1. Introduction

A Raman spectrum records the inelastic photon scattering through transfer of photon energy into excitations of low energies in the meV range. Raman spectroscopy is a non-contact and non-destructive method applicable to nearly all material classes that provides insight into the structural properties of a sample, such as crystal structure, strain, defects, and impurities, making it a highly effective and widely used instrument for characterizing materials [1]. Heterostructures are the basis of modern devices, therefore they became the most common samples for Raman spectroscopy studies [2–10]. Heterostructures consist of thin film layers deposited on a substrate which determines the crystalline properties of the films, ultimately linked to their electrical and optical performance. Substrate choice directly affects the presence of epitaxy, epitaxial orientation, crystal structure, and interface quality. Raman spectroscopy can help characterize various substrate induced effects. In fact, it can detect the influence of the substrate on the crystal structure, such as the presence of epitaxial strain [2,4]. Even in the absence of a direct influence of substrate on the crystal structure, for example in the case of transferred 2D materials, Raman spectroscopy can detect the effect of the screening properties and polarizability of the substrate on the optical and electrical properties of the films [11,12].

Thin film substrates are typically selected based on the lattice parameter mismatch between the film and the substrate with the goal to introduce a specific phase or level of strain in the thin film. However, if the sample is later to be studied by Raman spectroscopy, the optical response of the substrate is also of paramount importance. For instance, Raman characterization of the thin film is nearly impossible when the substrate below exhibits a rich and intense optical signature in the wavenumber region of interest for the thin film under investigation. When choosing a substrate for the Raman spectroscopy study of a thin film, one thus needs to ensure that the film’s active Raman modes of are not masked by the substrate response. Here, we aim to compare the suitability of common thin film growth substrates for Raman spectroscopy by measuring the relative intensities of spectral features, defining the free spectral range with no strong Raman response, and analyzing the background intensity of the optical response.

Though spectra of common substrates have been previously reported in the literature [11,13], and some can even be found in online databases, published spectra of individual substrates are often reported in arbitrary units and collected on different setups, preventing a direct comparison. We therefore do not aim to report new or discuss old Raman spectra, but to compare them in terms of intensity and density of spectral features. To the best of our knowledge this is the first comprehensive study showing Raman spectra for a number of substrates side-by-side. A single previous study examined the Raman response under a 514 nm laser excitation for ten substrates by comparing the intensity of most prominent spectral features, indicating the positions of other spectral features, and extrapolating the intensity of the Raman background [13]. However, the full spectra of each substrate were not shown, disabling the possibility to visualize the relative intensities of the peaks in each and across the substrates for wavenumbers other than the most prominent spectral feature. Furthermore, no attempt was made to evaluate the concomitant photoluminescent background signal of the substrates. In contrast, in this study we compare the optical response under two different laser excitations (633 nm and 785 nm). The dual excitation wavelength allows for a combined fingerprint of all substrates in different wavenumbers (higher and lower respectively) regions. We show the relative intensities of the spectral features for each and across the substrates, and discuss the intensity of the overall background, not only the Raman background.

2. Materials and method

We analyze the optical response for the following fourteen common substrates used in depositions of thin films: YAlO₃ (YAO), LaAlO₃ (LAO), NdGdO₃ (NGO), (LaAlO3)_0.3-(Sr₂AlTaO₆)_0.7 (LSAT), SrTiO₃ (STO), 0.5% doped Nb:SrTiO₃ (Nb:STO), DyScO₃ (DSO), GdScO₃ (GSO), MgO, rutile TiO₂, α-Al₂O₃, Y:ZrO₂ (YSZ), Si with a buffer of 150 nm of amorphous SiO₂ (SiO₂|Si), and Si. The samples utilized in our study were obtained from Crystec Gmbh, adhering to a criteria of root mean square roughness Rq below 0.1 nm (Supplement 1 (a)).

Raman spectroscopy is performed using a CW single-longitudinal-mode 633 nm and a 785 nm SLM laser (Integrated Optics) with an output power measured on the sample surface of 1.5 and 0.2 mW, respectively. The Raman signal is collected using an Andor Kymera 328i spectrometer coupled to a Newton EMCCD camera (Oxford Instruments). The optical path consists of bandpass filters, notch filters and silver mirrors guiding the laser to the sample and the spectrometer (Supplement 1 (b)).

The choice of laser excitation wavelength can have a significant impact on the optical response of the sample in several ways. First, changes in total intensity and intensity ratios of features typically result from moving between non-resonant and resonant Raman scattering [14,15]. Second, an optical excitation will excite not only the Raman response, but also photoluminescence which can be more pronounced at specific excitation wavelengths. It is therefore instructive to compare the optical response of several excitation wavelengths [16]. Further benefits of multiple excitation wavelengths are access to a wider spectral range of study, sensitivity to probing depth and surface phenomena [6]. Moreover, in our 785 nm Raman system, we can access the low wavenumber range (< 240 cm^-1). For this study, we cover 237-1750cm^-1 and 5-1340 cm^-1 with the 633 nm and 785 nm laser systems, respectively.

