Complex Refractive Index Extraction for Ultrathin Molybdenum Oxides Using Micro‐Photonic Integrated Circuit Chips

The emerging 2D nanomaterials with unique optical properties are promising for next‐generation miniatured on‐chip devices. One of the prerequisites is to precisely measure their optical parameters during their implementation. However, the inherent features of 2D layers, including limited lateral dimensions and ultra‐small thicknesses, are not favorable to the conventional characterization techniques applied in the bulk system, especially for optical complex refractive indices measurement. Here, this work proposes a silicon photonics‐enabled platform to evaluate the complex refractive indices of ultrathin 2D materials in a facile and reliable manner. Ultrathin molybdenum oxides (MoOx) with multiple stoichiometric states are selected as the target 2D material to provide sufficient complexity of the system for investigation. Upon the integration of ultrathin MoOx, the silicon photonic chip, in the form of a Mach‐Zehnder interferometer, exhibits wavelength shifts which are used for calculating the optical complex refractive indices. Compared with the theoretical calculation, the deviation is as low as 1% and generally less than 5%. This work demonstrates a highly accurate and reliable approach for measuring the complex refractive index of 2D films, possibly assisting future advances in 2D materials‐enabled optical and photonic applications.


Introduction
2D nanomaterials, with few nanometer thicknesses and large lateral dimensions, enjoy highly confined electrons and MoO 3 showing a layered phase, [7c,12] sub-stoichiometric MoO x in a non-layered or amorphous property. [13] MoO x has a high refractive index and a great number of defective states, which offers it highly tunable electronic and optic properties. Combined with the benefits of ultrathin thickness and large surface area, 2D MoO x can support surface plasmon resonance and vast surface functionalities, and hence plays a vital role in aspects of biosensing, photocatalyst, etc. [14] Most recently, the wafer-scale synthesis approach of ultra-thin metal oxides further extends their potential toward practical applications, [15] including biochemical sensing, [16] advanced imaging, [17] telecommunications, [18] energy storage, [19] and catalysis, [20] etc.
Precisely characterizing the 2D materials' optical and electronic properties is an important prerequisite for 2D materialenabled applications. Due to their ultra-thin nature and usually limited lateral dimension, it is challenging to characterize their optoelectronic properties through direct measurement. Particularly, the complex refractive index, indicating the optical refractive index and extinction coefficient of the 2D materials, is of key importance for optoelectronic device design. The commonly used techniques for measuring the complex refractive index include interferometer, [21] holography techniques, [22] ellipsometry, [23] and near-field measurement approaches, [24] etc. However, most of these approaches suffer from various limitations, such as a large and homogeneous material surface is required for direct measurements, [25] while the near-field approach needs a sophisticated setup and low tolerance to environmental disturbance. [24c,26] Furthermore, the high tunability of some 2D materials makes it more challenging to accurately determine their complex refractive indices in situ.
Silicon photonics is a mature technology with ready-to-go components such as passive optical waveguides, photodetectors, and modulators. [27] In particular, silicon-based waveguides and interferometers are highly sensitive to slight refractive index changes introduced by analytes or nanomaterials attached to the waveguides' surface, and translate those variations into measurable outputs like wavelength shifts. [28] In addition, 2D material-based research has made significant advances in the past decades, that the layers can be customized on the specific area of silicon photonics chips to create functional devices. [29] The hybrid of coplanar 2D material and waveguide can significantly increase the material's interaction length with light that travels in a lateral direction. Thus, even tens of micrometers-sized 2D layers adsorbed within the evanescent field of the silicon waveguide channel can arouse detectable output, [30] which in reverse makes integrated silicon photonics favorable to precisely capturing the complex refractive indices of 2D materials.
