- Email: [email protected]
Apertureless near-field scanning Raman microscopy using reflection scattering geometry
Ultramicroscopy 94 (2003) 237–244
Apertureless near-field scanning Raman microscopy using reflection scattering geometry W.X. Sun, Z.X. Shen* Physics Department, National University of Singapore, 2 Science Drive 3, Singapore 117542, Singapore Received 18 February 2002; received in revised form 27 August 2002
Abstract The combination of near-field scanning optical microscopy and Raman spectroscopy provides chemical/structural specific information with nanometer spatial resolution, which are critically important for a wide range of applications, including the study of Si devices, nanodevices, quantum dots, single molecules of biological samples. In this paper, we describe our near-field Raman study using apertureless probes. Our system has two important features, critical to practical applications. (1) The near-field Raman enhancement was achieved by Ag coating of the metal probes, without any preparation of the sample, and (2) while all other apertureless near-field Raman systems were constructed in transmission mode, our system works in the reflection mode, making near-field Raman study a reality for any samples. We have obtained the first 1D Raman mapping of a real Si device with 1 s exposure time. This is a very significant development in near-field scanning Raman microscopy as it is the first demonstration that this technique can be used for imaging purpose because of the short integration time. In addition, the metal tips used in our set-up can be utilized to make simultaneous AFM and electrical mappings such as resistance and capacitance that are critical parameters for device applications. r 2002 Elsevier Science B.V. All rights reserved. PACS: 68.37.Uv; 07.79.Fc; 07.60.j Keywords: Near-field; Spectroscopy; Simulation; Raman imaging
1. Introduction Near-field scanning optical microscope (NSOM) has emerged as an important tool in material characterization and life science due to its ability to convey optical information with nanometer resolution as well as topographic mapping. On the other hand, Raman spectroscopy has been used *Corresponding author. Tel.: +65-6874-6390; fax: +656777-6126. E-mail address: [email protected] (Z.X. Shen).
extensively in chemical and contaminant identification, phase identification, and strain/stress analysis because of its uniqueness in providing chemical and structural specific information. Raman technique is non-contact, non-invasive and does not require specific sample preparation. Similar to other optical microscopic techniques, the spatial resolution of conventional microRaman is governed by theoretical diffraction limit. The typical spatial resolution of the popular micro-Raman spectrometers is about 1 mm: Naturally, the combination of NSOM and Raman
0304-3991/03/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 3 9 9 1 ( 0 2 ) 0 0 3 3 4 - 0
238
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
spectroscopy is a logical step towards chemical and structural identification in the nanometer scale. However, Near-field Scanning Raman Microscopy (NSRM) has been plagued by the poor signal to noise ratio of the near-field Raman spectra. The conventional NSRM is embodied through an aperture NSOM, whereby an optical fiber tip with a small aperture (50–200 nm) is used to deliver laser light and the fiber tip is kept at a close distance (Btens of nanometers) above the sample surface. Raman signal is collected in farfield through either a microscope objective or a lens [1–3]. In this case, the tip aperture (near-field) rather than the laser wavelength defines the laser spot size on the sample. The scattered light (Raman signal) is coupled to a Raman spectrometer where the Raman spectra are recorded at each point on the sample by scanning the fiber tip across the sample surface. An NSRM image is constructed using these spectra. The laser emerging from such a fiber tip is extremely weak (typically 100 nW) due to the low optical throughput of the fiber tip. Although extensive efforts have been devoted to fabricate fiber tips with higher throughput [4–6], the throughput is still very low, especially for tips with an aperture less than 200 nm: In addition, Raman signal is intrinsically weak (typically less than 1 in 107 photons). A typical Raman spectrum takes several minutes to collect using the conventional near-field Raman method, making it prohibitive to construct a Raman image this way. The Raman images (no more than 32 32) reported in Refs. [1,3] took more than 9 h to collect the data. An alternative approach to the fiber tip is to use apertureless metal tips. With a suitable metal tip, Raman signal from the sample close to it can be enhanced by several orders of magnitude. In this case, the Raman enhancement is a result of local field enhancement by the tip, which can be either due to the enhancement of the electromagnetic field in the vicinity of sharply pointed irregularities on tip, the excitation of surface plasma by the laser electromagnetic wave, or increased polarizability of the sample due to the interaction between the tip and sample. The apertureless NSRM has been used to record Raman images in transmission mode [7].
In this paper, we simulate the local field enhancement using the generalized field propagator method [8]. Our results show that the laser intensity close to the metal tip can be enhanced by several orders of magnitude. The enhancement factor of a given metal tip depends on laser polarization, incident angle and wavelength. We have developed an NSRM system by integrating an NSOM and a Raman spectrometer using the apertureless configuration in reflection geometry. With this system, we succeeded in mapping nearfield Raman image on a real silicon device. While the transmission NSRM can only be applied to transparent and very thin samples which are very restrictive for applications, our method can apply to any samples in principle which is a giant step towards real applications of NSRM. The main drawback of our approach is that the unwanted far-field Raman signal is collected together with the near-field signal. The near-field signal has to be comparable to the far-field signal in order to obtain good-quality Raman images. In our experiments, the near-field enhancement factor is estimated to be more than 104 times and a goodquality Raman spectrum can be collected in 1 s integration time or less with modest laser power. Another crucial important advantage of our approach in comparison with other surface-enhanced Raman spectroscopy (SERS) is that the enhancement is achieved using the metal tip rather than by coating the sample with Ag as normally done. This means that our method will be applicable to any samples and no sample preparation is needed, making it an extremely powerful analytical technique for both fundamental research and applications.
