Coherent anti-Stokes Raman Scattering (CARS) microscopy is an advanced vibrational imaging technique [1]. CARS is a third-order nonlinear process where a pump beam (ω_p) and a Stokes beam (ω_s<ω_p) interact through χ⁽³⁾ of a sample to generate an anti-Stokes beam at a blue-shifted frequency ω_as = 2ω_p-ω_s. The CARS signal is significantly enhanced when ω_p-ω_s is tuned to the frequency of a Raman band, thus providing vibrational contrast. CARS microscopy has been used for the characterization of a wide range of biological systems [2–5] and materials [6,7]. However, like most other optical imaging techniques, its spatial resolution is limited by diffraction. As a nonlinear technique, tightly focused laser beams are usually used in CARS microscopy and the sample is only excited at the laser foci. Therefore the resolution is basically determined by the size of the excitation volume, which is related to the size of the focal spot. But care must be taken to distinguish the meaning of different terms. The size of the excitation volume is smaller than the size of the focal spot of a single laser beam in CARS microscopy because CARS polarization is proportional to the squared intensity of the pump field multiplied by the intensity of the Stokes field. Also, the size of the excitation volume is not the same as resolution. The excitation volume is defined using the point spread function of the imaging system, which is experimentally measured as the width of the image spot produced by a point-like object. The resolution is defined as the minimum separation between two point-like objects that can still be resolved in the images. For laser-scanning CARS microscopy with a pump wavelength λ_p, the lateral FWHM of the excitation volume can reach below 0.4λ_p, but the lateral resolution is about λ_p/2, slightly bigger than the former [8].

In the recent years, however, various techniques have been demonstrated to achieve super-resolution in CARS microscopy, such as depleting CARS radiation at the periphery of the focal spot [9,10], structured illumination [11,12], and near-field scanning with an aperture [13]. In particular, a spatial resolution of ~60 nm (or ~λ_p/12) was achieved with tip-Enhanced broadband CARS (TE-BB-CARS) microscopy [6,14]. And most recently, a lateral resolution of ~0.36 λ_p was achieved with switching laser mode (SLAM) CARS microscopy [15] which extracts the intensity difference between images obtained with Gaussian and donut-shaped laser beams. All these super-resolution methods expanded the horizon of CARS microscopy, but on the other hand they usually require extensive modifications of the conventional CARS microscope. In this paper, we demonstrate far-field super-resolution CARS microscopy by using photonic nanojets generated by microspheres. This method requires no modification of the optical setup, and could be very useful for the vibrational imaging of films or surfaces.

It was reported recently that microspheres could be a relatively simple way of achieving far-field super-resolution in optical microscopy [16–19]. A white light nanoscope using fused silica microspheres was demonstrated to have 50 nm resolution [16]. The microspheres functioned like far-field super lenses, but due to their small size, the magnification factor had to be explained by wave optics instead of geometric optics. A related phenomenon is the generation of a sub-diffraction photonic nanojet on the shadow side of the microsphere when it is illuminated by a laser beam [20]. The existence of such a nanojet provides the possibility to break the diffraction limit in nonlinear optical microscopy by greatly reducing the size of the excitation volume, and the signal generated in the excitation volume can still be collected in the far-field. Based on this principle, super-resolution CARS microscopy using microspheres has been realized and reported recently [21]. Here we report our independent work on the same method. However, our work has two main differences from the previous work [21]. First, in the previous work, the resolved sub-diffraction lines on a grating sample had a minimum width of 280 nm, while in our work we resolved lines on a DVD sample with a smaller minimum width of 100 nm. Second, the theoretical study in the previous work was focused on explaining the contrast enhancement, while in our work we quantified the lateral magnification provided by the microspheres under different conditions and our simulation was focused on explaining the observed behavior. Therefore, our work provided information which was not available in the previous work, and together they can give a more complete picture of this super-resolution imaging technique.

According to Mie theory, the condition of nanojet formation is linked to the refractive index (n) of the sphere and the size parameter defined as S = πD/λ [16,20,22], where D is the sphere diameter. For SiO₂ spheres (n = 1.46), the photonic nanojet occurs only for S< 70. In our CARS setup, the pump and Stokes laser wavelength were 796 nm and 1028 nm, respectively. And for them nanojets can be generated for spheres with D<18 μm and D <23 μm, respectively. The SiO₂ spheres we used had 1-6 μm diameters, which should generate nanojets for both lasers. The CARS signal should then be generated in the overlapped region of the pump nanojet and Stokes nanojet.