In this paper, we proposed an innovative platform based on silicon photonics to determine the optical complex refractive indices of ultrathin materials. By integrating the 2D MoO x nanofilms onto a silicon photonic waveguide, the optical output signal variation aroused by the thin film can be detected and utilized to extract its optical refractive index and absorption. [30a] we have first experimentally measured the complex refractive index of the metal oxide thin film using an asymmetric Mach-Zehnder interferometer (MZI) structure on a silicon-on-insulator (SOI) platform. The MoO x was deposited, and in situ tuned in a cycle from MoO 3 to MoO 2 on the waveguide. Their complex refractive indices at seven different stoichiometric states were then extracted through our proposed method and compared with the ones obtained from ellipsometer measurement. The theoretical and experimental results are well aligned. This proposed methodology for extracting complex refractive indices adds an alternative pathway for investigating the optical property of ultra-thin 2D material and possibly assists their enabled optoelectronic device development.

Material Characterization
The ultra-thin molybdenum (Mo) oxide nanofilm was prepared through precisely controlled oxidization or reduction of the deposited Mo film by tuning the operation temperature and oxygen concentration (Figure 1a). The oxidation-reduction process follows the reaction equation as below: The annealing temperature of MoO x samples is maintained at a relatively low value to avoid sample crystallization under excessive heating (see Experimental Section). This is because the formation of crystalline islands leads to the crack of the film ( Figure S1, Supporting Information). As a result, a significant scattering loss of light is occurred, [31] and the circular tunability of the film is hindered. In the meantime, excessive heating also causes the evaporation of the film. [32] The thickness and morphology of ultrathin Mo and MoO x films were measured using an atomic force microscopy (AFM). The freshly deposited Mo metal nanofilm exhibits a thickness of 7.5 nm ( Figure S2, Supporting Information) while the oxidized MoO 3 nanofilm appear slightly thicker at ≈10 nm ( Figure S3, Supporting Information). The further tuning of the MoO x nanofilm by the oxidation-reduction process does not create significant variation on its thickness as shown in the Figure 1b, Figures S4 and S5, Supporting Information. The large variance in thickness at the edge of the nanofilm is due to the lift-off process. Large-area deposition of the nanofilm on substrates is also feasible ( Figure S6, Supporting Information). A cross-sectional transmission electron microscopy (TEM) image ( Figure 1c) shows a representative MoO x on a silicon substrate. No obvious crystal lattice structure is observed from the sample, indicating that the MoO x nanofilm is amorphous. Selected area electron diffraction (SAED) pattern of the corresponding area was obtained with no diffraction pattern shown, which agrees with the observation from bright field TEM. The X-ray diffraction (XRD) results of the nanofilm in different conditions ( Figure S7, Supporting Information) display a peak at ≈22°which originates from the silica slide. [33] No molybdenum or molybdenum oxide peaks are observed which again confirms the amorphous nature of the prepared nanofilm.
X-Ray photoelectron spectroscopy (XPS) was performed to verify the chemical composition of the MoO x nanofilm. Figure 1d shows the Mo 3d spectrum of the as-deposited molybdenum film (labeled as Mo). From the figure, two peaks at ≈228.0 and ≈231.15 eV can be identified as Mo metal 3d 3/2 and 3d 5/2 , respectively. The Mo 6+ doublets are located at ≈232.3 and ≈235.45 eV. By deconvoluting the spectrum, the other two pairs of doublets can be found at ≈230.91 and ≈228.85 eV, which are assigned to Mo 5+ and Mo 4+ , respectively. [34] The initial native oxide on the deposited Mo metal surface is found to be sub-stoichiometric. The full-stochiometric MoO 3 was obtained by oxidizing the deposited Mo nanofilm in an ambient air environment (see Experimental Section). After that, the MoO x nanofilm was tuned under 5 successive steps to obtain different sub-stoichiometric states of the material, and the corresponding samples were labeled as R1, R2, R3, O1, and O2 in turns (detailed operation conditions are presented in the Experimental Section). A comparison of the Mo 3d spectra of all the samples (Figure 2a) illustrates the changes in Mo's stoichiometry. On the top is the spectrum of the MoO 3 nanofilm which has a dominant Mo 6+ doublet located at 232.4 and 235.5 eV. The MoO 3 nanofilm was then annealed in an oxygen-deficient environment to obtain the