1. Introduction

Lithium niobate (LN) on insulator (LNOI) is a widely used material for many applications in nonlinear optics, electro-optics, and acousto-optics and a very promising material for the realization of photonic integrated circuits (PICs) [1–3]. The utilization of rare-earth lanthanide ions in the LN crystal structure makes integrating active elements in the PICs possible [4–8]. For instance, the rare-earth (RE) doped LNOI (RE:LNOI) have been an attractive platforms for applications in quantum information processing and communications [9]. Realization of broadband quantum memories with a relatively long coherence times and nondestructive detection of photonic qubits demonstrated based on RE:LN waveguides could be named as other applications of RE:LNOI substrate [10,11].

Among different lanthanide ions, Er³⁺ attracts special attention due to its ⁴I_13/2 $\rightarrow$ ⁴I_15/2 transition and corresponding fluorescence emission at the telecommunication C-band. Thanks to the new properties of LN after adding the Er³⁺ ions, the doped substrate can find a wide range of applications in photonics, sensing, and telecommunications when structured into waveguides, resonators, and other configurations [12–15].

One method to introduce Er³⁺ ions to the crystal of LNOI is using the modified Czochralski method during the production of the LN crystal [16,17] followed by a process using the smart-cut method [18]. This method is well-established and has notable advantages such as high-quality surface and uniform doping profile. However, a limitation is the complete doping of the entire LN substrate with Er. This affects the properties of all waveguide elements integrated on the same substrate, although only a limited region actually makes use of the doping, which can be avoided using local doping [19].

Alternatively, Er³⁺ ions can be added to the LNOI later via different processes. Different methods have been developed and tested to incorporate Er³⁺ into the LNOI [20]. Among these methods, ion implantation, ion in-diffusion, and ion exchange are widely investigated and used.

In the ion implantation method, a flux of the Er ions with high energy is accelerated to the sample’s surface and the ions penetrate the crystal structure. A post-annealing at high temperature is necessary to repair the damaged lattice from the implantation process [21,22]. Although this method has been widely used for the fabrication of different structures [22,23], it requires sophisticated equipment and processes.

An alternative method is ion in-diffusion from a metal layer, for which a layer of Er with a thickness of 10 to 20 nm is deposited on the surface of LNOI, and the ions diffuse into the LNOI crystal using a specific annealing process. The annealing should be done for about 100 hours and around 1000 °C [24–26]. Compared to ion implantation, this method is simpler, however, the annealing temperature can cause adhesion problems at the bonding layer of LNOI to the SiO₂ buffer layer [27].

The method we use in this paper for doping LNOI is a low-temperature ion exchange process using a salt melt of Er [28–31]. In this method, a solution of Er³⁺ ions is prepared and the sample is immersed in the solution for a few hours at a temperature of 400-500 °C. During the doping, a thin layer at the LNOI surface is doped with Er³⁺, where the doping concentration as a function of the depth depends on a subsequent thermal annealing process. Compared to previously mentioned methods, this method does not require high-temperatures and, hence, does not apply a high thermal load on the sample [19]. Besides, it is a simpler method than the ion implantation method. The ion-exchange method is widely used for doping bulk LN substrates [28–31], however, up to our knowledge it has not been applied to LNOI substrates.

In this paper, we developed the doping of Er³⁺ ions in LNOI by optimizing experimental parameters for Er³⁺ doping and finding the best conditions for the fabrication of high-quality substrates. The Er³⁺-doped samples are optically characterized in the near-infrared wavelength range. Also, we demonstrate the possibility of local doping in SiO₂ masked samples. Our fabricated substrates can be potentially used for many applications, particularly in PICs. These substrates offer a promising platform for the integration of active elements such as amplifiers and lasers [23,32,33]. The capability of local doping enables doping of the substrate in the desired areas without affecting the properties of the adjacent regions, which gives another flexibility for fabrication of customized PICs.

