In this work, a novel chemical composition characterization method of direct-write focused electron beam-induced deposition (FEBID) gold nanostructures is presented. The reliable determination of the chemical content for nanostructures has been challenging due to its limited interaction volume. We hereby propose an alternative technique for measuring the chemical composition of nanostructures with limited volume. By measuring the differences in the optical absorption of the nanostructure due to the differences in the chemical composition with the resonance frequency detuning of a nanomechanical resonator and the assistance of analytical optical modeling, we demonstrate the possibility of characterizing the carbon content in the (FEBID) gold nanostructures. From our characterization method, the post-purification process with water successfully reduced the carbon content from ∼65 at. % to ∼20 at. %. This method presents a new technique for the chemical analysis of nanostructures.
Localized surface plasmon resonance (LSPR) is capable of creating gigantic field enhancements focused within the subwavelength region and has thus enabled a wide variety of applications in the recent decays, such as single-molecule Raman spectroscopy,1,2 biomolecule detection,3–5 high-efficiency solar cells,6 and light-sensitive devices.7 With LSPR, the plasmonic resonance frequency can be tailored by the geometry of the nanostructures. To fabricate well-defined plasmonic nanostructures, various nanofabrication techniques including interference lithography and electron beam lithography have been used.8–10 However, these photoresist-based conventional techniques rely on a photoresist layer of homogeneous thickness and can only be applied on planar surfaces and depend on a complex sequence of multiple process steps. In order to fabricate complex plasmonic structures on arbitrary samples, an alternative flexible direct-write process for plasmonic materials is required. Focused electron beam-induced deposition (FEBID) is such a direct-write approach that allows for a mask-free and resist-free deposition of plasmonic nanostructures.11–17 In FEBID, volatile precursor molecules of the desired material are injected into a scanning electron microscope through a gas injection system (GIS) and are locally deposited by the impinging electron beam.
FEBID direct-write approaches for fabricating gold nanostructures have been presented recently.18–20 It has been demonstrated that complex 3D gold nanostructures deposited by FEBID display plasmonic properties.20 Beside numerous benefits including direct patterning and 3D nanoprinting, the low purity of the noble metal structures deposited by FEBID is a major drawback of this technique. Generally, about 25–30 at. % gold is present in typical as-deposited FEBID gold structures. The rest of the chemical composition is mainly contributed by the residual carbon content from the deposition precursor, with minor measurable contribution from oxygen.21,22 Therefore, various purification methods, such as electron beam exposure,22 annealing,23 laser-assisted FEBID,24 substrate heating,25,26 post-deposition exposure to water,27,28 and oxygen plasma,29,30 have been explored to increase the metal content and reduce the carbon content of the structure recently. By optimizations of the FEBID process with an additional oxidizing gas, it has become possible to directly deposit chemically pure gold,21 which is a prerequisite for obtaining a strong plasmonic resonance. In this work, we use a simple and effective post-deposition purification approach in which water acts as the oxidative gas species, purifying already deposited gold nanostructures.20 The bulk chemical composition of the structure is obtained using scanning electron microscopy (SEM) equipped with energy dispersive spectroscopy (EDX), and the morphology of the structure is measured using atomic force microscopy (AFM).
However, the conventional chemical composition analysis methods such as aforementioned EDX require a certain interaction volume of the sample material with the probing electron or the radiation beam. For a direct-write plasmonic nanostructure, the volume is limited to nanoscale. As a result, in addition to the standard analysis methods, we introduce a new technique based on nanomechanical scanning absorption microscopy (NSAM),31–33 which allows for the direct measurement of the chemical composition of individual plasmonic nanostructures. The NSAM method is based on the frequency detuning of a nanomechanical resonator due to the photothermal effects by plasmonic nanostructures. The optical absorption cross section of the nanostructures can be extracted from the frequency shift of the nanomechanical resonator precisely. Similar characterizations of the optical absorption of nanoparticles have also enabled the displacement detection down to picometers resolution.34 In this study, the change in resonance frequency is directly correlated to the level of impurity within the single plasmonic nanostructures by the theoretical analysis based on effective medium approximation analysis and Mie theory. Hence, NSAM represents a viable alternative to electron energy loss spectroscopy (EELS), Auger electron spectroscopy, and high-angle annular dark-field imaging. Finally, we use NSAM to map and analyze the purified FEBID Au bowtie antenna with various gap sizes to demonstrate the strong LSPR in the purified nanostructures.