reduced nanofilm R1. The dramatically increased Mo 5+ doublet at 231.2 and 234.3 eV in the R1 spectrum indicates the formation of oxygen vacancies in the nanofilm. [35] Further twice reductions were performed on the sub-stoichiometric R1 sample to obtain R2 and R3, and the corresponding XPS spectra show that the intensity of Mo 5+ peaks increases and Mo 4+ peaks start to appear. This result implies the creation of more oxygen vacancies after further reduction of the nanofilm. Compared with the R2 result, the Mo 4+ peaks at 229.5 and 232.7 eV take more portion of the total Mo 3d spectrum while the percentage of Mo 5+ alters little. [36] Considering that Mo 5+ is a transition state of Mo ions, the increased ratio of Mo 4+ over Mo 5+ indicates that the sample is in transient from MoO 3-x to nearly MoO 2. Later, the oxidation processes were conducted in air at 200 and 250°C to produce the O1 and O2 samples. The Mo 4+ peaks diminish in the Mo 3d spectrum of the O1 and the intensity of the Mo 5+ doublet starts to drop which eventually decreased to a negligible value in the O2 sample spectrum. At this stage, the sample is again oxidized to full-stoichiometric MoO 3 . To have a clear view of changes in samples' stoichiometry, the percentage of Mo 4+ , Mo 5+ , and Mo 6+ of the as-synthesized MoO x nanofilm R1, R2, R3 and O1, O2 are calculated from the Mo 3d spectra, and the corresponding values of each sample are presented in Table S1, Supporting Information.
In addition to chemical composition, the nanofilm's tunability in optical absorption was examined using UV-visible (UV-vis) spectroscopy ( Figure 2b). The deposited Mo metal nanofilm has a broad absorption band with a peak at 255 nm which represents the interband absorption of its native oxide layer. [37] The interband absorption peak of the nanofilm is narrowed and shifts to 260 nm after oxidizing the metal to MoO 3 , and further redshifts to 268 and 272 nm in R1 and R2, respectively. The peak-shift to lower energy reflects the narrow in the band gap of MoO x after reduction. [38] Moreover, the interband peak blueshifts back to 261 nm in O1 and O2 when the sample was tuned from sub-stoichiometric MoO x to stoichiometric MoO 3 again as indicated by the XPS result. In addition to the shifts of the interband absorption peak position, the absorption spectra of the sample R1, R2, and R3, have a moderate absorption peak centered at ≈540 nm. The appearance of this peak possibly indicates the occurrence of oxygen vacancies associated with the substoichiometric Mo ions [16b,39] after the reduction as it diminishes in the spectra of O1 and O2 after the oxidation of the nanofilm. In addition to the absorption spectra, the photoluminescence (PL) spectra ( Figure S8, Supporting Information) of the nanofilms under 375 nm laser excitation exhibit peaks at 440 and 470 nm which both are associated with Mo 5+ defect centers. [40] These two peaks become stronger as the sample is tuned from R1 to R3, implying the formation of oxygen vacancies during the reduction process. On the other hand, their intensities drop in O1 and O2, suggesting the refilling of oxygen atoms after oxidation in the air. Peaks at 402, 409, and 417 nm are a result of the laser beam undergoing multiple reflections and refractions on the silica substrate that was utilized for the MoO x nanofilm fabrication ( Figure S9, Supporting Information).
The complex refractive indices of the as-prepared wafer-scale MoO x nanofilms were estimated through ellipsometry measurement. Its real part, refractive index (n), and the imaginary part, extinction coefficient (k) at the communication wavelength ( = 1550 nm) of all samples are extracted and summarized in Figure 2c. The refractive index (n) varies from ≈2.0 to 2.7 when the MoO x nanofilms are tuned into different states, at the same time the extinction coefficient (k) ranges from ≈0.2 to 1.0 accord-ingly. Among all the samples, the deposited Mo nanofilm has the highest n and k value due to its metallic nature. The n value of R1 to R3 gradually decreases compared with the n value of MoO 3 nanofilm, which implies the lower density of the nanofilm after reduction due to loss of oxygen in the structure. The rise in Mo 4+ and Mo 5+ peaks in the corresponding XPS spectra also shows the increase in the number of oxygen vacancies from R1 to R3. The n values of O1 and O2 rise to 2.47 from 2.0 of R3 as gaining in molecular weight after the oxidation in air. This can also be reflected in the attenuation in the Mo 4+ and Mo 5+ peaks when the amount of oxygen vacancies drops. In the meantime, the changes in k values also illustrate that the more sub-stoichiometric, that is, less Mo 6+ , the sample is, the larger the k value is and the higher the absorption is.