The imaging target with sub-diffraction features was a commercial Blu-ray disc (Verbatim BD-R, LTH type, 25 GB). The disc consists of several layers in order: a 100 µm thick transparent protective (PT) layer, a 20 nm thick recording (RC) layer, a 50 nm thick metallic reflective (RF) layer, and a 1.1 mm thick polycarbonate substrate (SB) layer [23,24]. We cut the disc into small pieces, and peeled off the PT layer to expose the RC layer. The RC and RF layers had periodic patterns as shown in Fig. 1(a). The surface profile of the RC layer measured by atomic force microscopy (AFM) [Fig. 1(b)] showed 200 nm wide stripes separated by 100 nm wide and 20 nm deep grooves. These were the sub-diffraction features we aimed to image. Similar AFM images of Blu-ray discs could be found in the literature [25]. The RC layer was made of organic azo-dye in LTH type discs which had an index of 1.6. Due to the existence of the RF layer, the optical imaging of RC layer had to use reflection mode.

 

Fig. 1 (a) Detailed sketch of the microsphere imaging region: 1.1 mm-thick-polycarbonate substrate (SB) layer, 20 nm-thick reflective (RF) layer, 20 nm-thick recording (RC) layer, photonic nanojet (PNJ), Microsphere (MS), (b) Atomic force microscopy (AFM) image of the RC layer shows 200 nm width stripes separated by 20 nm deep and 100 nm width groove. The height-color bar is on the right side of the image. (c) An image acquired with a white light microscope with a 100X 0.9-NA objective through 3-µm SiO2 spheres on the RC layer, (d) Scanning electron microscopy (SEM) image of a few microspheres on the disc sample. (e) Schematic of the epi-CARS microscope. MO-W: water immersion objective; DM: 650 nm short-pass dichroic mirror; BPF: 650/60 nm band-pass filter; FL: focusing lens; PMT: Photomultiplier tube.

Download Full Size | PDF

The SiO₂ spheres with diameters in the range 1-10 μm suspended in ethanol-water solution were purchased from EPRUI Nanoparticles & Microspheres Co. Ltd., Shanghai, China. We needed to place the spheres on the RC layer, but couldn’t drop the solution on it directly because ethanol dissolves azo-dye. So first we dropped the sphere suspension on a No. 1.5 cover glass, and left it to dry. Then a piece of the disc was placed on the spheres with the RC layer facing down. Due to unevenness there was a varying gap between the cover glass and the disc, but by pressing the disc down some spheres would attach to it. The biggest spheres we found to be attached were 6 μm ones. This cover glass-sphere-disc system was the sample for optical imaging. Multiple samples prepared in the same way were imaged.

We first used a white light microscope with a 100X, 0.9 numerical aperture (NA) objective to image the sample. An image with 3-μm SiO₂ spheres is shown in Fig. 1(c). We can see that without the spheres the microscope couldn’t resolve the sub-diffraction features on the RC layer, because the Rayleigh resolution – 0.61(λ/NA) – was 407 nm at the peak wavelength of 600 nm for a halogen lamp. But the microspheres provided a lateral magnification which allowed the parallel features to be resolved in their viewing window. The image was similar to those in previous works [17]. This confirmed that the spheres were contacting the RC layer. Scanning electron microscopy (SEM) images of disc samples with microspheres on top were also taken, and one example is shown in Fig. 1(d). Because the disc was not conductive, the contrast of SEM images was always low. But nonetheless we could see the spheres on the disc.

Before CARS imaging, we measured the spontaneous Raman spectra of the RC and RF layers (data not shown). The RC layer showed a resonant peak at 2745 cm⁻¹ which was broad and extended from 2400 cm⁻¹ to 3000 cm⁻¹. This peak is associated with O-H stretching vibration in aryl sulphonic acid groups in the blue azo dye [26]. The RF layer did not show any obvious peak in this region. Because the Stokes laser in our setup had a fixed wavelength of 1028 nm and the epi dichroic mirror in Fig. 1(e) had a cutoff at 650 nm, we couldn’t use the 2745 cm⁻¹ Raman shift. Instead we chose the Raman shift of 2840 cm⁻¹ for CARS imaging, for which the CARS wavelength was 649 nm. Considering that the pump and Stokes lasers had bandwidths of 123 cm⁻¹ and 67 cm⁻¹, respectively, the CARS signal from the RC layer should be partially resonant.