II. RESULTS and DISCUSSIONS
A. Optimization of the FEBID purification process
In order to investigate purification strategies and optimize the parameters, first Au nanostructures were deposited using a conventional FEBID procedure as shown in Fig. 1(a). The Au precursor is injected into the vacuum chamber (base pressure of <2 × 10−6 mbar) of a scanning electron microscope (SEM) via a gas injection system (GIS). The precursor is decomposed by secondary effects caused by electrons from the focused e-beam impinging on the surface. Various post-deposition purification strategies have been explored to purify gold nanostructures. Among them, the simplest way to purify gold nanostructures is annealing by extended irradiation with the focused electron beam. In addition, it has been shown that transferring the deposited structures to an environmental SEM and introducing water as an oxidative gas improve the purification significantly. However, this requires an additional, specialized equipment. Here, we test the water-based post-deposition electron beam purification strategy in situ without changing to an environmental SEM but perform purification in the conventional SEM, where also the deposition was conducted.
First, a series of Au rectangles as shown in Fig. 1(b) was deposited by FEBID. After deposition, the structure was exposed to extensive electron beam irradiation (1 nA) at a pressure ∼2.8 × 10−6 mbar for 32 mins. For the 1.2 × 1.2 μm2 area of the Au-deposit, this corresponds to 1.33 μC/μm2. The exposure of the e-beam results in visible morphological changes on the surface. Second, in order to optimize an electron-beam activated purification with reactive species, de-ionized water was used as an oxidative additive. Though the reaction of carbon and water vapor to carbon oxide would happen spontaneously based on the Ellingham diagram, the electron beam provides sufficient stimulus for overcoming the activation energy barrier of the reaction between carbon and H2O. In this case, a freshly deposited structure was purified with the electron beam in the presence of water for 32 min at a pressure of ∼2.0 × 10−4 mbar. After purification, the surface looks smoother compared to e-beam curing alone. Since this process is effective mainly on reducing the carbon content, other minor chemical impurities are regarded to remain constant.
While the morphological differences are apparent in Figs. 1(d) and 1(e), it is interesting to note that the EDX spectra differ even more (see Fig. S1 in the supplementary material); the EDX spectrum of the pristine deposit shows a significant count at the characteristic energy for carbon. The e-beam only purified structure exhibits a slightly lower carbon peak compared to a pristine deposit, which can result from the annealing effect and enhanced diffusion of carbon in gold.22 When water is added, the EDX shows that significant amount of carbon is removed due to the oxidative species formed by the electron beam. The chemical composition of the structures is shown in Table I. A different carbon content of the gold structures should also be reflected by a different plasmonic response of the structures. This was investigated by nanomechanical scanning absorption microscopy (NSAM) using a laser Doppler vibrometer.
|Treatment .||C (at. %) .||Au (at. %) .|
|E-beam + H2O||21||76|
|Treatment .||C (at. %) .||Au (at. %) .|
|E-beam + H2O||21||76|
B. Characterization of carbon content with NSAM and the effective medium modeling of dielectric function
In order to characterize the plasmonic properties of post-purified FEBID gold structures using NSAM, Au nanodisks with different sizes were fabricated on multiple nanomechanical drums. The silicon nitride nanomechanical drums were prepared by the bulk micromachining lithography process. A 50 nm silicon nitride thin film was first deposited with the low-pressure chemical vapor deposition (LPCVD) process on the silicon substrate with ∼150 MPa tensile stress and subsequently released by a KOH etching process of silicon. Gold plasmonic nanostructures were then deposited on the drum with FEBID. Part of the Au nanodisks were purified using the previously introduced water assisted post-deposition purification method. Both post-purified and as-deposited nanodisks were analyzed with NSAM under high vacuum condition below a pressure of 10−4 mbar, as shown in Fig. 2(a). The mechanical resonance frequency of the silicon nitride drum was monitored with a laser-Doppler vibrometer (LDV). When the nanodisks are scanned, the absorption of the laser by the FEBID nanodisks