as-grown and low annealing temperatures Normalized intensity 28 Figure 1. Effect of the annealing temperature on the a) XRD pattern and b) normalized room-temperature PL emission spectra of the 5 D 0 ! 7 F 2 transitions of 87 nm-thick Y 2 O 3 films doped with 5% Eu 3þ and deposited on a-plane Al 2 O 3 (11)(12)(13)(14)(15)(16)(17)(18)(19)(20) substrates. The XRD data in part (a) are scaled to facilitate comparison. The Y 4 Al 2 O 9 (YAM) peaks present a good agreement with the JCPDS 00-034-0368. The PL spectrum of the sample annealed at 1100 C (in red) has been vertically translated for better clarity. Nanoparticles and a ceramic with the same composition (in gray) are, respectively, used as references for the XRD pattern and the PL spectra.
www.advancedsciencenews.com www.pss-a.com (<700 C), the film is poorly crystallized and preferentially textured along the <100> direction. [18] When the annealing temperature is increased, crystallinity improves and a change in texture occurs because a significant contribution from the (222) planes appears. Other groups have already observed similar changes in orientation during the postannealing treatment. [28,31,32] Until 1000 C, no additional phases are observed on the XRD pattern. At 1100 C, the XRD pattern changes significantly and new broad lines become visible, in particular at 29 and 35 . Those new lines are attributed to the monoclinic yttrium aluminate (YAM) Y 4 Al 2 O 9 (JCPDS card 00-034-0368). [33] This indicates that the deposited yttria layer has reacted with the substrate at this temperature and that the ions are located in a different crystalline environment. This solid-state reaction has already been reported for Y 2 O 3 sol-gel films deposited on sapphire. [34] The PL spectra shown in Figure 1b confirm this analysis. In a similar fashion to films on Si substrates, a narrowing of the PL peaks is observed when the annealing temperature increased. The spectra became closer to that measured for a reference transparent ceramic. Furthermore, emission arising from the monoclinic phase of yttria (shown by an arrow in Figure 1b) reduced with an increase in the annealing temperature. When the temperature reached 1100 C, however, the full width at half maximum (FWHM) increased significantly confirming solid-state reaction with the substrate. A blueshift of the main emission line from 611 to 607 nm suggests the formation of a yttrium-aluminum mixed phase such as Eu:YAM in good agreement with the XRD data. [35] In this phase, Eu 3þ ions substitute Y 3þ ions located in a different C 1 symmetry site. This multiplication of substitutional sites together with the lowering of crystal symmetry is responsible for the broadening of the line observed in PL. [36] Therefore, although the use of sapphire as a substrate for Eu:Y 2 O 3 thin-film growth allows slightly increasing the maximum annealing temperature compared with a standard Si substrate, a limitation also exists at about 1100 C due to the formation of parasitic phases.

Deposition on Y 2 O 3 Transparent Ceramics
Undoped Y 2 O 3 transparent ceramics have been selected because they have the same composition as the layer and no lattice and thermal mismatch. They may allow limiting stress and interface reactions. The main drawback of this substrate is its noncommercial availability and difficult preparation. [37] The substrates were cut into small pieces of typically 5 Â 5 mm 2 and mirrorpolished. Structural analysis by XRD does not allow distinguishing signal from the substrate and the layer. For this reason, only PL results are presented. In Figure 2, the 5 D 0 ! 7 F 2 PL for different annealing temperatures is shown. For 900 C annealing and below, an important contribution from the monoclinic phase (indicated by an arrow) is visible showing again that crystallinity remains relatively poor when low annealing temperatures are used. When the annealing temperature is raised above 900 C, a significant narrowing of the main line at 611 nm is observed. Simultaneously disappearance of the monoclinic contribution occurs. Interestingly in this case, we were able to raise the annealing temperature up to 1200 C with no obvious limitation by parasitic phases unlike other substrates considered previously.
At the highest temperature, the PL spectrum is very narrow and the different transitions between Stark sublevels at 622 and 627 nm are resolved. Therefore, the use of such substrates does not present limits in terms of annealing temperatures due to the absence of interfacial reactions. However, one has to consider that diffusion of ions into the bulk ceramic might occur for higher temperatures and longer annealing times. [38] This can be a limiting factor for the foreseen application in QTs for which spatial localization of the emitters in a thin layer at the surface may be required. [39] Although the transparent ceramic proves to be an ideal candidate for obtaining optimized PL properties of ions in a very high crystalline quality environment, the difficult preparation and poor substrate availability led us to turn to other strategies. In particular, silicon remains an attractive platform with easier scaling-up and postprocessing capabilities. The possibility to limit surface reactions through the intercalation of a buffer layer was thus attempted.