2. Instrumentation The NSRM system used in our experiments consists of a standard Nanonics NSOM system and a Renishaw micro-Raman spectrometer. Nanonics NSOM system uses a bent tip (cantilevered tip), which makes the scanning head very compact. The compact design allows the easy integration of the spectrometer and NOSM by putting the scanning head under the microscope
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
objective (Nikon, x50, WD 13:8 mm and NA 0.45) of the Renishaw system, as shown in Fig. 1. An argon ion laser (488 nm line) is focused onto the metal tip, which is made from tungsten wire through electrochemical etching (similar to the method used in STM tip preparation) and then coated with silver through RF sputtering. The interaction between the tip and the laser enhances the local electrical field near the tip, which in turn enhances the Raman signal nearby. The same objective lens collects the Raman signal in the backscattering geometry, which includes both the far-field and near-field components. The Raman signal is then directed into the spectrometer by the coupling optics. The Renishaw spectrometer is driven by the dynamic data exchange technique, which allows third-party software control. The Burleigh controller of the NSOM system is used to control the Renishaw spectrometer as well as running an AFM scan. During an experiment, the tip is fixed relative to the laser spot while the sample runs a raster scan. The Raman spectra obtained point by point are then processed using an Array Basic program developed by us. In Fig. 1, the metal tip is brought to the sample surface by the cantilever. A red (or infrared) laser emitted from laser diode 4 is focused onto the
239
cantilever, and then reflected into the quadrant detector 8 by the cantilever, which acts as a curved mirror in this case. The position of the reflected laser is determined by the output of the quadrant detector. During a scan, the tip–sample distance is controlled by either the contact mode or tapping mode. In the tapping mode, the cantilever is driven by piezo 3, and the vibrational amplitude, which is a measure of atomic force between the tip and sample, is detected by quadrant detector 8. In contact mode, the quadrant detector detects the bending of the cantilever, which is also a measure of the atomic force between the tip and the sample. The laser used to excite the Raman signal is much stronger than the laser used for force sensing. If some of the scattered light (excitation laser) goes into the quadrant detector, the force detection will be severely affected. To avoid this, a band pass (or long pass) filter is placed just before the quadrant detector to block the scattered light. On the other hand, although the laser used for force sensing is weak compared with the Raman excitation laser, it is still much stronger than the Raman signal. It will have deleterious effect on the signal-to-noise ratio of the acquired Raman spectrum if it gets into the spectrometer. Hence a short pass filter (with its cut-off wavelength shorter
Fig. 1. Schematic diagram of apertureless NSOM-Raman in backscattering geometry. 1 – XYZ scanning stage, 2 – tip, 3 – driving piezo, 4 – laser diode, 5 – microscope objective, 6 – laser, 7 – Raman signal, 8 – quadrant detector, 9 – sample, 10 – controller, 11 – spectrometer, 12 – computer. 1,2,3,4 and 8 constitute the scanning head, which is put on the mechanical stage of the microscope. Laser is delivered by microscope objective 5, which is also used to collect Raman signal in this embodiment.
240
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
or equal to that of the force sensing laser) or a notch filter is placed before the entrance of the spectrometer to prevent the laser diode light from entering the spectrograph. The polarization of excitation laser has great effect on near-field enhancement. A half-wave plate is placed in the excitation laser beam and the polarization direction can be rotated by adjusting the half-wave plate. Besides polarization, incident angle of excitation laser is another factor affecting near-field enhancement. To maximize near-field enhancement, the metal tip should not be perpendicular to the sample surface in our backscattering geometry. One reason is that the tip shaft may shadow the tip end, at least to some extent, although the converging angle of the focused laser is greater than the half angle of the tip. Another reason is that the electrical field of the excitation laser is perpendicular to tip axis in this configuration and can be considered as totally s-polarized laser, which is not the preferred polarization in near-field enhancement as shown by our computer simulation. In our experiment, the tip is set at an oblique angle to the sample. We have also used a perpendicular tip and oblique incident laser. Rather than using an oblique objective, which requires a specially designed microscope and an objective with longer working distance, a mirror was glued to the tip mount to reflect the converging laser beam onto the tip. In this configuration, the bottom edge of the mirror is very close to the tip ðo1 mmÞ and it does not require much increase in the working distance of the objective. The existing Nikon objective (NA 0.45, WD 13:8 mm) can still work in this case. Both the tip end and cantilever can be observed in the microscope and there is no difficulty in aligning the laser spot to the tip. Besides polarization, the incident angle also has great effect on Raman enhancement and it must be optimized for maximum enhancement. In our case, this was first calculated by computer simulation and then realized in experiments by trial and error. The metal tip should be sharp and the cantilever must be flexible and have high resonance frequency. Flexible cantilever is more sensitive in force detection while high resonance frequency can suppress noise. Hence the cantilever should be thin
and short. We prepared the tungsten tip through electro-chemical etching, which is similar to but different from STM tip preparation. In STM application, the tip should be sharp and short in order to reduce tip vibrations. In our case, the etched tip should be long enough to form a cantilever, although it should not be too long. With a tip etching kit (Omicron), a cylindrical stainless-steel cathode with an inner diameter of 10 mm; 2–3 M NaOH solution, 3.5–4:5 V etching voltage, a suitable tip can be obtained [9]. The etching conditions affect the tip sharpness and length. After coated with silver, the tip was bent into a cantilevered tip through a home-designed apparatus with two independently adjustable blazes, operating under a microscope.