Optical Characterization of the Waveguides
To demonstrate the capability of the proposed method to extract the complex refractive indices of MoO x nanofilms, we integrated MoO x on silicon photonics circuits by depositing a thin layer of Mo onto a Si waveguide (see Experimental Section). The optical waveguides used are 220 nm-height and 450 nm-width silicon waveguides fabricated on an SOI platform. An asymmetric MZI device was used as illustrated in Figure 3a, showing a longer arm as the modulating arm and the shorter one as the reference arm. The length difference (ΔL) between the arms will lead to a slight phase difference for light to pass through, forming an interference pattern at the optical output. [41] A layer of ≈10 nm MoO x was deposited onto a specific area of the modulating arm as indicated in the Figure 3a, and the length of the material is 75 μm. The MoO x nanofilm that adheres uniformly on the waveguide, as indicated by the dashed line in Figure 3b,c, is of an averaged surface roughness of 361.92 pm determined by AFM. The cross-section view of the waveguide is illustrated in Figure 3d, and the corresponding transverse electric (TE) mode (Figure 3e) inside the silicon waveguide was simulated using in-house Full-vectorial Eigen-mode-expansion software REME. [42] After the deposition of MoO x , the optical mode field (Figure 3e(ii)) slightly shifts up compared with the bare silicon waveguide (Figure 3e(i)) due to the high refractive index of MoO x film, leading to the weaker optical confinement in the silicon waveguide.
Based on the refractive index (n) and extinction coefficient (k) obtained from the ellipsometer and the thickness of MoO x nanofilm, the effective refractive index (n eff ) and effective extinction coefficient (k eff ) values at 1550 nm are calculated accordingly using the REME (Figure 3f). The n eff after incorporation with the MoO x layer ranges from 2.28 to 2.3 while the k eff ranges from 0.005 to 0.027, which both share the exact trend with the measured n and k values of MoO x . The n eff is higher than the value of the bare silicon waveguide but lower than the n values of MoO x itself. Similarly, the k eff of the bare device is nearly zero, the larger k eff after incorporating the MoO x onto the device implies that extra absorption loss is introduced by the material. Knowing the ΔL, the theoretic interference wavelength ( ) after material deposition can be obtained using the calculated n eff of each condition of MoO x and the interference index (m) (details in Note S1, Supporting Information). The difference (Δ ) between calculated and interference wavelength without material ( 0 ) is plotted in Figure 3g. It can be known that the of the asymmetric MZI device with the MoO x layer red-shifts as the n eff becomes larger, and the exact number of shifted periods can be calculated by dividing the free spectral range (FSR). For instance, the n eff of the device after Mo nanofilm deposition changes from 2.268 to 2.296, hence the shifts ≈19 nm, that is four periods of FSR (≈5.36 nm) and extra ≈3 nm. The calculation indicates that small shifts in the interference pattern occur when the n eff changes due to further tuning the nanofilm on the waveguide. As the result, n eff ( Figure 3f) and Δ (Figure 3g) are directly related to each other: the shifts red and Δ increases when n eff increases and the blue-shifts, Δ drops when n eff decreases. Therefore, tuning the properties of the MoO x layer can eventually influence the optical mode of the MZI device. On the other hand, by analyzing the interference wavelength shift and the transmission power variation in the MZI device, n and k values of the deposited MoO x nanofilm can be extracted.