Deposition on Silicon with a 100 nm-Thick Si 3 N 4 Buffer Layer
The first buffer layer studied is a 100 nm-thick low-stress Si 3 N 4 deposited on silicon by plasma enhanced chemical vapor deposition (purchased from University Wafer). Deposition of Eu:Y 2 O 3 luminescent films has already been reported on Si 3 N 4 membrane by e-beam evaporation. Furthermore, Si 3 N 4 is a particularly attractive option for capping passivation or buffer layer due to its high barrier properties, low porosity, and high chemical resistance. [40,41] Reactivity between silicon nitride and Y 2 O 3 has been largely reviewed because yttria is used as a sintering additive in Si 3 N 4 ceramics. [42][43][44] During the high-temperature sintering process, an intergranular crystalline secondary phase of yttrium silicate such as β-Y 2 Si 2 O 7 was shown to appear at the grain interface of β-Si 3 N 4 . [42][43][44]  www.advancedsciencenews.com www.pss-a.com The XRD pattern and PL spectra are shown in Figure 3 for different annealing temperatures. The XRD pattern for the 900 C annealing presents broad and weak peaks indicating a poor crystalline quality. Furthermore, ratio of the (222) peak area on the (400) one presents a value of about 2, whereas the expected ratio is about 3.6 for a random orientation. The use of Si 3 N 4 buffer layer gives a polycrystalline film weakly textured and crystallized. Indeed, the diffraction signal is low and lines are broad, which indicates small crystalline domains. At 1000 C, a significant change in the XRD pattern occurred with the appearance of new lines possibly related to a reaction with the nitride layer and that are overlapping with that originating from the Y 2 O 3 phase. At 1100 C, indexation of the new lines suggested the presence of two yttrium silicate phases (Y 4 Si 3 (SiO 4 )O 10 and Y 2 Si 3 O 3 N 4 ) confirming that a solid-state reaction between the silicon substrate, the nitride buffer layer, and the Y 2 O 3 layer occurred.
The PL spectra confirm the XRD analysis. Indeed, for a 900 C annealing post-treatment, the characteristic PL of europium both in the cubic and in the monoclinic phases are present. When the post-treatment annealing rose up to 1000 C and above, a significant change in the PL spectrum was observed, in a similar way as for a bare silicon wafer, indicating that a reaction occurred between the oxide and the buffer layer or substrate. Therefore, the use of a Si 3 N 4 buffer was not sufficient to prevent the formation of parasitic phases.

Deposition on Silicon with an Undoped Y 2 O 3 Buffer Layer
To move away the reactive interface between the substrate and the luminescent layer, we interspaced a Y 2 O 3 undoped layer. This 20 nm-thick buffer layer was deposited in situ by ALD before the active luminescent layer of Eu:Y 2 O 3 . XRD pattern for different annealing temperatures is shown in Figure 4. As already observed for deposition on silicon, a strong texture along the <100> direction is observed. [18] The <100> texture remains whatever the annealing temperature, in contrast to growth on sapphire. Furthermore, we observed that the 28     www.advancedsciencenews.com www.pss-a.com diffraction peaks are shifted toward higher angles with respect to the reference which suggests the presence of strain as already described by Scarafagio et al. [18] At 1000 C, a careful analysis of the XRD pattern revealed small additional peaks due to the formation of yttrium silicate compounds. Those new silicate phases (such as Y 2 SiO 5 and Y 2 Si 2 O 7 ) are due to chemical reaction between the Y 2 O 3 film and the silicon wafer. For 1150 C annealing temperature, the intensity of the XRD peaks of the silicate phase increased significantly and is indicated by a star on Figure 4a. On the contrary, no additional lines were observed on the PL spectrum (Figure 4b) even for annealing treatment up to 1100 C, whereas on a bare silicon wafer (i.e., without the buffer), emission of RE ions would start to be modified at temperatures as low as 1000 C. In addition, the Stark sublevels were resolved for the highest annealing temperatures. This result indicates that the silicate parasitic phase is well localized at the interface between the silicon wafer and the undoped Y 2 O 3 buffer layer but does not extend to the top Eu-doped layer in which the optical properties of the ions are preserved. In this case, the maximum temperature could be increased to values as high as 1150 C leading to a significant improvement of the crystalline environment. Figure 5 and 6 show the effect of the annealing temperature on the PL intensity and area of the 5 D 0 ! 7 F 2 transition for deposition on different substrates. To better visualize the effect, we normalized the data to that collected for a 900 C annealing. The results are presented up to temperatures at which parasitic phases appear. We can appreciate that the annealing temperature has two effects. For low-temperature annealing (less than 900 C), a significant increase in the integrated emission (PL area of Figure 5b) is observed by a factor of 5-8. We attribute this larger emission to an increase in the number of Eu 3þ ions that participate to the PL spectrum. Indeed in this temperature window, the post-treatment annealing is expected to remove quenching centers that would otherwise limit total luminescence emission. This hypothesis has been confirmed for deposition on silicon for which a substantial change in the decay time is observed (data not presented here). Indeed, for as-grown or low-temperature annealed films, the decay time presents a short lifetime with a nonexponential behavior indicating a strong quenching of the europium emission or energy transfer, whereas at higher annealing temperatures the decay time approaches that measured for bulk crystals. [18] It is remarkable on Figure 5b that the PL area remains rather constant when the annealing temperature is further increased. In contrast, the maximum PL intensity continues to rise (Figure 5a). This suggests that crystalline quality improves and that, although the total number of ions contributing to luminescence remains constant, their emission falls within a smaller inhomogeneously broadened line. This is particularly striking for the highest annealing temperatures that show very high PL intensity.