A typical epi-CARS (E-CARS) microscope [Fig. 1(e)] was used. The laser sources were Orpheus-Twin OPAs pumped by Pharos-9W femtosecond oscillator/amplifier (Light Conversion Ltd., Lithuania). The Pharos-9W laser produced 220 fs pulses at 1028 nm, 1 MHz repetition rate, and 9.5 W power. A small part of it (~700 mW) was split off as the Stokes laser for CARS. The twin OPAs can generate 120 fs pulses in 630-2600 nm range with more than 100 mW average power at 1 MHz repetition rate. One of the OPA outputs at 796 nm was the pump laser for CARS, resulting in a Raman shift (ω_p-ω_s) centered at 2840 cm⁻¹. The laser beams were both horizontally polarized. They were collinearly combined and sent into an inverted laser-scanning microscope (Nikon Ti-Eclipse), and were focused into the sample by a 60X 1.0-NA water objective (MO-W). The average pump and Stokes laser power before the objective were 0.3 mW each. The CARS signal was collected by the same objective, separated from the laser beams by a 650 nm short-pass dichroic mirror (DM), filtered with a 650/60 nm band-pass filter from Semrock Inc. (BPF), and then detected by a photomultiplier tube (PMT). The images were acquired at 1 frame/sec or 2 μs pixel dwell time, and averaged over 8 frames. The scanning area was 40 μm X 40 μm consisting of 512X512 pixels. A smaller scanning area would lead to photodamage to the sample over time. The same PMT gain was used in all experiments. At these settings, the diffraction-limited resolution of the E-CARS microscope was about 400 nm.

The sample was placed on the microscope with the RC layer facing down as in Fig. 1(e). For places with microspheres, nanojets could be formed at the RC layer upon laser illumination as drawn in Fig. 1(a). When the laser beams were scanned, it would be the nanojets that actually scanned the RC layer. The E-CARS signal should be mainly generated inside the RC layer, and reflected by the RF layer towards the epi-detector. In some samples, the RC layer was peeled off in some regions but the RF layer still existed, which could be confirmed by AFM. The E-CARS signal from the sample region without the RC layer was weaker by a factor of 17. Therefore, the signal generated by the RF layer was essentially negligible compared with that generated by the RC layer in our images. Since the RC layer was only 20 nm thick, the CARS signal was expected to be sensitive to the z position of the laser focal plane. We took image series at different focal plane positions by moving the objective in the z-axis as indicated in Fig. 1(e). The z-scan range was ~15 μm with a step size of 1 μm.

We generally observed that in the z-scan image series, for the part of the sample without microspheres, the strongest E-CARS signal happened only when the laser focal plane was right on the RC layer (taken to be the z = 0 plane), and the signal became much weaker when it moved outside or inside the disc, as expected. For the part of the sample covered by microspheres, we found that the E-CARS signal was strong within a certain z range when the focal plane was on or inside the disc, and this range was different for microspheres of different sizes. Outside this range the CARS contrast would fade. It shows that strong nanojets could only be formed when the microsphere was located within a certain region at the incident side of the laser foci. This observation was consistent with the simulation results derived from Lorentz-Mie theory in Ref [27] (especially the movie which can be accessed through a link in Fig. 4 captions in Ref [27]): to form a strong 3D-confined nanojet, it’s the best to have a tightly focused incident field and some high angular incident components passing aside the sphere. If the sphere was too close to the focal center so that no angular component passes aside, the nanojet would be mostly inside the sphere and couldn’t excite the sample. If the sphere was too far away from the focal center so that it only interacted with a small portion of angular components, the nanojet intensity would become low and the CARS signal would fade due to its cubic intensity dependence.

Selected images obtained on three different samples are shown in Fig. 2. Sample #1, #2, #3 have spheres with diameters in the range 1-4 μm, 1-5 μm, and 1-6 μm, respectively. Figures 2(a) and 2(d) were acquired when the laser focal plane was on the RC layer, and hence strong CARS signal was observed everywhere, while in the other images with a non-zero z, CARS signal was always weak outside the microspheres as expected. It was also obvious that outside the microspheres, E-CARS could not resolve the sub-diffraction patterns on the RC layer (100-nm grooves and 200-nm stripes) due the diffraction-limited resolution. On the other hand, microspheres enabled us to resolve the sub-diffraction patterns in their viewing window because the nanojets provided much better resolution. The contrast of the pattern was produced because the stripes had more matter in the nanojet than the grooves, and therefore they produced higher CARS signal when nanojets scanned across them. Thus, the bright parallel lines in Fig. 2 are the images of the stripes. However, the images were a magnified version of the actual patterns on the RC layer, because the scanning distance of nanojets was related to the scanning distance of laser foci by a magnification factor. We will discuss it in detail later.