As illustrated in the schematic diagram (Figure 3a), the experiment set-up employed a continuous wave laser source connected to one side of the device through an optical fiber as an input signal and a photoreceiver on the other side to measure the transmission spectra over the wavelength. A set of seven straight waveguides with different material coverage are first designed to explore the relation between the absorption intensity and the amount of material. Each waveguide in the group is covered by a rectangular area of 10 nm-thick MoO x whose length ranges from 10 to 45 um with an increment of 5 um. The propagation loss variation due to the differences in MoO x coverage area can be observed from the intensity changes in the transmis-sion spectra of the waveguide set, where the propagation loss from material absorption grows gradually as the coverage increases. The loss coefficient of the MoO x nanofilm is then calculated by the cutback method by plotting the peak transmission power of each waveguide in the group which is determined by averaging the maximum intensity over 1 nm wavelength range, against the length covered with the MoO x nanofilm. [31] Take the sample R2 as an example, the maxima transmission power of each waveguide (indicated by the squared range in Figure S10, Supporting Information) exhibit a clear decline trend as the coverage of the material increases is observed (Figure 4a) because of the higher extinction that light experiences when it passes through the waveguide with a larger area of material. A linear relation can be found between the area of MoO x on the waveguides and the light transmission, which is fitted in the dashed line in Figure 4a. The line fitting clearly indicates the decreasing trend in power as the sample R2 gets longer whose slope reflects the propagation loss of light per unit length of material (≈0.06 dB μm −1 ) and can be used to describe the absorbance of the MoO x in R2 condition. The transmission spectra of the device group with deposited Mo and MoO x nanofilm in all the other conditions ( Figure S11, Supporting Information) share similar observation results. For example, the propagation loss of the device with deposited Mo is ≈0.07 dB μm −1 obtained from the linear fitting, which is presented in Figure S11, Supporting Information, right column. Its metallic nature results in the high absorption of the device when Mo was first deposited onto the waveguides. After oxidation to MoO 3 , the loss of the device drops dramatically and the intermediate state sample R1 has an absorption of 0.01 dB μm −1 . The absorption per unit length of the reduction state R1, R2, and R3 becomes stronger and stronger and reaches 0.08 dB μm −1 as the amount of sub-stoichiometric Mo 5+ and Mo 4+ was raised. The propagation loss due to the material absorption decreases to 0.01 dB μm −1 again in the case of sample O2 since the MoO x nanofilm is oxidized from nearly MoO 2 to almost full stoichiometric MoO 3 . The unit absorption of MoO x nanofilm is summarized in Figure 4b along with the unit transmission power of all sample conditions to examine the variation of the material's absorbance against its condition. In Figure 4b, the deep gradient from R1 to R2 indicates that the reduction result of MoO x nanofilm on the device is dramatic at the beginning and turns moderate after reaching a certain level. As the reaction condition is kept same for the R2 and R3, this means that MoO x approaches the operation limit. The gradual and sharp change in the measurement result after the device is treated by another cycle of oxidation. This change in absorbance reflects stoichiometry changes of tuning of MoO x as discussed in the XPS result. Moreover, it can be found that after the reduction of the MoO x nanofilm, the transmission slope becomes steeper and the k eff value rises while the oxidation process leads to drops in the transmission loss and k eff , which means that the waveguide propagation loss is majorly due to the material absorption and is closely related to its condition. As the result, the intensity of the device transmission can be used to estimate the k value of the material.
The variation in amplitude and the phase shift in the interference pattern caused by refractive index differences can be observed from the transmission spectra of the asymmetric MZI device with MoO x nanofilm. The normalized result of two  Figure S12, Supporting Information, along with the spectrum of the bare device. A 1.56 dB drop in device transmission is introduced due to the deposition of Mo nanofilm. Although oxidation of the nanofilm to MoO 3 does not affect the intensity very much, a clear decrease is observed in the spectrum of R1. The transmission intensity declines 4.23 and 7.41 dB for sample R2 and R3, respectively which reflects the rise in propagation loss due to material absorption. The propagation loss recovers for the sample O1 and O2. The O1 results in a transmission level between R1 and MoO 3 while the one of O2 has the almost same level as the bare device. The overall trend in the change of asymmetric MZI device propagation loss reconciles with the observations from the straight waveguide group.