Discussion
The results are finally shown in Figure 6 which compare the FWHM of the main PL emission at 611 nm for films grown on different substrates. We observe that it strongly decreases with increasing the annealing temperature until the maximum temperature of formation of parasitic phases (silicate or aluminate) is reached. Both depositions on the Y 2 O 3 buffered substrate and the transparent ceramic allow advantageously extending the posttreatment to more than 1100 C without encountering this issue. This translates into much narrower PL linewidths that are comparable or even improved with respect to the reference bulk ceramic material.   www.advancedsciencenews.com www.pss-a.com

Conclusion
In this work, we fabricated Eu 3þ -doped Y 2 O 3 thin films on different substrates by ALD and compared the structural and PL properties with standard deposition on Si (100). For all the substrates, crystalline films were obtained exemplifying the great versatility of the ALD deposition technique. To maximize the PL intensity, a high-temperature annealing post-treatment enhancement is required but the appearance of parasitic interfacial phases is usually a strong limiting factor. For the growth on Al 2 O 3 substrates, the maximum annealing temperature is slightly improved (1050 C) compared with that of bare silicon (950 C). Furthermore, the use of a buffer layer such as Si 3 N 4 did not help in increasing this temperature. When a transparent ceramic is used as a substrate, no interfacial reactions are formed allowing extending post-treatment up to 1200 C. We note though that diffusion of the dopants deep into the matrix might occur and could be an issue for QT applications for which localized dopants are needed. Given the difficulty of preparation and poor availability of the material, transparent Y 2 O 3 ceramics also do not constitute an ideal platform. The use of a thin undoped Y 2 O 3 buffer layer was found to be the most appealing strategy allowing annealing up to 1150 C without modifying Eu 3þ PL on the top Eu-doped active layer, whereas significant changes in the XRD pattern were observed. This suggests that this thin layer acted as a sacrificial interface that reacted with the silicon substrate and prevented the development of the silicate parasitic phase further into the Eu-doped film. The use of this approach of sacrificial undoped layer is a versatile method that could be applied to other photonics systems.

Experimental Section
Film depositions were conducted with a Picosun Sunale R200 using conventional β-diketonate precursors: Y(tmhd) 3 and Eu(tmhd) 3 from (99.9%, Strem Chemicals). These stable precursors were held at 140-160 C and delivered using 300 sccm N 2 carrier gas. The vaporized precursors were sent sequentially with an oxidant gas (O 3 ) into the thermalized deposition chamber at 300-350 C. To maximize the film crystallinity, we used ozone as a strong oxidizing agent. A typical ALD sequence consisted of a 3 s Y(tmhd) 3 pulse followed by a 3 s purging pulse of N 2 ; a 3 s ozone pulse also followed by a 3 s purge with N 2 . Eu doping of the oxide films was easily tuned by sequentially introducing pulses of the dopant element in the standard yttrium oxide cycle. In our case, a doping level of 5% was used by introducing five pulses of Eu(tmhd) 3 every 95 pulses of Y(tmhd) 3 . More details about the impact of the different deposition parameters and their optimization were discussed in details in Scarafagio et al. [18] The following different substrates were considered for deposition: silicon (100), a-plane Al 2 O 3 sapphire (11)(12)(13)(14)(15)(16)(17)(18)(19)(20), Y 2 O 3 transparent ceramic and low thermal stress 100 nm-thick Si 3 N 4 deposited on Si(100). All substrates were used as received after cleaning with ethanol, acetone with ultrasonic bath. They were all purchased from private companies except for the Y 2 O 3 ceramics that were provided by Dr Ikesue (World Lab Co., Nagoya, Japan). To remove experimental bias, deposition on a reference silicon (100) wafer was done for systematic comparison for each experiment. To improve the crystallinity of the films and the optical properties of the emitters, we investigated a postgrowth annealing treatment at different temperatures in air for 2 h. The crystallinity of the films was evaluated using XRD with a Panalytical X'Pert Pro diffractometer using an incident beam Ge monochromator, and a stainless-steel sample holder. For all XRD patterns, baseline from Fullprof software was subtracted.
Film thickness was measured by white light interferometry in the range of 250-1000 nm with an Ocean Optics NanoCalc system using Y 2 O 3 optical constants. Finally, PL measurements were performed in a Renishaw InVia micro-PL apparatus with a 50Â objective and a 532 nm laser as the excitation source.