3. Measured data

We analyze the optical response of substrates with perovskite crystal structure and lattice constants below 4 Å (section 3.1) and substrates of binary oxides and Si with lattice constants above 4 Å (section 3.2), highlighting their lattice constants and bandgaps (Table 1–2) and their Raman spectral features under 633 nm and 785 nm laser excitation (Fig. 1, 2).

3.1 Raman spectra of perovskite substrates

All perovskite substrates exhibit significant and distinctive spectral characteristics. Above 1250 cm^-1 (150 meV), a growing background response is observed with the 633 nm laser system. This is a typical photoluminescence signature, which appears as a broad, intense band centred at extremely high wavenumbers. As the 785 nm system is less energetic, the background remains unchanged [23]. For the YAO substrates under 785 nm excitation, spectral features at 1199 cm^-1 and 1323 cm^-1 become dominant over all features present at lower wavenumbers (< 700 cm^-1) under 633 nm excitation (Fig. 1(a)). LAO, NGO, LSAT, STO and Nb:STO show similar spectral features in both excitation systems (Fig. 1(b)-(f)), with NGO showing dominant peaks in the low (< 700 cm^-1) and high (> 1000 cm^-1) wavenumbers range for the 633 nm and 785 nm excitation wavelength, respectively. Though the STO and Nb:STO spectra are almost identical under the 633 nm excitation, the dominant feature of the Nb:STO (Fig. 1(g)) for the 785 nm excitation is in the low wavenumber range (< 100 cm^-1), likely attributable to the Nb dopant. However, the spectral features of Nb:STO remain similar to the ones observed with the 633 nm laser when the effect of the dominant feature is removed (Fig. 1(g) - inset). Some changes are observed between the 633 nm and 785 nm spectra of DSO (Fig. 1(g)) due to a changes in their peaks relative intensities. GSO shows well defined spectral peaks in the 785 nm system (Fig. 1(h)). When extending the available range beyond 1300 cm^-1 by using the 633 nm laser, the drastic increase in background becomes the dominant feature. Raman peaks position, dominant spectral features, and frequency range free from Raman activity are reported in Table 2 for both 633 nm and 785 nm laser excitation.

Table 1. Lattice constant and band gap of common perovskite substrates

View Table | View all tables in this article

 

Fig. 1. (a)-(h) Normalized Raman spectra of perovskite substrates with lattice constant below 4 Å under 633 nm (black) and 785 nm (red) excitation lasers.

Download Full Size | PDF

Table 2. Raman spectral features and free spectral range of common perovskite substrates

View Table | View all tables in this article

3.2 Raman spectra of binary oxides and Si substrates

Figure 2(a)-(f) shows the optical response from binary oxides and Si substrates (Table 3) collected under 633 nm (black) and 785 nm (red) excitation laser. MgO (Fig. 2(a)) shows a dominant sharp peak at 1255 cm^-1 under 785 nm laser illumination. When extending the available range beyond 1300 cm^-1 by using the 633 nm laser, the peak at 1513 cm^-1 becomes the dominant one, and the 1255 cm^-1 broadens. TiO₂ (Fig. 2(b)) shows a dominant sharp Raman peak at 600 cm^-1 and comparable spectral features under both excitation wavelengths. Al₂O₃ (Fig. 2(c)) shows no significant Raman response < 1300 cm^-1, while two Raman lines (1381 cm^-1 and 1409 cm^-1) are observed when extending the available range beyond 1300 cm^-1 using the 633 nm laser. For the YSZ (Fig. 2(d)), the relative intensities of the spectral features change for different excitation wavelengths. Silicon (Fig. 2(e)) shows its widely recognizable peak at 520 cm^-1, with no significant alteration of any spectral feature with the addition of an amorphous SiO₂ layer (Fig. 2(f)). Raman peaks position, dominant spectral features, and frequency range free from Raman activity are reported in Table 4 for both 633 nm and 785 nm laser excitation.

 

Fig. 2. (a)-(f) Normalized Raman spectra of MgO, TiO₂, Al₂O₃, Y:ZrO₂, SiO₂|Si, Si under excitation with 633 nm (black) and 785 nm (red).

Download Full Size | PDF

Table 3. Lattice constant and band gap of common binary oxide and Si substrates

View Table | View all tables in this article

Table 4. Raman spectral features and free spectral range of common binary oxide and Si substrates.

View Table | View all tables in this article

3.3 Discussion: selection criteria of substrates for optical studies

In order to compare the relative intensities of spectral features and the free spectral range, we build a 2D heat map (Fig. 3). We investigate spectra for both excitation systems using three equally relevant criteria. First, we remove the background (Fig. 3(a), (b) - 633 nm and 785 nm, respectively) across substrates to identify the position and shape of spectral peaks. Second, we plot the non-normalized Raman spectra (for both 633 and 785 nm system) with a capture time of 2s (Fig. 3(c), (d)) to highlight the relative spectral intensity across substrates in both laser systems. Third, we extract the spectra background across substrates (Supplement 1 (c)) and laser systems when recorded with a 2s capture time (Fig. 3(e), (f)).