2. Fabrication

For the samples, we used LNOI substrates with a thickness of 600 nm on a 2-µm-thick SiO₂ buffer layer and a 500-µm-thick layer of LN for handling. The fabrication process is shown in Fig. 1. In the first step, we prepare a solution of KNO₃ and Er(NO₃)₃. The initial proportion we used is 27 g (90 wt%) of KNO₃ and 3 g (10 wt%) of Er(NO₃)₃. In parallel, the sample is cleaned with acetone, isopropanol, and deionized water. Then the sample and the solution are tempered separately to 465 °C with a heating rate of 30 °C/min. This temperature exceeeds the melting temperature of Er(NO₃) and KNO₃[34], causing the salts to completely melt (Fig. 1(a)). The doping step is done by immersing the sample in the solution for 3 to 5 hours (Fig. 1(b)). In this step, the Li⁺ ions are replaced with the Er³⁺ and K⁺ ions in an amount and distribution depending on the reaction temperature and time.

 

Fig. 1. Fabrication of Er-doped LNOI. (a) Preparation of the sample and the solution by heating them separately to the temperature 465 °C. (b) Doping of the sample by immersing it in the solution for few hours. (c) Annealing the substrate on a hot plate. A sample temperature profile of the hot plate during the annealing process is shown in the figure, where the annealing takes place in sequence at 200, 300, and 400 °C for 4, 4, and 23 hours respectively. The temperature ramps up and down between different stages as 5 °C/min

Download Full Size | PDF

In the next step, the sample undergoes a combined process involving annealing and the diffusion of doped ions deeper into the substrate, following a pre-defined temperature profile. A typical temperature profile is depicted at Fig. 1(c) where the process is done at three different stages with temperatures of 200, 300, and 400 °C and duration of 4, 4, and 23 hours respectively. For simplicity we use the following notation: 200/300/400 °C and 4/4/23 hours. The temperature changes between different stages for all of our experiments at 5 °C/min. In our experiments, we optimized the temperature profile by changing the number of stages as well as the temperature and duration time of each stage, since it is important to facilitate the diffusion of the Er³⁺ ions deeper into LN [30] and to anneal the damaged crystal structure [22].

3. Discussion and results

3.1 Doping of bulk and thin film LN

In the first experiment, we compared the results of the doping process for LNOI and bulk LN. We doped X-cut LNOI and X-cut bulk LN with a thickness of 500 µm with the same process. The weight ratio of salt was 27g:3g for KNO₃:Er(NO)₃ and the samples were immersed in the solution for 5 hours. Both samples were annealed for one annealing stage at 400 °C for 25 hours. The measurement results from secondary-ion mass spectrometry (SIMS) are shown in Fig. 2. In principle, there are 4 possibilities for the position of Er³⁺ in the crystal of LN: either it substitutes the Li or Nb ions or it stays at the octahedral or tetrahedral positions of the crystal [35]. For both of our samples, as it is shown in SIMS measurements, the Li⁺ concentration correlates with the concentration of Er³⁺ and K⁺ ions, while the O and Nb concentrations are unchanged in the whole measured thickness. It means that the ion exchange process happens mainly between Er³⁺ and Li⁺ ions. This is in agreement with the previously reported results with a similar doping method [28].

 

Fig. 2. Secondary-ion mass spectrometry measurements of doped bulk LN and LNOI. Both samples are X-cut and the doping and measurement conditions are kept the same.

Download Full Size | PDF

One significant point in this figure is that the concentration of Er³⁺ ions doped into LNOI is much higher than the bulk LN. The explanation of this observation is not simple, however, we think that the higher surface-to-volume ratio of the LNOI can be the reason.

3.2 Optimization of doping process

The fabrication of high-quality Er³⁺ doped substrates with a desired ion concentration profile depends on the fabrication process parameters. In this part we explain our systematic study for finding a proper doping recipe to fabricate high-quality Er-doped LNOI platforms. In the first part, we investigated the effect of the doping and annealing steps by fabricating samples at different temperatures and with different durations. We compared the concentration profiles of doped ions in LNOI using SIMS measurements. Afterwards, we changed the ratio of salts used in the doping process and investigated its effect on the ion concentration profiles and the substrates’ final quality. The fabrication parameters of our samples are shown in Table 1 and the details are explained in the following.

Table 1. Sample fabrication process details. For samples 5,6, and 8, the cracks are detected only at the edges of the samples. Sample 9 is our best sample and we did not detect any cracks on the sample’s surface. The result is confirmed by processing 5 identical samples.