3. Computer simulation of near-field enhancement We used the generalized field propagator method to compute the electrical field on the sample plane. The details of this method were presented in Ref. [8]. This model considers a scattering system with dielectric function eðr; oÞ ¼ er ðoÞ þ es ðr; oÞ; where er ðoÞ is the dielectric function of the infinite homogeneous reference medium and es ðr0 Þ is the dielectric contrast of the scattering object (metal tip in this case) to the reference medium. When a monochromatic incident wave E0 ðrÞ impinges on the system, the scattered electrical field EðrÞ is a solution of the vectorial wave equation: r r EðrÞ þ k2 er ðoÞEðrÞ þ k2 es ðr; oÞEðrÞ ¼ 0;
ð1Þ
where k is the wave number in vacuum ðk ¼ jkjÞ: Similarly, the incident field E0 ðrÞ satisfies r r E0 ðrÞ þ k2 er ðoÞE0 ðrÞ ¼ 0:
ð2Þ
Let Eq. (1) equal to Eq. (2) and make use of the following identity: r r Gðr; r0 ; oÞ þ k2 er ðoÞGðr; r0 ; oÞ þ k2 es ðr; oÞGðr; r0 ; oÞ ¼ dðr r0 ÞI:
ð3Þ
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
We obtain Z EðrÞ ¼ dr0 ½dðr r0 Þ k2 Gðr; r0 ; oÞ es ðr0 ; oÞ E0 ðr0 Þ;
ð4Þ
where Gðr; r0 ; oÞ is the Green’s tensor associated with the whole (scattering and background) system. Once Gðr; r0 ; oÞ is obtained, electrical field EðrÞ at any point can be calculated using Eq. (4). To obtain Gðr; r0 ; oÞ; we first find the Green’s tensor G0 ðr; r0 ; oÞ associated with the homogenous reference system, which can be expressed analytically as G0 ðr; r0 ; oÞ 1 ikr R 3 þ 3ikr þ kr2 R2 RR ¼ 1 1 kr2 R2 kr2 R4 exp½ikr R ; ð5Þ 4pR where R ¼ jRj ¼ jr r0 j and kr2 ¼ er k2 : Then the Green’s tensor associated with the whole system can be obtained through the discretized Dyson’s equation Gi;j ¼ G0i;j k2
N X
G0i;p esp Gp;j Dp ;
241
The incident beam used in our simulation is a parallel beam of l ¼ 488 nm and intensity of 1. The tip is a conical silver tip of 151 half angle and the size of tip apex l=80 (around 6 nm for 488 nm laser), as shown in Fig. 2. The tip is discretized into small meshes. The closer the mesh is to the tip end, the finer is the mesh. In our simulation, the tip was divided into 1887 meshes. The dimension of the smallest mesh is l=250 ð0:871 nmÞ and that of the biggest mesh is about l=12 ð41:2 nmÞ: The influence of sample on the electrical field is neglected. Figs. 3a and b show the simulation results with p-polarized and s-polarized incident beams, respectively. Obviously, the near-field enhancement factor for p-polarized incident beam is much
ð6Þ
p¼1
where Dp is the volume of the pth mesh and N is the total number of meshes.
Fig. 2. The schematic diagram of the model used in simulation. y is incident angle, k is the wave vector. Tip length is l and tip size is l=80:
Fig. 3. Simulation results with p-polarized (a) and s-polarized (b) incident beams. The incident angle is 451 and the observation plane is l=200 below the tip end.
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
242
greater than that for s-polarized incident beam. Therefore, p-polarization is preferred for near-field enhancement. If we take the full-width at halfmaximum of the peak in Fig. 3a as the spatial resolution of the apertureless NSOM, the spatial resolution is about one and a half times of the tip size, as shown in Fig. 3a. To simulate the spatial resolution in the z-direction, we calculated the near-field enhancement factor for different distances between the tip end and the observation plane. As shown in Fig. 4, the enhancement factor
decays rapidly with the increase of the distance. Hence the near-field signal is confined to a very thin layer of the sample about 7 nm thick. Besides polarization, we have also simulated the effect of incident angle on the enhancement factor, shown in Fig. 5. As the electrical component along the tip axis plays an important role in near-field enhancement, one might expect maximum enhancement when the incident angle tends to p=2: However, our results show that the maximum enhancement does not occur at p=2 and the incident angle at which maximum enhancement occurs changes with the tip geometry.
Dependence of enhancement on distance 14000 12000
4. Experimental results
10000
|E| 2
8000 6000 4000 2000 0 0
10
20
30
40
Distance (nm)
Fig. 4. Distance dependence of near-field enhancement factor. The beam is incident at 451 with p-polarization.
Dependence of Enhancement on incident angle 14000
To observe the near-field enhancement, an experiment on pure silicon crystal was performed. In this experiment, a silicon wafer was placed on the piezo scanning stage of the NSOM system and the silicon Raman peak at 520 cm1 was recorded for two tip positions: (1) the tip ‘‘touching’’ the silicon surface to record the Raman spectrum which includes both the near- and far-field components, and (2) the tip lifted (for about 100 mm) from the surface to record the far-field Raman spectrum, which are shown as spectra a & b in Fig. 6, respectively. The near-field component consists about 35% of the total signal in spectrum a. It should be noted that the near-field Raman signal comes from a small region on the sample, while the far-field signal comes from a much larger
12000
8000
5000
&Far-field)
6000
4500
B(Far-field only)
|E|
Intensity (AU)
5500
2
10000
4000 2000 0
A(Near-field
4000 3500 3000 2500
0
20
40
60
80
Incident Angle (degree) Fig. 5. Angular dependence of near-field enhancement factor. The incident beam is p-polarized. The sampling point is l=200 below the tip end.