In addition to the amplitude changes, the phase of the spectrum shifts corresponding to the tuning of the material. When the nanofilm is tuned into R1, the is ≈1.0 nm larger than the measurement of the bare device. Moreover, reducing the sample to R3 leads to another 1.1 nm-shift of the spectrum, which corresponds to 2.1 nm from the of the bare device. According to the theoretical calculation, the for a device with sample R1 will redshift 17.24 nm, which corresponds to 3 FSRs and 1.2 nm. For the device with sample R3, the calculated is 12.22 nm, corresponding to 2 FSRs and 1.53 nm. The theoretical shift of sample R1 is only 0.1 nm different from the experiment result which is 1% of the total shift. While the difference between the calculated and measured value of the sample R3 is also within 5% upon the subtle shift of , which is probably due to the variance introduced during the preparation of the nanofilm.
According to the theoretical model in the previous section, the interference wavelength will redshift because of the high refractive index of Mo and MoO x . As shown in Figure S12, Supporting Information, an 18.9 nm redshift in the interference pattern occurs for the device with the Mo nanofilm, corresponding to 3 FSRs and 2.83 nm. While the difference in wavelength grows to 21.05 nm after oxidation to MoO 3 , which corresponds to 3 FSRs with 4.98 nm. When the nanofilm is reduced from full stoichiometric MoO 3 to MoO x (R1, R2, and R3), the amount of wavelength shift (Δ ) drops as the interference pattern blueshifts with the decrease in the corresponding n eff . For instance, moves 3.95 nm to shorter wavelength from MoO 3 to R1, which is more than one FSR, and further blueshifts 0.7 and 3.7 nm for the sample R2 and R3, respectively. Hence, Δ of the sample R3 gives rise to 12.73 nm blueshift, just over 2 FSRs, as it has the smallest n value among all the samples, which presents as an interference wavelength located at 2.05 nm longer than the peak of the bare device. In contrast, the wavelength shift increases when the nanofilm on the device is oxidized from the sub-stoichiometric state (O1 and O2). As the n value of O2 is larger than the one of O1, the corresponding interference pattern of O2 shifts 0.5 nm longer than O1's. The interference wavelength differences measured from these spectra are summarized along with the calculated value against the sample condition in Figure 4d. To obtain the complex refractive index from the optical propagation loss of the device, the measured interference wavelength shift is compared with the calculated value for all the conditions (Table S2, Supporting Information) to evaluate its feasibility first. Surprisingly, the experimental and theoretic values highly align with each other as illustrated in Figure 4d, which demonstrates the experimental measurement experiences appreciable accuracy with the theoretical model. Therefore, the measured interference shift and refractive index of the material on the device will enjoy a oneto-one relationship. Table 1 summarizes the n and k values of MoO x in each condition at 1550 nm and the corresponding interference wavelength shift measured from the transmission spectra, which can be used to find the complex refractive index of the material in future experiments. In addition, Tables S3 and S4, Supporting Information, list all the n and k values of MoO x in each condition at 850, 1300, and 1550 nm as the reference. Overall, the refractive index of MoO x slightly rises as the wavelength increases from 850 to 1550 nm for all the conditions while the absorption coefficient does not exhibit an explicit wavelength dependency.

Conclusion
We have successfully demonstrated a highly accurate method to extract the complex refractive index of ultrathin 2D nanomaterial using an on-chip integrated optical waveguide. The MoO x nanofilm with 10 nm thickness was synthesized using E-beam evaporation deposition and tuned into different stoichiometric states which exhibit various refractive indices and extinction coefficients. The ultrathin MoO x film used as a proof-of-concept was precisely integrated onto an asymmetric MZI device whose optical transmission spectrum at communication wavelength will be influenced by changes in the complex refractive index of the nanofilm. We facilitated in situ tuning of the MoO x nanofilm on the device and recorded the wavelength shift and intensity change of the transmission spectra, which are caused by the slight varia-tion in refractive index and extinction coefficient, respectively. A theoretical model is adopted to calculate the interference wavelength shift based on the refractive index obtained using ellipsometry from wafer-scale MoO x nanofilm, and less than 0.5 nm difference between the theoretic shift and experiment result was achieved. The result hence proved that the interference wavelength shifts of the R1, R2, R3, O1, and O2-covered MZI devices correspond to measure refractive indices. This one-to-one relation provides an accurate and convenient pathway to determine the complex refractive index of nanomaterial solely through their output optical transmission. We believe this work offers a new direction for characterizing the optical properties of various types of ultrathin nanomaterial regardless of its size and morphology, which may eventually advance the development of on-chip photonic applications in sensing and communication.