 

Fig. 2 E-CARS images of three samples with spheres of diameters in the range 1-4 μm [(a), (b)], 1-5 μm [(d), (e)] and 1-6μm [(g), (h)] on the surface. The insets in (b) are the images of the same sphere at the Raman shift of 2840 cm⁻¹ and 3160 cm⁻¹, respectively. Intensity profiles along the line indicated by letters p to u are shown in (c), (f), (i). z is the focal plane position relative to the RC layer.

Download Full Size | PDF

To confirm that the signal in our images was from CARS but not fluorescence, we imaged the same sample at two different Raman shifts: 2840 cm⁻¹ and 3160 cm⁻¹. As mentioned earlier, the signal from the RC layer at 2840 cm⁻¹ should be partially resonant, while the signal at 3160 cm⁻¹ should be non-resonant. The insets in Fig. 2(b) are the images of the same sphere at these Raman shifts. The signal at 2840 cm⁻¹ was four times stronger than that at 3160 cm⁻¹ with the same excitation powers, which confirmed that we had CARS signal with resonant enhancement. Furthermore, when we adjusted the delay between the pump and Stokes pulses so that they no longer overlapped temporally, the signal essentially disappeared. These observations showed that fluorescence was negligible in our images.

As pointed out in Ref [21], the microspheres could generate non-resonant CARS signal within themselves. We could observe such signal when z became negative, i.e., the laser focal plane was outside the disc but at the spheres. However, the signal was weak and it didn’t show any pattern. Therefore, the line patterns in Fig. 2 were clearly from the disc surface but not the microspheres.

As mentioned above, spheres of different sizes provided strong nanojets in different z ranges. At z = 0 [Figs. 2(a) and 2(d)], we could see the disc pattern through 1-4 μm spheres, but not through 5-6 μm spheres like the sphere ‘r’ in Fig. 2(d). The 1-μm spheres could resolve the pattern when z was near 0, but the viewing window was too small to see a complete period. For 2-4 μm spheres, the pattern could be seen with the best contrast when the focal plane moved into the disc for a few μm as in Fig. 2(b). For 5-6 μm spheres, the pattern was the best resolved when the focal plane moved into the disc by 5 μm or more, as seen in Figs. 2(e), 2(g) and 2(h).

The intensity line profiles of the super-resolution patterns in selected spheres are shown in Figs. 2(c), 2(i) and 2(f). From these profiles we could estimate the range of imaging resolution. In the line profiles the peaks were produced by the 200-nm stripes and the dips were produced by the 100-nm grooves. The grooves were closer to ideal line objects (with negative contrast) and they were well resolved; the signal at the grooves was almost always less than half of that at the stripes. In some cases such as the profile ‘t’ in Fig. 2(i), the dip signal even came close to the background. These results indicate that an imaging resolution of at least 200 nm could be reached. On the other hand, it is also noticeable that the FWHM of peaks and dips were rather close and didn’t follow the 2:1 ratio. This indicates that the resolution was probably worse than 100 nm.

From the line profiles we could also measure the magnification factors (denoted as ‘M’) since we know that the distance between the peaks corresponds to 300 nm on the RC layer (the period of pattern). However, this simple measurement of M was based on the assumption that the CARS excitation within the viewing window was uniform. This could be a little over simplified because there may be a slow intensity fall-off of CARS signal when the laser beams moved away from the sphere center, as shown by some line profiles in Fig. 2. This intensity fall-off may be due to the reduction of nanojet intensity at the disc surface, and it varied from sphere to sphere. Some line profiles such as ‘r’ in Fig. 2(f) showed no intensity fall-off within a large range. The reason for this difference will be discussed with the simulation results later. If the intensity fall-off was present, it would cause the peaks to slightly shift towards the sphere center, and therefore, the measured M value should be slightly smaller than the real value. It is probably difficult to remove such effects by data processing, and here we present the measured M value as it is. M values for the profiles in Figs. 2 (c), 2(i) and 2(f) are tabulated in Table 1. From this data, M was usually 3~5, and it is clear that it depended on both the sphere size and the focal plane position.