View Table

The first parameter we changed is the doping immersion time. We produced two samples (samples 1 and 2) with an immersion time of 5 and 3 hours respectively. The SIMS measurements of these two samples are shown in Fig. 3(a). As expected, the immersion time clearly affects the amount of Er³⁺ ions which are inserted into LNOI. For the sample with an immersion time of 3 hours, the SIMS measurement shows that the Er³⁺ ions are doped into LNOI for 100 nm. However, in the sample with higher immersion time, the amount of Er³⁺ ions is much higher and deeper into the LNOI.

 

Fig. 3. (a) Effect of changing doping immersion time for sample 1 (5 hours) and sample 2 (3 hours). (b) Effect of annealing profile on ion concentration in LNOI for sample 2 (300 °C and 25 hours), sample 3 (350 °C and 25 hours), and sample 4 (400/200 °C and 13/8 hours).

Download Full Size | PDF

To check the effect of the annealing temperature on the doping profile, we fabricated three samples (samples 2 to 4) with the same immersion conditions and different annealing processes. The SIMS measurements for these three samples are shown in Fig. 3(b). For samples 2 and 3, the annealing is done with a one-stage profile at 300 and 350 °C respectively for 25 hours, while sample 4 is annealed with a two-stage annealing profile of 400/200 °C and 13/8 hours. As mentioned before, the temperature change for all experiments is 5 °C/min. We observe that the Er³⁺ ion concentration profiles are almost the same for samples 2 and 3, while the higher temperature results in higher K⁺ concentration. For sample 4, the K⁺ ions are doped in the whole thickness of LN and Er³⁺ ions have a deeper ion concentration profile.

One of the challenges we faced after the doping process are cracks on the LNOI surface that appear during the fabrication process. These cracks make the doped substrate not suitable for further fabrication processing. Scanning electron microscope (SEM) images of the surface cracks for sample 1 are shown in Fig. 4. We have observed similar cracks in all other samples 2 to 4. To find the cause of the cracks, we carried out an experiment with only the annealing process. The results revealed the absence of cracks, which means the cracks stem from the doping process and originate from the damage caused by the presence Er³⁺ ions within the crystal structure.

 

Fig. 4. Surface cracks of doped LNOI substrate. (a) SEM image from top view and (b) cross-section image of the cracks with focused ion beam (FIB) cutting.

Download Full Size | PDF

The volume of Er³⁺ ions in the LN crystal should be reduced to overcome this problem. For samples 5 to 9, we decreased the amount of Er³⁺ ions step by step and observed its effect on the surface cracks. For all these samples, we used 27 mg of KNO₃ and changed the amount of Er(NO)₃ salt. For sample 6 with 0.8 g Er(NO)₃, the surface cracks are observed after doping only at the edges of the substrate, and these cracks propagate through the whole sample during the annealing process (sample 7). Decreasing the mass of Er(NO)₃ salt to 0.6 g results in less cracks (only at the edges even after annealing) and we observed no crack for 0.4 g of Er(NO)₃.

This recipe (used for sample 9) is our optimized recipe for fabrication of crack-free high-quality substrates. The SIMS measurement in Fig. 5 shows the Er³⁺ and Li⁺ concentrations for this sample and we observe that the Er is doped to a thickness of up to 50 nm. We performed this recipe on five identical samples and no surface cracks are observed in any of them. For all these five repeating samples, we observed that the depths of doping of Er ions are between 50 to 90 nm, which shows the repeatability of the doping process. We measured the surface roughness of the doped substrate by atomic force microscopy in a randomly selected region on its surface. The measured area dimensions were 10 µm $\times$ 2.5 µm, revealing a mean roughness depth (R_z) of 1.8 $\pm$ 0.2 nm. This value is comparable to the value observed in an undoped LNOI substrate, suggesting that the doping process has no significant impact on the substrate’s surface quality.

 

Fig. 5. The SIMS measurements from the sample 9 which is optimized to achieve a high quality and crack-free substrate. The measurement shows that the Li⁺ ions are replaced by Er³⁺ ions for a thickness of up to 50 nm.