490
500
510
520
530
540
550
560
Raman Shift (cm-1) Fig. 6. The silicon Raman spectra with the tip ‘‘touching’’ the sample surface (spectrum a) and the tip withdrawn (spectrum b).
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
243
than 1.5 times of the tip. Hence the optical resolution for the near-field Raman mapping shown below should be less than 50 nm: In the third experiment—the NSRM experiment, a total of 128 Raman spectra were recorded with 1 s integration time across the lines. The Raman intensity, bandwidth and frequency at each point on the device were derived by fitting the spectra with Gaussian profile using the Array Basic program. In Fig. 8, we plot the Raman intensity of Si against the serial number of the Raman spectra that corresponds to different positions on the device shown in Fig. 7b, e.g. number 1 and 128 are the left-most and right-most points, respectively. The two curves a and b shown in Fig. 8 are two Raman mappings of the same Si lines with different tips, showing good reproducibility of this method. As the SiO2 is very thin and its Raman scattering cross section is much smaller than that of Si, the Raman spectrum of SiO2 that consists a predominant at 465 cm1 was not observed in our experiments with 1 s integration time.
region (about 2–3 mm in our instrumental set-up). Assuming the near-field region is about several tens of nanometers, a simple estimation shows the near-field enhancement factor is no less than 104 : In our second experiment, we demonstrate the improvement in spatial resolution using a featured sample—a silicon device which consists of SiO2 lines on Si substrate. Fig. 7a shows schematically the cross section of the silicon device sample. The separation between the SiO2 lines is 300 nm; and the SiO2 lines are 380 nm in width, 350 nm in depth and 30 nm higher than the silicon substrate. Fig. 7b is a topographic AFM image of the sample. By comparing the width of the SiO2 strip in the AFM profile (curve at the bottom of Fig. 7b) and the real width indicated in Fig. 7a, the topographic spatial resolution of our set-up is derived to be 34 nm: The tip size is smaller than 34 nm as the non-linearity of the piezo stage is also included. This is confirmed by our SEM images of the tips [9]. As our simulation result shows, the near-field signal comes from a region not bigger
SiO 2
Silicon (A)
nm
30 20 10 0 0.0
(B)
0.5
1.0
1.5
2.0
2.5
3.0
3.5
X (µm)
Fig. 7. (a) Schematic cross-sectional diagram of the Si device sample. (b) AFM image of the Si device sample. The image was acquired by the NSOM with a metal tip used in this experiment.
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
244
Position (Micron) 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Intensity of Si Peak (AU)
1400
stronger because of near-field enhancement. Hence with a metal tip, the total Raman signal varies with the position on the device.
B 1300
5. Conclusion
1200
A 1100
C 1000 0
20
40
60
80
100
120
Spectrum No.
Fig. 8. Raman mapping of the Si device sample across the Si lines. Curves A and B are generated from two sets of Raman spectra acquired using two different metal tips. Curve C is generated from Raman spectra acquired while the tip is out of the laser spot.
We have also collected Raman mapping of the same Si lines using the micro-Raman setting and the Raman intensity does not show any variation with position. This can be understood easily by considering the fact that far-field Raman signal comes from a region of 2–3 mm; which is much larger than the size of the Si or SiO2 lines. Hence the far-field Raman signal does not show variation while the sample is scanned, and contributes a constant background to the total signal. On the other hand, the near-field Raman signal comes from a small region close to the tip. When the tip sits on the SiO2 line, there is no near-field contribution because the SiO2 is too thick to allow any near-field enhancement of the Si peak. When the tip sits on the Si line, Raman peak becomes
We simulated near-field enhancement in apertureless NOSM configuration and developed an apertureless near-field Raman microscope in reflection geometry by integrating an NSOM and a Raman spectrometer. The apertureless configuration is a breakthrough to the limitation set by the low-optical throughput of metal-coated optical fiber tips, reducing the integration time for Raman spectrum drastically. The reflection scattering geometry makes the system applicable to any samples without preparation. Hence this is a giant step towards the real application of NSRM.
References [1] S. Webster, D.A. Smith, D.N. Batchelder, Vib. Spectrosc. 18 (1998) 51. [2] E.J. Ayars, H.D. Hallen, Appl. Phys. Lett. 76 (2000) 3911. [3] C.L. Jahncke, M.A. Paesler, H.D. Hallen, Appl. Phys. Lett. 67 (1995) 2483. [4] M.N. Islam, X.K. Zhao, A.A. Said, S.S. Mickel, C.F. Vail, Appl. Phys. Lett. 71 (1997) 2886. [5] T. Yatsui, M. Kourogi, M. Ohtsu, Appl. Phys. Lett. 73 (1998) 2090. [6] Y.H. Chuang, K.G. Sun, C.J. Wang, J.Y. Huang, C.L. Pan, Rev. Sci. Instrum. 69 (1998) 437. . [7] R.M. Stockle, Y.D. Suh, V. Deckert, R. Zenobi, Chem. Phys. Lett. 318 (2000) 131. [8] O.J.F. Martin, C. Girard, A. Dereux, Phys. Rev. Lett. 74 (1995) 526. [9] W.X. Sun, Z.X. Shen, Rev. Sci. Instrum. 73 (2002) 2942.