results in a local heating of the drum, causing a thermoelastic detuning of the drum mechanical resonance frequency. This detuning of the resonance frequency is tracked by a phase-locked loop while scanning the drum, as shown in Fig. 2(a). The extend of the absorption of the laser depends on the absorption and plasmonic effects of the material, which—among other factors such as size and geometry—also depends on the purity of the Au nanodisks. A test measurement signal with three as-deposited and three post-purified nanodisks was plotted as in Fig. 2(b). In Fig. 2(b), the post-purified nanodisks induce an almost 10-fold stronger frequency shifts than the as-deposited nanodisks. This indicates a much higher optical absorption for purified nanostructures. The near-Gaussian frequency shift profile from the nanodisks in Fig. 2(b) results from the convolution of the Gaussian laser beam and the absorption of the nanodisk, represented by a point absorber, as discussed in the previous studies.35,36
To investigate the effect of the carbon content on the optical absorption of the FEBID nanostructures systematically, the peak frequency shifts of nanodisks with different sizes were extracted with a one-dimensional Gaussian fit, as plotted in Fig. 2(c). The root-mean-square fitting error was indicated by the error bar. The frequency shift mapping of selected nanodisks, labeled as #1 to #6 in Fig. 2(c), is plotted in Fig. 2(d). Nanodisks #1 to #3 have similar radius with nanodisks #4 to #6, and the only difference should be the purification process. The radius of the nanodisks was measured by SEM, and the thickness was measured by AFM to be around 40 nm for such deposition parameter. The thickness can be reduced by 5–10 nm after the post-purification process due to the removal of the carbon content. However, this thinning has only a minor effect on optical absorption, since the linearly polarized light with a perpendicular incident angle only excites the lateral in-plane modes rather than out-of-plane modes. To avoid plasmon coupling between the nanodisks and the grating effect, all the nanodisks were deposited with at least 5 μm spacing. In general, from both Figs. 2(c) and 2(d), for nanodisks with different radius, the post-purified nanodisks demonstrated higher frequency shift with higher optical absorption, implying an enhanced LSPR for post-purified nanodisks.
To quantify this effect of the carbon content on LSPR, the optical absorption cross sections σabs of each nanodisk can be extracted, respectively, from drum frequency detuning , namely,
with the previously measured responsivity R of the silicon nitride drum of 850 W−1, peak irradiance I of 250 μW/μm2, and is the reference resonance frequency of the silicon nitride drum to be around 52 kHz, measured by the LDV, as shown in Fig. S2 in the supplementary material. The peak irradiance I can be calculated from the measured power P and beam radius r, based on Gaussian power distribution of the beam, such that
The absorption cross sections were extracted from these vibrometer measurements and are summarized in Fig. 2(e). A goal of this study was to establish a theoretical relation between the carbon content and the absorption cross section and fit this theoretical model with the measured absorption cross section extracted from the frequency shift. In this way, we can compare the carbon content in the as-deposited and post-purified nanodisks and can extract the level of carbon content.
First, we consider the theoretical absorption cross section σabs of individual gold nanosphere. This can be calculated based on the Mie theory absorption model, which is the solution of Maxwell equations for particles with sizes similar to the incident wavelengths, and is given by37
with wavelength and polarizability . Since the dipolar resonance behaviors are similar between nanospheres and nanodisks along the axial polarization, this solution can also be used for our following analysis. For a homogeneous nanosphere of pure gold, the polarizability for a dipolar mode is dependent on the dielectric function of the gold material and its surrounding medium , such that38
in which V is the volume of the nanosphere, is the optical wavelength, and for the case of vacuum. From Eq. (3), it can be clearly seen that the polarizability and thus the theoretical absorption cross section σabs of a nanosphere are dependent on the dielectric function of the gold material . For our FEBID nanodisks, the dielectric function material is strongly dependent on the carbon content level of the gold nanodisks.