Download Full Size | PDF

The achieved thickness of 50 nm is relatively thin compared to the total thickness of LNOI layer. While it has been demonstrated in previous studies [22] that a thin layer of Er doping at the surface of the LNOI substrate can be sufficient for the fabrication of functional devices, we tried to increase the penetration depth of the Er³⁺ ions in the crystal of LN by changing the doping procedure. In sample 10, we increased the doping time to 5 hours, and in sample 11, we did two complete doping processes each for 4 hours one after the other. For both of these two samples, the surface cracks appeared after the doping process. Hence, from our experiments, the fabrication recipe used for sample 9 is the optimized recipe to get a crack-free Er-doped LNOI substrate.

3.3 Characterization of Er³⁺ ions by optical measurements

To optically characterize the Er³⁺ ions doped into the LN crystal, we used a setup as shown in Fig. 6. The input light is a pulsed Ti:Sapphire laser (Mai Tai HP, Spectra-Physics) with a wavelength of 980 nm, a pulse width of <100 fs, and a repetition rate of 80 MHz. We control the polarization as well as the power of the input light with a polarizer and a half-wave plate. The input light is focused on the sample with a 20x objective (Mitutoyo with NA = 0.4) and the generated light is collected with the same objective. A long pass filter removes the input frequency and detects only the generated fluorescence light from samples. Another polarizer is used before the detector to characterize the polarization of the generated light. The fluorescence signal is detected with a near-infrared spectrometer (Oxford Instruments - Shamrock 163).

 

Fig. 6. Optical characterization setup to illuminate the doped sample with a 20x objective and collect the generated fluorescence signal with the sample objective. The polarizers and half-wave plate are used to control the input and output polarizations.

Download Full Size | PDF

The Er³⁺ ions have a transition at the telecommunication C-band (⁴I_13/2 $\rightarrow$ ⁴I_15/2), where they absorb light with a wavelength of 980 nm and emit at around 1530 nm [36–38]. To make sure that the detected fluorescence signal is from Er³⁺ ions, we did each measurement on both doped (sample 9) and undoped LN chips and subtracted the detected spectra to get the final spectrum. The final spectra for 5 different points on the same sample are shown in Fig. 7(a). For all these points, the measurement conditions are kept the same and the detected fluorescence spectra have a peak value at 1530 nm. However, the difference in the peak intensity values can be associated with the homogeneity of the concentration of Er³⁺ ions in these points, which shows that the Er³⁺ concentration is not homogeneous on different points of the sample.

 

Fig. 7. (a) The detected fluorescence signal at 5 different points of sample 9. (b) Maximum value of the detected fluorescence spectrum for 5 different input wavelengths from 960 to 1000 nm.

Download Full Size | PDF

In the rest of this paper, for comparing the fluorescence spectra of different measurements, we compare the maximum peak values of the final spectra (subtraction of spectra from doped and undoped substrates under same measurement conditions) as a representation of the whole fluorescence spectrum.

In the next experiment, we checked if the peak in the final spectra is coming from the Er³⁺ ions but not from other elements or materials in the measurement. As mentioned before, we expect that the Er³⁺ ions absorb light at 980 nm and emits at 1530 nm. Hence, we changed the input wavelength of the laser from 960 to 1000 nm and measured the peak values of the spectra. As it is shown in Fig. 7(b), the maximum detected fluorescence is for the input wavelength of 980 nm and it decreases by increasing and decreasing the input wavelength. This corresponds to the ⁴I_13/2 $\rightarrow$ ⁴I_15/2 transition of Er³⁺ ions and indicates that the observed fluorescence originates from Er³⁺ ions.

We changed the input power up to 250 mW for sample 1 and measured the peak values. The results are shown in Fig. 8(a). From this graph and the linear fit to the measurement points, we conclude that the generated fluorescence signal is directly proportional to the intensity of the excitation light at 980 nm.

 

Fig. 8. Effect of varying (a) input power and (b) output polarization on the maximum value of the generated fluorescence spectrum. The fitted lines for the power and polarization graphs are linear and sinusoidal, respectively.