Table 1. Magnification factor calculated from the CARS intensity profiles in Fig. 2.

View Table

To explain the magnification factors we saw, we preformed 2D Finite Element Method (FEM) simulations based on electromagnetic theory. Our 2D simulation method used the same conditions as in Ref [20], which was a cylinder model but verified to be valid for particles by comparing with the solutions based on the separation-of-variables method. The simulation model is illustrated in Fig. 3. The simulation area (36 µm X 20 µm) is divided into an air region (n = 1.0) and a disc region. Figure 3(a) shows part of the simulation area for a 5-μm sphere. The disc region contains three flat layers: the first layer has n = 1.6 and 20 nm thickness corresponding to the RC layer; the second is a 50 nm thick silver layer. These two thin layers appear as the dark line in Fig. 3(a). The third layer fills the rest of the space and has n = 1.6 corresponding to the SB layer. A microsphere of n = 1.46 is in the air region and in contact with the RC layer. A tightly focused laser beam usually has a complicated focal field, but here we simulate it by a TM Gaussian beam propagating along the + z-axis with a FWHM beam waist diameter of (λ/2NA) or an e⁻² radius of (λ/2.35NA) in intensity, where λ is 800 nm or 1000 nm and NA is 0.8 or 1.0. The beam waist is always centered laterally in the simulation area (x = 0), but the z-position varies for different simulations. The z = 0 plane coincides with the lower surface of the RC layer. In each simulation, the scattered field is calculated, and the intensity distribution of the total field is derived. The generation of a nanojet is clearly visible in Fig. 3(a). To calculate M, two simulations are needed. The first simulation has the sphere laterally centered at x = 0, i.e., it’s centered on the principle axis of the beam. By symmetry the nanojet should be centered at x = 0 if it exists. The second simulation has the sphere shifted in the lateral direction with its center at x = Δ. It’s equivalent to shifting the laser beam laterally which was the case in our experiments. Now the resulting nanojet should shift to a new location x = δ, and (Δ–δ) is the lateral shift of the nanojet relative to the sphere center [Fig. 3(b)]. M is then calculated as M = Δ/(Δ -δ), which is the scanning distance of the laser beam over that of the nanojet if the sphere was fixed.

 

Fig. 3 An example of numerical simulation done by using Finite Element Method (FEM). (a) Cross-sectional view of the local intensity distribution (|E|²) around a single sphere (n = 1.46, D = 5.0 μm) on a substrate. The incident laser (λ = 800 nm) beam is linearly polarized and propagates along the z-axis. The focus of Gaussian beam is identified with ‘O’ which is + 7μm above the RC layer. The dark line consists of a 20 nm flat layer with n = 1.6 and a 20 nm flat silver layer. (b) Line scan profiles showing a shift in the nanojet position (δ = 0.74 μm) when the microsphere is shifted in the x-direction for Δ = 1 μm. The M is calculated to be 3.8. (c) The magnified view of nanojet when x = 0 μm, i.e., the sphere is centered. (d) The magnified view of nanojet when x = 1 μm.

Download Full Size | PDF

Before discussing the results about M, it is helpful to take a look at other insights provided by the simulation. Figures 3(c) and 3(d) are the magnified view of nanojets generated by a centered and a shifted 5-μm sphere, respectively. Although it might be difficult to see, the nanojet in Fig. 3(d) was not only shifted in x direction, but was also slightly shifted in the z direction. This should be expected because with a lateral shift, the sphere surface at the location of the nanojet also curved away from the sample surface, and the nanojet could only reach the sample through a small gap. In Fig. 3(d), this gap was calculated to be 14 nm. As a result, the nanojet intensity may be slightly attenuated when reaching the sample surface, which explains the intensity reduction of the shifted peak in Fig. 3(b). This could be the main cause of the intensity fall-off of CARS signal when the laser beams moved away from the sphere center. However, in certain experimental data such as the line profile in Fig. 2(f), the intensity fall-off seemed to be absent within most part of the viewing window. It was probably because the real sample surface was complicated and the local contact condition varied from sphere to sphere. As a result, different spheres may show different intensity variations within the viewing window. On the other hand, the FWHM of both peaks in Fig. 3(b) was 300 nm. Therefore, although there may be an intensity variation of nanojet at the sample surface, the lateral FWHM of the excitation volume could still be rather constant. Again, the CARS excitation volume must be smaller than 300 nm due to the cubic intensity dependence. If assuming Gaussian peaks, one might expect the FWHM of the excitation volume to be close to 300 nm/$\sqrt{3}$ = 170 nm, which is consistent with our discussion about image resolution earlier.