Second, we need to consider the effect of carbon content on the effective dielectric function of the deposited gold nanodisks. Due to the effect of interband transition in gold starting around the wavelength below 700 nm, modeling the dielectric function with the Drude model by varying the electrical conductivity and damping rate would result in large discrepancy for the 633 nm wavelength used in the present study.39 To overcome this, also to provide a more general model independent of material and excitation wavelength, we apply the effective medium theory,40 which obtain the effective dielectric function of the FEBID gold based on the dielectric function of pure gold,41 pure graphite,42 and the level of impurity. By micro-Raman spectroscopy, the carbon matrix in FEBID deposits was shown to be amorphous or graphite-like so that using the dielectric function pure graphite42 was the closest match to model the carbon contents. For FEBID-Pt-deposition, Poratti et al. also describe the material composition as a granular metal embedded in a carbonaceous matrix, and the effective dielectric function of the FEBID gold with carbon content can be obtained by the Maxwell-Garnett effective-medium expression as in Ref. 40,
in which and are the dielectric functions of pure gold and pure graphite, respectively, and is the volume fraction of the carbon content. For consistency with the EDX data, which is given in atomic fraction, the atomic fraction will be used in the following discussions. As indicated by AFM thickness measurements with our experiments, no significant change in the volume with different carbon contents was observed. Hence, the volume fraction can be translated to mass fraction and subsequently atomic fraction by
where and are the densities of carbon and gold (2.26 and 19.30 g/cm3), respectively. and are the atomic masses of carbon and gold (12.01 and 196.96 g/mol), respectively. The effective dielectric function obtained by the Maxwell-Garnett effective-medium expression is then applied to the Mie theory model in Eq. (3),40 and the absorption cross section of different carbon contents can be calculated, as shown in the model in Fig. 2(e). By fitting this model to different sizes of nanodisks instead of just single one, the individual deviations can be corrected to give a more robust picture, and the model can also be better verified.
As shown in Fig. 2(e), the measured absorption cross sections of the post-purified FEBID nanodisks fit with the Maxwell-Garnett model of ∼20 at. % of carbon content, which is quite consistent with the chemical composition obtained by EDX shown in Table I. However, the measured absorption cross sections of the as-deposited FEBID nanodisks fit with the model of ∼65 at. % of carbon content, which deviates from the EDX value of 44 at. %. This deviation can be the fact that the Maxwell-Garnett effective medium expression assumes a host material with higher chemical composition and an incorporated material with lower chemical composition, which in our case are gold and graphite, respectively. As briefly mentioned, the incorporated material would form nanoscopically heterogenous domains in the matrix material. However, when the content of the carbon approaches the content of the gold, the heterogenous micro-domain picture of the incorporation in the Maxwell-Garnett approximation starts to show some discrepancies. Instead, a more randomly mixed microstructure with no clear differences between the host and incorporated materials proposed by Bruggeman would be more appropriate for modeling highly carbon-contaminated gold structures.40,43,44 As the present study intended to quantify small carbon content in plasmonic Au nanostructures, the Maxwell-Garnett model is clearly more suitable to meet the demands of this study. Numerical simulations can always provide more accurate modeling results; however, this is beyond the scope of this discussion. The emphasis of this work is on the proof of the physical concept. Another interesting observation in Fig. 2(e) is that the measured absorption cross section of larger post-purified nanodisks seems to deviate from the 20 at. % carbon content model to higher carbon content. This can result from the reduced surface to volume ratio for the larger nanodisk, so that the purification process was less effective with the same parameters.
To remove the contribution of the nanostructure geometry on absorption, measured absorption cross sections of as-deposited and post-purified nanodisks both with 100 nm in diameter are plotted in Fig. 2(f). The calculated absorption cross sections for 100 nm nanodisks from the proposed model with different levels of carbon content are also plotted in Fig. 2(f) for comparison. The measured values fit well with the model at ∼65 at. % and ∼20 at. % for the as-deposited and post-purified. As shown in Fig. 2(f), the increased content of carbon would reduce the absorption cross section, implying a reduction in LSPR. Gold, as a low-loss plasmonic material, has a small imaginary part of the dielectric function in the visible regime. The contamination with graphitic carbon increases the imaginary part of the dielectric function and introduces additional damping to the plasmonic resonance. The proposed quantification analysis based on plasmonic absorption proves the reduction of carbon content in gold nanodisks after the purification process aside from the conventional EDX and also further provides a novel technique for the material characterization of nanostructures with a small volume for which a conventional point-probing technique would be impossible.
C. LSPR in purified Au bowtie nanostructures
In order to show the enhanced plasmonic resonance in the post-purified FEBID nanostructures, as well as to show the versatility and potential of our technique for probing the absorption cross section, bowtie nanostructures with different gap sizes were deposited and post-purified, as shown in Figs. 3(a)–3(c). From the finite-difference time-domain (FDTD) simulations shown in Figs. 3(d)–3(f), it can be seen that the electromagnetic field enhancement scales inversely proportional to the gap size, as expected from existing models.45 The same trend can be observed readily from the corresponding NSAM scans as presented in Figs. 3(g)–3(i). Based on the responsivity of R = 850 W−1 and irradiance of I = 250 μW/μm2, the absorption cross section of the bowtie nanostructructures can be extracted from the measured relative peak frequency shifts of 9.8%, 5.1%, and 2.3% in Figs. 3(g)–3(i) to be σabs = 0.46 μm2, 0.24 μm2, and 0.11 μm2, respectively. The extracted absorption cross sections are comparable with the resonant absorption cross sections of an e-beam-evaporated pure gold bowtie nanoantenna,46,47 which is a good sign of the purity of our nanostructures. The deposition of bowtie antenna with high purity and controllable gap size shows the applicability of FEBID for the deposition of complex nanostructures.