Download Full Size | PDF

Afterwards, we characterized the polarization of the generated fluorescent light from Er³⁺ ions by changing the angle of output polarizer from 0 to 180 degrees. The measurement result (Fig. 8(b)) is almost constant where the measured counts are almost unchanged. However, we detected a weak sinusoidal behaviour. Previously, it has been shown that in Er doped Ti in-diffused waveguides, the Er ions emit both TE and TM polarizations with different cross-sections, which can be the reason for the weak polarity of our detected signal [39]. This effect can be because of the local asymmetry of the LN crystal structure at the place of the Er³⁺ in the crystal lattice. Understanding the main reason for this effect requires further investigations.

One crucial aspect to investigate is the impact of doping on the nonlinear optical properties of LNOI. We check the nonlinearity of the doped LNOI (sample 1) by comparing the second-harmonic generation (SHG) between the doped and undoped samples. In this regard, we used a nonlinear microscope setup and illuminated sample 1 and an undoped sample with a femtosecond pulsed laser (Mai Tai HP, Spectra-Physics) together with an optical parametric oscillator (OPO Inspire HF 100). The laser has a pulse width of 100 fs, repetition rate of 80 MHz, FWHM of 10 nm, a wavelength of 1500 nm, and an average power of 8.2 mW. The corresponding plot of the measurements is shown in Fig. 9, where the SHG counts for different excitation polarizations are shown as blue and orange for doped (sample 1) and undoped samples respectively. As expected, the nonlinear properties of the LN crystal are affected significantly since the Li⁺ ions are replaced with K⁺ and Er³⁺ ions. This is the penalty of our low-temperature doping method and it is important for considering this doping method to fabricate substrates for PIC applications. One of the solutions to this problem is to dope the sample locally in only the parts of the sample which should be used as active elements.

 

Fig. 9. The SHG signal from doped (sample 1) and undoped LNOI substrates.

Download Full Size | PDF

3.4 Local doping

We used a layer of SiO₂ with a thickness of 350 nm to protect the LN from the doping process. We covered half of a sample with SiO₂ using the electron-beam physical vapour deposition method. Afterwards, the sample is processed and doped with the optimized doping methods. After the process, SIMS measurements at two points with and without SiO₂ mask are used to characterize the doping process. The results are shown in Fig. 10(a). In this figure, the schematic of the sample with and without a mask, as well as the distribution of the Er ions are presented. The SIMS measurements for concentration of Er ions for masked and unmasked areas are shown on the left and right sides of the schematic. As we see, a thickness of more than 350 nm SiO₂ mask can effectively block the Er doping process, however, in the unmasked area, a 50-nm-thick layer of LN is doped.

To compare doped and undoped areas, the SiO₂ layer is removed by etching in hydrofluoric acid (HF) for 4 minutes. Fig. 10(b) shows a microscope image of an area of the sample close to the border between the masked and unmasked area. We can see, these two regions are clearly distinguishable. From these results, we conclude that the method is effective for local doping. In this way, only small areas are affected, while the other parts of the substrate are kept unaffected for integration of other functional devices on the same PIC.

 

Fig. 10. Local doping of LN using SiO₂ mask layer. (a) Schematics of the doped sample with a SiO₂ mask layer on part of the sample. The left and right graphs show the SIMS measurement of the density of Er³⁺ ions in the areas with and without mask respectively. (b) Microscope image of the sample after removing the mask with HF acid. The doped and undoped areas show a contrast in the microscope.

Download Full Size | PDF

4. Conclusion and outlook

This work used an ion-exchange method using molten salt to dope Er³⁺ ions in an LNOI substrate. We fabricated several samples and found an optimized recipe to dope the crystal without any damage and surface cracks. With our recipe, we doped a 50 nm surface layer of LN. We characterized the doped LN with a reflection optical setup with an input wavelength of 980 nm and detected wavelength of 1530 nm. The generated fluorescence is linearly dependent on the input power and is unpolarized. We have observed that our doping method affects the nonlinearity of the LN crystal since the Li⁺ ions are replaced with Er³⁺ and K⁺ ions. We implemented local doping by using an SiO₂ mask to overcome this problem. In the next step, we would fabricate nanostructures such as resonators and metamaterials to investigate the effectiveness of our doping method compared to other methods.