Now we shall discuss the results about M. Experimentally we measured the average M for spheres of different sizes at different focal plane positions. The M for 2-µm spheres was 2.6 at z = + 1 µm and it was the only data point we could extract. The M for 3-µm spheres was 3.3, 3.8 at z = + 3, + 4 µm, respectively. For 4-μm, 5-μm, and 6-μm spheres, the measured M vs. z data are shown as the circle curves in Figs. 4(a)–4(c), respectively. Clearly, M increased when the focal plane moved into the disc. We don’t have a full theoretical explanation for this phenomenon, but intuitively it is analogous to the behavior of a convex lens. The focused laser beam at the back of the sphere is analogous to a virtual object, and the nanojet is analogous to its real image. M, or the ratio between the lateral shifts of object and image, is expected to increase when the virtual object moves farther away from the lens. We performed numerical calculations of M for spheres of various sizes at various focal plane positions as well. The results are plotted as the triangle curves in Fig. 4. An excellent agreement can be seen between numerical and experimental results. The numerical points don’t seem to form smooth lines because the position of nanojets may have an uncertainty of one grid size which translates into a numerical error of ~0.1 in M values.

 

Fig. 4 Lateral magnification obtained by 4, 5, 6 μm-diameter SiO₂ spheres on the Blu-ray disc as a function of focal plane position z. The circles (o) and triangles (Δ) represent the experimental and simulation results, respectively.

Download Full Size | PDF

Through simulation, we also found that M has very weak dependence on λ (800 nm or 1000 nm) or NA (0.8 or 1.0) for 2-6 μm spheres. The M values in these cases for the same sphere showed a difference of less than 0.1, smaller than the numerical error. However, for a 1 μm sphere, the M value at λ = 1000 nm is ~10% larger than that at λ = 800 nm, showing increased wavelength dependence. These results indicate that when the sphere is several times larger than the wavelength, the wavelength dependence of M is probably dominated by only material dispersion just like a common lens, and it is a relatively weak effect. But when the sphere becomes as small as the wavelength, a stronger dependence arises because the size of the sphere could directly affect the diffraction behavior of light. The wavelength insensitivity is important for CARS microscopy because CARS generation requires good overlapping between the pump and Stokes field. If M varies with wavelength significantly, we would expect CARS signal to have a much faster fall-off when the laser beams scan away from the sphere center. But in the images, we clearly saw nearly uniform patterns inside the viewing window of some big spheres. This is a confirmation of our theoretical speculations above. In addition, we performed E-CARS imaging experiments with a 40X 0.8-NA water objective and found that there was no significant change in magnification as predicted by simulation. The only difference seemed to be a weaker CARS signal.

Based on these results, bigger microspheres seem to be better for this imaging method because they provide larger viewing windows and less wavelength dependence. On the other hand, if the microspheres are too big, they lose the capability to generate nanojets. The upper bound of microsphere diameter for our laser wavelengths is 18 μm as mentioned earlier, while the biggest microsphere in our imaging experiments had a diameter of 6 μm. Therefore, it is possible to produce even better super-resolution CARS images than what we had with larger microspheres.

In conclusion, we have demonstrated super-resolution CARS imaging with photonic nanojets generated by SiO₂ microspheres. A virtual magnification up to 5.0X and a far-field lateral resolution of at least 200 nm at 796 nm pump laser wavelength were observed, and we could clearly resolve the sub-diffraction patterns on the surface of a Blu-ray disc sample which were unresolvable under normal CARS imaging. We found that the magnification varied with the microsphere size and the focal plane position of the laser beams, and these results were roughly explained by our theoretical simulations. From our results, it also seems preferable to use bigger microspheres for this imaging method. This method is not limited to SiO₂ microspheres. In fact, microspheres with higher refractive index (n>1.8) can be used to achieve better resolution [16,18]. And although we demonstrated the method on a reflective sample, it should work equally well for transmissive samples [16]. The use of photonic nanojets is an economical and straightforward way to achieve super-resolution in CARS microscopy and possibly other nonlinear optical imaging methods. It could be a very useful technique for applications that desire the vibrational imaging of nano-scale objects on films and surfaces.