Apertureless near-field scanning Raman microscopy using reflection scattering geometry W.X. Sun, Z.X. Shen* Physics Department, National University of Singapore, 2 Science Drive 3, Singapore 117542, Singapore Received 18 February 2002; received in revised form 27 August 2002
Abstract The combination of near-field scanning optical microscopy and Raman spectroscopy provides chemical/structural specific information with nanometer spatial resolution, which are critically important for a wide range of applications, including the study of Si devices, nanodevices, quantum dots, single molecules of biological samples. In this paper, we describe our near-field Raman study using apertureless probes. Our system has two important features, critical to practical applications. (1) The near-field Raman enhancement was achieved by Ag coating of the metal probes, without any preparation of the sample, and (2) while all other apertureless near-field Raman systems were constructed in transmission mode, our system works in the reflection mode, making near-field Raman study a reality for any samples. We have obtained the first 1D Raman mapping of a real Si device with 1 s exposure time. This is a very significant development in near-field scanning Raman microscopy as it is the first demonstration that this technique can be used for imaging purpose because of the short integration time. In addition, the metal tips used in our set-up can be utilized to make simultaneous AFM and electrical mappings such as resistance and capacitance that are critical parameters for device applications. r 2002 Elsevier Science B.V. All rights reserved. PACS: 68.37.Uv; 07.79.Fc; 07.60.j Keywords: Near-field; Spectroscopy; Simulation; Raman imaging
1. Introduction Near-field scanning optical microscope (NSOM) has emerged as an important tool in material characterization and life science due to its ability to convey optical information with nanometer resolution as well as topographic mapping. On the other hand, Raman spectroscopy has been used *Corresponding author. Tel.: +65-6874-6390; fax: +656777-6126. E-mail address: [email protected] (Z.X. Shen).
extensively in chemical and contaminant identification, phase identification, and strain/stress analysis because of its uniqueness in providing chemical and structural specific information. Raman technique is non-contact, non-invasive and does not require specific sample preparation. Similar to other optical microscopic techniques, the spatial resolution of conventional microRaman is governed by theoretical diffraction limit. The typical spatial resolution of the popular micro-Raman spectrometers is about 1 mm: Naturally, the combination of NSOM and Raman
0304-3991/03/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 3 9 9 1 ( 0 2 ) 0 0 3 3 4 - 0
238
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
spectroscopy is a logical step towards chemical and structural identification in the nanometer scale. However, Near-field Scanning Raman Microscopy (NSRM) has been plagued by the poor signal to noise ratio of the near-field Raman spectra. The conventional NSRM is embodied through an aperture NSOM, whereby an optical fiber tip with a small aperture (50–200 nm) is used to deliver laser light and the fiber tip is kept at a close distance (Btens of nanometers) above the sample surface. Raman signal is collected in farfield through either a microscope objective or a lens [1–3]. In this case, the tip aperture (near-field) rather than the laser wavelength defines the laser spot size on the sample. The scattered light (Raman signal) is coupled to a Raman spectrometer where the Raman spectra are recorded at each point on the sample by scanning the fiber tip across the sample surface. An NSRM image is constructed using these spectra. The laser emerging from such a fiber tip is extremely weak (typically 100 nW) due to the low optical throughput of the fiber tip. Although extensive efforts have been devoted to fabricate fiber tips with higher throughput [4–6], the throughput is still very low, especially for tips with an aperture less than 200 nm: In addition, Raman signal is intrinsically weak (typically less than 1 in 107 photons). A typical Raman spectrum takes several minutes to collect using the conventional near-field Raman method, making it prohibitive to construct a Raman image this way. The Raman images (no more than 32 32) reported in Refs. [1,3] took more than 9 h to collect the data. An alternative approach to the fiber tip is to use apertureless metal tips. With a suitable metal tip, Raman signal from the sample close to it can be enhanced by several orders of magnitude. In this case, the Raman enhancement is a result of local field enhancement by the tip, which can be either due to the enhancement of the electromagnetic field in the vicinity of sharply pointed irregularities on tip, the excitation of surface plasma by the laser electromagnetic wave, or increased polarizability of the sample due to the interaction between the tip and sample. The apertureless NSRM has been used to record Raman images in transmission mode [7].
In this paper, we simulate the local field enhancement using the generalized field propagator method [8]. Our results show that the laser intensity close to the metal tip can be enhanced by several orders of magnitude. The enhancement factor of a given metal tip depends on laser polarization, incident angle and wavelength. We have developed an NSRM system by integrating an NSOM and a Raman spectrometer using the apertureless configuration in reflection geometry. With this system, we succeeded in mapping nearfield Raman image on a real silicon device. While the transmission NSRM can only be applied to transparent and very thin samples which are very restrictive for applications, our method can apply to any samples in principle which is a giant step towards real applications of NSRM. The main drawback of our approach is that the unwanted far-field Raman signal is collected together with the near-field signal. The near-field signal has to be comparable to the far-field signal in order to obtain good-quality Raman images. In our experiments, the near-field enhancement factor is estimated to be more than 104 times and a goodquality Raman spectrum can be collected in 1 s integration time or less with modest laser power. Another crucial important advantage of our approach in comparison with other surface-enhanced Raman spectroscopy (SERS) is that the enhancement is achieved using the metal tip rather than by coating the sample with Ag as normally done. This means that our method will be applicable to any samples and no sample preparation is needed, making it an extremely powerful analytical technique for both fundamental research and applications.