Experimental Section
Ultra-Thin MoO x Film Preparation: Molybdenum nanofilms on the optical waveguide were prepared using an electron beam evaporator deposition system (Kurt J. Lesker PVD75, Country) under a non-reactive environment. The molybdenum metal chunks (99.99% American Elements) were used as an evaporation source. Prior to the deposition, the target deposition area was specified using the maskless photolithography technique (Heilberg Instruments MLA150). After the evaporation, the chip was lifted off in the acetone solvent.
The waveguide chip with Mo metal nanofilm was oxidized in the air on a hot plate at 250°C for 10 min to obtain the MoO 3 nanofilm. The chip was then processed into five different states successively, in the order of R1→R2→R3→O1→O2, in which "R" refers to "reduction" and "O" refers to oxidation. Either reduction or oxidation can be realized by simply playing the matrix of temperature and oxygen level. The detailed synthesis parameters for each condition are listed as follows: All samples were annealed using a hotplate with the temperature increasing rate of 5°C s −1 . The reduction process was carried out under an inert gas environment with an oxygen level at 0.1 ppm, while the oxidation process was conducted in air. And a glovebox accompanied with an automated circulating/purification system (SIEMENS programmable logic controller (PLC)) was used to provide a low-oxygen environment for sample reduction experiment.
Morphological, Structure, and Optical Characteristics: The thickness and surface roughness of deposited MoO x nanofilm were examined using a Bruker Dimension Icon AFM. The acquired AFM images were analyzed using a Gwyddion software. The morphology of the nanofilm on the waveguide was studied with an FEI Nova nanoSEM 200. Bright-field optical images of the waveguide were captured using a BX41LED-Olympus (7418) microscope. The TEM imaging, SAED pattern and cross-sectional view of the MoO x nanofilm were measured using a Jeol JEM-2100F HRTEM.
The MoO x film on the silicon substrate was covered by a 50 nm Au capping layer via E-beam evaporation deposition first to protect the material from exposure under the ion beam for cross-section sample preparation. Then, the lamella sample was cut by the focused ion beam (FEI Scios Dualbeam FIBSEM) and in situ transferred to a TEM grid. A Bruker D4 ENDEAVOR was used for XRD measurements (monochromatic Cu K source = 0.154 nm) of samples on a glass substrate. XPS measurements were carried out using a Thermo Scientific K-alpha XPS (monochromated XPS, Aluminium K 1486.7 eV). UV-vis-NIR spectroscopy was performed using a CRAIC Apollo Microspectrophometer (CRAIC Tech) to obtain the absorption spectra of MoO x nanofilm on quartz substrates. Steady-state photoluminescence spectra of the samples were acquired by a QE Pro fluorimeter (Ocean Optics) under 375 nm excitation. The refractive index and extinction coefficient were acquired by Woollam Ellipsometer M-2000.
Optical Waveguide Fabrication: The silicon optical waveguides used in this paper were fabricated on a silicon-on-insulator (SOI) platform with a 220 nm guiding silicon layer and 3 μm buried oxide layer using electronbeam-lithography (EBL) and reactive ion etching (RIE). The width and height of the silicon optical waveguide were 450 and 220 nm, respectively, with grating couplers as input and output interfaces. The straight waveguide was ≈3 mm long while the asymmetric MZI-structured waveguide is constructed with a modulating arm of 980 μm and a reference arm of 880 μm with a difference of 100 μm.
Optical Waveguide Measurement: A grating coupler measurement setup was used to measure the transmission spectra of MoO x -waveguide devices. An Agilent 8164B tunable laser was used as the source for the measurement at ≈1450-1640 nm, and the built-in InGaAs photodetector was used for measuring the output optical signal. The input transverse electric (TE) light was provided to the grating coupler on one end of the device via a polarization-maintaining optical fiber, while the output light was collected by another optical fiber from the grating coupler on the other end of the device and measured with an optical detector.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.