This works shows that FEBID as a resistless direct-write method is capable of producing complex nanostructures even on fragile substrates with a high precision of tens of nanometers. We showed that post-deposition purification with H2O can significantly reduce the carbon content of the FEBID nanostructures and hence significantly enhance their plasmonic response. FEBID Au nanostructures of high purity open the door to fast plasmonic prototyping of complex structures on non-standard surfaces. To investigate the carbon residuals in the as-deposited and post-purified nanostructures, we introduced nanomechanical absorption scanning microscopy (NSAM) as a new tool to quantitatively characterize chemical composition of single plasmonic nanostructures, more specifically the carbon content. NSAM combined with Mie theory and effective medium modeling can measure the chemical composition of individual Au nanostructures and allows us to analyze the FEBID purification process. Due to its high single particle sensitivity, NSAM has the potential to be a unique analysis tool, especially for nanometer-sized samples, and has potential applications in many fields and plasmonics in particular.
All FEBID experiments were performed inside a Zeiss Leo 1530VP scanning electron microscope that has a home-built multi-nozzle gas injection system (GIS). The base pressure of the system was ∼2 × 10−6 mbar.To deposit gold, commercially available dimethyl gold (III) trifluoroacetylacetonate precursor was used. The nozzles are cylindrical in shape with an inner diameter of 400 μm. The deposition area was approximately 1 mm away from the center of the nozzle outlet. Generally, during Au deposition, the gold precursor reservoir was heated up to ∼50 °C in order to yield a working pressure of ∼2 × 10−5 mbar. In order to obtain reproducible results, the water was first outgassed in an ultrasonic bath. During the curing process, the precursor reservoir was surrounded by a larger containment of water at room temperature to compensate for evaporation cooling inside the reservoir caused by the exposure to vacuum.
A. Energy dispersive spectroscopy
EDX analysis was performed using a Zeiss Neon 40Esb cross-beam microscope with an Oxford Instruments EDS 7427 detector. The obtained spectra were normalized to the gold peak, and the zero peak was removed.
B. Nanomechanical scanning absorption microscopy
A Polytec laser Doppler vibrometer (MSA−500) was used to characterize the samples. The laser wavelength used in the present study is 633 nm, and the laser featured a spot size of ∼1.5 μm. For this, a silicon nitride drum, which functions as a nanomechanical resonator with low background absorption, was used as the sample. The absorption of silicon nitride in the visible regime has been reported to be below 1%.48 With the silicon nitride resonators in our study, the absorption has been measured to be around 0.5% based on the mechanical frequency shift.32 To obtain the genuine frequency shift resulting only from the nanostructures, this background absorption of silicon nitride is subtracted from the measurements by a reference frequency measured with the laser beam focused on the bare silicon nitride without nanostructures.
C. Fabrication of silicon nitride drum
The samples are fabricated with a bulk micromaching process. A silicon wafer with a thickness of ∼520 μm is coated with 50 nm silicon-rich silicon nitride (SiN) with low pressure chemical vapor deposition (LPCVD). The release of the drum is done by a wet-etching process of the silicon in KOH for ∼8 h. The prestress of the released drum is approximately 150 MPa, which is extracted from the measured mechanical resonance frequency of the drum.
See the supplementary material for the EDX spectra of cured FEBID structures, the resonance peak of the silicon nitride drum measured by laser-Doppler vibrometry, and the dimensions of the fabricated Au nanodisks.
M.-H.C. and M.M.S. contributed equally to this work.
The authors are grateful for the assistance of Sophia Ewert on device fabrication and Center for Micro and Nanostructures (ZMNS) for the cleanroom support. This work has received funding from the European Research Council under the European Union's Horizon 2020 Research and Innovation Program (Grant Agreement No. 716087—PLASMECS).
The data that support the findings of this study are available from the corresponding author upon reasonable request.