2. Instrumentation The NSRM system used in our experiments consists of a standard Nanonics NSOM system and a Renishaw micro-Raman spectrometer. Nanonics NSOM system uses a bent tip (cantilevered tip), which makes the scanning head very compact. The compact design allows the easy integration of the spectrometer and NOSM by putting the scanning head under the microscope
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
objective (Nikon, x50, WD 13:8 mm and NA 0.45) of the Renishaw system, as shown in Fig. 1. An argon ion laser (488 nm line) is focused onto the metal tip, which is made from tungsten wire through electrochemical etching (similar to the method used in STM tip preparation) and then coated with silver through RF sputtering. The interaction between the tip and the laser enhances the local electrical field near the tip, which in turn enhances the Raman signal nearby. The same objective lens collects the Raman signal in the backscattering geometry, which includes both the far-field and near-field components. The Raman signal is then directed into the spectrometer by the coupling optics. The Renishaw spectrometer is driven by the dynamic data exchange technique, which allows third-party software control. The Burleigh controller of the NSOM system is used to control the Renishaw spectrometer as well as running an AFM scan. During an experiment, the tip is fixed relative to the laser spot while the sample runs a raster scan. The Raman spectra obtained point by point are then processed using an Array Basic program developed by us. In Fig. 1, the metal tip is brought to the sample surface by the cantilever. A red (or infrared) laser emitted from laser diode 4 is focused onto the
239
cantilever, and then reflected into the quadrant detector 8 by the cantilever, which acts as a curved mirror in this case. The position of the reflected laser is determined by the output of the quadrant detector. During a scan, the tip–sample distance is controlled by either the contact mode or tapping mode. In the tapping mode, the cantilever is driven by piezo 3, and the vibrational amplitude, which is a measure of atomic force between the tip and sample, is detected by quadrant detector 8. In contact mode, the quadrant detector detects the bending of the cantilever, which is also a measure of the atomic force between the tip and the sample. The laser used to excite the Raman signal is much stronger than the laser used for force sensing. If some of the scattered light (excitation laser) goes into the quadrant detector, the force detection will be severely affected. To avoid this, a band pass (or long pass) filter is placed just before the quadrant detector to block the scattered light. On the other hand, although the laser used for force sensing is weak compared with the Raman excitation laser, it is still much stronger than the Raman signal. It will have deleterious effect on the signal-to-noise ratio of the acquired Raman spectrum if it gets into the spectrometer. Hence a short pass filter (with its cut-off wavelength shorter
Fig. 1. Schematic diagram of apertureless NSOM-Raman in backscattering geometry. 1 – XYZ scanning stage, 2 – tip, 3 – driving piezo, 4 – laser diode, 5 – microscope objective, 6 – laser, 7 – Raman signal, 8 – quadrant detector, 9 – sample, 10 – controller, 11 – spectrometer, 12 – computer. 1,2,3,4 and 8 constitute the scanning head, which is put on the mechanical stage of the microscope. Laser is delivered by microscope objective 5, which is also used to collect Raman signal in this embodiment.
240
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
or equal to that of the force sensing laser) or a notch filter is placed before the entrance of the spectrometer to prevent the laser diode light from entering the spectrograph. The polarization of excitation laser has great effect on near-field enhancement. A half-wave plate is placed in the excitation laser beam and the polarization direction can be rotated by adjusting the half-wave plate. Besides polarization, incident angle of excitation laser is another factor affecting near-field enhancement. To maximize near-field enhancement, the metal tip should not be perpendicular to the sample surface in our backscattering geometry. One reason is that the tip shaft may shadow the tip end, at least to some extent, although the converging angle of the focused laser is greater than the half angle of the tip. Another reason is that the electrical field of the excitation laser is perpendicular to tip axis in this configuration and can be considered as totally s-polarized laser, which is not the preferred polarization in near-field enhancement as shown by our computer simulation. In our experiment, the tip is set at an oblique angle to the sample. We have also used a perpendicular tip and oblique incident laser. Rather than using an oblique objective, which requires a specially designed microscope and an objective with longer working distance, a mirror was glued to the tip mount to reflect the converging laser beam onto the tip. In this configuration, the bottom edge of the mirror is very close to the tip ðo1 mmÞ and it does not require much increase in the working distance of the objective. The existing Nikon objective (NA 0.45, WD 13:8 mm) can still work in this case. Both the tip end and cantilever can be observed in the microscope and there is no difficulty in aligning the laser spot to the tip. Besides polarization, the incident angle also has great effect on Raman enhancement and it must be optimized for maximum enhancement. In our case, this was first calculated by computer simulation and then realized in experiments by trial and error. The metal tip should be sharp and the cantilever must be flexible and have high resonance frequency. Flexible cantilever is more sensitive in force detection while high resonance frequency can suppress noise. Hence the cantilever should be thin
and short. We prepared the tungsten tip through electro-chemical etching, which is similar to but different from STM tip preparation. In STM application, the tip should be sharp and short in order to reduce tip vibrations. In our case, the etched tip should be long enough to form a cantilever, although it should not be too long. With a tip etching kit (Omicron), a cylindrical stainless-steel cathode with an inner diameter of 10 mm; 2–3 M NaOH solution, 3.5–4:5 V etching voltage, a suitable tip can be obtained [9]. The etching conditions affect the tip sharpness and length. After coated with silver, the tip was bent into a cantilevered tip through a home-designed apparatus with two independently adjustable blazes, operating under a microscope.