 

Fig. 3. 2D Raman maps (a),(b) normalized between [0,1] and baseline corrected (multi-polynomial fit of degree 2), and (c),(d) at uniform exposure time of 2s (intensity values shown in log-scale) with 633 nm and 785 nm lasers. Background intensity captured over 2s exposure time for all substrates using (e) 633 nm (inset compares DSO intensities of processed and raw spectrum) and (f) 785 nm excitation laser. Green-boxes compare Raman from GSO with capture times of 2s in 633 and 785 nm laser systems, respectively.

Download Full Size | PDF

Although the shape and location of the spectral peaks are reported in section 3.1, Fig. 3 allows for a nuanced understanding of the Raman response. Taking MgO as an example, we can immediately identify a dominant peak at 1513 cm^-1 in 633 nm system from Fig. 3(a) and a dominant peak at 1257 cm^-1 for the 785 nm system from Fig. 3(b). Furthermore, identification of free spectral range becomes easy in Fig. 3(a), (b), simply coinciding with the darkest ranges for the substrate of interest. However, viewing the raw spectrum at a fixed capture time (Fig. 3(c), (d)) becomes advantageous to discern relative spectral signal intensities across substrates and appreciate backgrounds strength. In fact, the Raman signal from a thin film will be obscured by the presence of a high background coming from the substrate. Background in Raman spectrum can have different origins, notably either due to photoluminescence or noise sources particularly, shot noise, quantum efficiency, dark current noise and thermal noise [31–33]. This background can manifest itself in various ways across substrates depending on irradiation power. A deep discussion towards discerning the type of background and the method to filter it is discussed in Supplement 1 (c). A strong background can distort the thin-film's Raman peaks, making them broader and less distinct. If Raman spectral features typical of the thin film of interest cover a wide range of wavenumbers, hence requiring an optical characterization extended over a broad range, Fig. 3(c), (d) are the most useful maps for guiding the choice of substrate for thin-film growth.

In Fig. 3(e) and (f), we depict the background intensities for all substrates with 2s exposure time in the 237-1750cm^-1 ranges for 633 nm laser excitation and 5-1340 cm^-1 for 785 nm excitation, respectively. As expected from higher photon energy, the 633 nm excitation yields an overall higher background than the 785 nm excitation across all substrates. It becomes immediately visible that NGO and MgO show the strongest background for the 785 nm system (Fig. 3(f)), with GSO showing the strongest background when extending our detection range beyond 1300 cm^-1 with the 633 system (Fig. 3(e)). To further appreciate the influence of the background on the Raman response from a thin-film over a substrate, we compare the baseline corrected and non-normalized Raman response for DSO using the 633 nm excitation (Fig. 3(e) - inset). Although the overall DSO Raman signal across any wavelength is rather low compared to other substrates, i.e., it appears almost black in Fig. 3(a), (c), it exhibits a background intensity, i.e., minimum intensity of the raw Raman spectrum in the range 237-1750cm^-1 which is approximately 44% of the dominant spectral feature (Fig. 3(c)). A conventional way to mitigate background influence is to use a lower photon energy excitation. By comparing the response of GSO (Fig. 3(c), (d) - green boxes) in the two systems, it is evident that spectral peaks are well defined and visible in 785 nm excitation. This is also supported by the outcomes depicted in Fig. 1(h).

As a result, Fig. 3(a)-(e) provides superior visual aids for materials scientists and device engineers to identify distinctive Raman spectral characteristics and guide the choice of substrate for thin-film growth. To give an example, by looking at Fig. 3(d), a growth engineer interested in Raman characterization (785 nm) may prefer LAO over NGO (both having similar lattice constant), as NGO has a strong background and denser spectral characteristics.

4. Conclusion

Raman spectroscopy is a highly effective method for characterizing thin films. This work aimed to establish the suitability of substrates for optical characterization of thin films by discussing the Raman spectra of the fourteen substrates commonly used for growth of thin films. We identified the spectral features using 633 nm and 785 nm excitation lines and compared the overall intensity of background signals across substrates. According to our findings, LAO and Al₂O₃ have the largest free spectral range for both 633 nm and 785 nm laser excitations, while Al₂O₃ has the lowest background levels. The substrates STO and Nb:STO have the narrowest free spectral range, whereas the highest background levels are recorded for GSO and MgO at 633 nm and NGO and MgO for 785 nm, rendering them unsuitable for optical investigations. As a result, we build a 2D map to be used as a universal and convenient handbook to guide the choice of substrate for thin-film growth compatible with optical characterization.