3. Computer simulation of near-field enhancement We used the generalized field propagator method to compute the electrical field on the sample plane. The details of this method were presented in Ref. [8]. This model considers a scattering system with dielectric function eðr; oÞ ¼ er ðoÞ þ es ðr; oÞ; where er ðoÞ is the dielectric function of the infinite homogeneous reference medium and es ðr0 Þ is the dielectric contrast of the scattering object (metal tip in this case) to the reference medium. When a monochromatic incident wave E0 ðrÞ impinges on the system, the scattered electrical field EðrÞ is a solution of the vectorial wave equation: r r EðrÞ þ k2 er ðoÞEðrÞ þ k2 es ðr; oÞEðrÞ ¼ 0;
ð1Þ
where k is the wave number in vacuum ðk ¼ jkjÞ: Similarly, the incident field E0 ðrÞ satisfies r r E0 ðrÞ þ k2 er ðoÞE0 ðrÞ ¼ 0:
ð2Þ
Let Eq. (1) equal to Eq. (2) and make use of the following identity: r r Gðr; r0 ; oÞ þ k2 er ðoÞGðr; r0 ; oÞ þ k2 es ðr; oÞGðr; r0 ; oÞ ¼ dðr r0 ÞI:
ð3Þ
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
We obtain Z EðrÞ ¼ dr0 ½dðr r0 Þ k2 Gðr; r0 ; oÞ es ðr0 ; oÞ E0 ðr0 Þ;
ð4Þ
where Gðr; r0 ; oÞ is the Green’s tensor associated with the whole (scattering and background) system. Once Gðr; r0 ; oÞ is obtained, electrical field EðrÞ at any point can be calculated using Eq. (4). To obtain Gðr; r0 ; oÞ; we first find the Green’s tensor G0 ðr; r0 ; oÞ associated with the homogenous reference system, which can be expressed analytically as G0 ðr; r0 ; oÞ 1 ikr R 3 þ 3ikr þ kr2 R2 RR ¼ 1 1 kr2 R2 kr2 R4 exp½ikr R ; ð5Þ 4pR where R ¼ jRj ¼ jr r0 j and kr2 ¼ er k2 : Then the Green’s tensor associated with the whole system can be obtained through the discretized Dyson’s equation Gi;j ¼ G0i;j k2
N X
G0i;p esp Gp;j Dp ;
241
The incident beam used in our simulation is a parallel beam of l ¼ 488 nm and intensity of 1. The tip is a conical silver tip of 151 half angle and the size of tip apex l=80 (around 6 nm for 488 nm laser), as shown in Fig. 2. The tip is discretized into small meshes. The closer the mesh is to the tip end, the finer is the mesh. In our simulation, the tip was divided into 1887 meshes. The dimension of the smallest mesh is l=250 ð0:871 nmÞ and that of the biggest mesh is about l=12 ð41:2 nmÞ: The influence of sample on the electrical field is neglected. Figs. 3a and b show the simulation results with p-polarized and s-polarized incident beams, respectively. Obviously, the near-field enhancement factor for p-polarized incident beam is much
ð6Þ
p¼1
where Dp is the volume of the pth mesh and N is the total number of meshes.
Fig. 2. The schematic diagram of the model used in simulation. y is incident angle, k is the wave vector. Tip length is l and tip size is l=80:
Fig. 3. Simulation results with p-polarized (a) and s-polarized (b) incident beams. The incident angle is 451 and the observation plane is l=200 below the tip end.
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
242
greater than that for s-polarized incident beam. Therefore, p-polarization is preferred for near-field enhancement. If we take the full-width at halfmaximum of the peak in Fig. 3a as the spatial resolution of the apertureless NSOM, the spatial resolution is about one and a half times of the tip size, as shown in Fig. 3a. To simulate the spatial resolution in the z-direction, we calculated the near-field enhancement factor for different distances between the tip end and the observation plane. As shown in Fig. 4, the enhancement factor
decays rapidly with the increase of the distance. Hence the near-field signal is confined to a very thin layer of the sample about 7 nm thick. Besides polarization, we have also simulated the effect of incident angle on the enhancement factor, shown in Fig. 5. As the electrical component along the tip axis plays an important role in near-field enhancement, one might expect maximum enhancement when the incident angle tends to p=2: However, our results show that the maximum enhancement does not occur at p=2 and the incident angle at which maximum enhancement occurs changes with the tip geometry.
Dependence of enhancement on distance 14000 12000
4. Experimental results
10000
|E| 2
8000 6000 4000 2000 0 0
10
20
30
40
Distance (nm)
Fig. 4. Distance dependence of near-field enhancement factor. The beam is incident at 451 with p-polarization.
Dependence of Enhancement on incident angle 14000
To observe the near-field enhancement, an experiment on pure silicon crystal was performed. In this experiment, a silicon wafer was placed on the piezo scanning stage of the NSOM system and the silicon Raman peak at 520 cm1 was recorded for two tip positions: (1) the tip ‘‘touching’’ the silicon surface to record the Raman spectrum which includes both the near- and far-field components, and (2) the tip lifted (for about 100 mm) from the surface to record the far-field Raman spectrum, which are shown as spectra a & b in Fig. 6, respectively. The near-field component consists about 35% of the total signal in spectrum a. It should be noted that the near-field Raman signal comes from a small region on the sample, while the far-field signal comes from a much larger
12000
8000
5000
&Far-field)
6000
4500
B(Far-field only)
|E|
Intensity (AU)
5500
2
10000
4000 2000 0
A(Near-field
4000 3500 3000 2500
0
20
40
60
80
Incident Angle (degree) Fig. 5. Angular dependence of near-field enhancement factor. The incident beam is p-polarized. The sampling point is l=200 below the tip end.
490
500
510
520
530
540
550
560
Raman Shift (cm-1) Fig. 6. The silicon Raman spectra with the tip ‘‘touching’’ the sample surface (spectrum a) and the tip withdrawn (spectrum b).
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
243
than 1.5 times of the tip. Hence the optical resolution for the near-field Raman mapping shown below should be less than 50 nm: In the third experiment—the NSRM experiment, a total of 128 Raman spectra were recorded with 1 s integration time across the lines. The Raman intensity, bandwidth and frequency at each point on the device were derived by fitting the spectra with Gaussian profile using the Array Basic program. In Fig. 8, we plot the Raman intensity of Si against the serial number of the Raman spectra that corresponds to different positions on the device shown in Fig. 7b, e.g. number 1 and 128 are the left-most and right-most points, respectively. The two curves a and b shown in Fig. 8 are two Raman mappings of the same Si lines with different tips, showing good reproducibility of this method. As the SiO2 is very thin and its Raman scattering cross section is much smaller than that of Si, the Raman spectrum of SiO2 that consists a predominant at 465 cm1 was not observed in our experiments with 1 s integration time.
region (about 2–3 mm in our instrumental set-up). Assuming the near-field region is about several tens of nanometers, a simple estimation shows the near-field enhancement factor is no less than 104 : In our second experiment, we demonstrate the improvement in spatial resolution using a featured sample—a silicon device which consists of SiO2 lines on Si substrate. Fig. 7a shows schematically the cross section of the silicon device sample. The separation between the SiO2 lines is 300 nm; and the SiO2 lines are 380 nm in width, 350 nm in depth and 30 nm higher than the silicon substrate. Fig. 7b is a topographic AFM image of the sample. By comparing the width of the SiO2 strip in the AFM profile (curve at the bottom of Fig. 7b) and the real width indicated in Fig. 7a, the topographic spatial resolution of our set-up is derived to be 34 nm: The tip size is smaller than 34 nm as the non-linearity of the piezo stage is also included. This is confirmed by our SEM images of the tips [9]. As our simulation result shows, the near-field signal comes from a region not bigger
SiO 2
Silicon (A)
nm
30 20 10 0 0.0
(B)
0.5
1.0
1.5
2.0
2.5
3.0
3.5
X (µm)
Fig. 7. (a) Schematic cross-sectional diagram of the Si device sample. (b) AFM image of the Si device sample. The image was acquired by the NSOM with a metal tip used in this experiment.
W.X. Sun, Z.X. Shen / Ultramicroscopy 94 (2003) 237–244
244
Position (Micron) 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Intensity of Si Peak (AU)
1400
stronger because of near-field enhancement. Hence with a metal tip, the total Raman signal varies with the position on the device.
B 1300
5. Conclusion
1200
A 1100
C 1000 0
20
40
60
80
100
120
Spectrum No.
Fig. 8. Raman mapping of the Si device sample across the Si lines. Curves A and B are generated from two sets of Raman spectra acquired using two different metal tips. Curve C is generated from Raman spectra acquired while the tip is out of the laser spot.
We have also collected Raman mapping of the same Si lines using the micro-Raman setting and the Raman intensity does not show any variation with position. This can be understood easily by considering the fact that far-field Raman signal comes from a region of 2–3 mm; which is much larger than the size of the Si or SiO2 lines. Hence the far-field Raman signal does not show variation while the sample is scanned, and contributes a constant background to the total signal. On the other hand, the near-field Raman signal comes from a small region close to the tip. When the tip sits on the SiO2 line, there is no near-field contribution because the SiO2 is too thick to allow any near-field enhancement of the Si peak. When the tip sits on the Si line, Raman peak becomes
We simulated near-field enhancement in apertureless NOSM configuration and developed an apertureless near-field Raman microscope in reflection geometry by integrating an NSOM and a Raman spectrometer. The apertureless configuration is a breakthrough to the limitation set by the low-optical throughput of metal-coated optical fiber tips, reducing the integration time for Raman spectrum drastically. The reflection scattering geometry makes the system applicable to any samples without preparation. Hence this is a giant step towards the real application of NSRM.
References [1] S. Webster, D.A. Smith, D.N. Batchelder, Vib. Spectrosc. 18 (1998) 51. [2] E.J. Ayars, H.D. Hallen, Appl. Phys. Lett. 76 (2000) 3911. [3] C.L. Jahncke, M.A. Paesler, H.D. Hallen, Appl. Phys. Lett. 67 (1995) 2483. [4] M.N. Islam, X.K. Zhao, A.A. Said, S.S. Mickel, C.F. Vail, Appl. Phys. Lett. 71 (1997) 2886. [5] T. Yatsui, M. Kourogi, M. Ohtsu, Appl. Phys. Lett. 73 (1998) 2090. [6] Y.H. Chuang, K.G. Sun, C.J. Wang, J.Y. Huang, C.L. Pan, Rev. Sci. Instrum. 69 (1998) 437. . [7] R.M. Stockle, Y.D. Suh, V. Deckert, R. Zenobi, Chem. Phys. Lett. 318 (2000) 131. [8] O.J.F. Martin, C. Girard, A. Dereux, Phys. Rev. Lett. 74 (1995) 526. [9] W.X. Sun, Z.X. Shen, Rev. Sci. Instrum. 73 (2002) 2942.