Femtosecond Stimulated Raman Microscopy
Speeding up Raman Microscopy
- Fig. 1: a) Principle of Raman scattering: Incident photons with frequency nL are annihilated generating new photons with different frequencies nS,i (inelastic scattering). Hereby nR,i = nL - nS,i corresponds to the frequencies of molecular vibrations (left). The Raman spectrum is given by the frequency distribution of these photons. b) In Raman microscopy spectra are recorded at many locations in the sample. From those a false color image can be calculated using a variety of algorithms. c) Femtosecond stimulated Raman microscopy employs one broad- (“probe”) and one narrow-band (“pump”) laser beam. If the frequency difference matches the one of a molecular vibration, stimulated Raman loss is observed in the probe beam. This can be used to calculate the Raman spectrum.
- Fig. 2: Set-up used for FSRM: The pulses of a Ti:Sa-laser, which are only a few femtoseconds long, serve directly as Raman probe. The narrow-band pump pulses are generated using a custom-built fiber amplifier. After interaction of the two pulses in the sample the spectra are recorded using an imaging spectrograph and a diode array. Spatial resolution is achieved by raster-scanning using a xy-piezo-stage.
- Fig. 3: Top: Spectrum of neat benzonitrile recorded within 0.1ms using the FSRM set-up. Bottom: Raman images of a binary polymer-blend consisting of SAN droplets (blue) embedded in a PMMA matrix (red) recorded with two different acquisition speeds. Typical spectra of PMMA rich (left) and SAN rich (right) areas recorded with an acquisition time of 0.1 ms are shown below. Adapted from .
Raman microscopy yields images with chemical contrast without relying on molecular labels. Its application is hampered by long acquisition times. This can be mitigated by applying non-linear Raman techniques. The one described here, Femtosecond Stimulated Raman Microscopy (FSRM), retains the spectral coverage of conventional Raman microscopy.
Raman microscopy was first described more than 40 years ago . Since then it has evolved into an imaging tool with (potential) applications in areas ranging from materials science  to clinical diagnostics . Various manufacturers offer turnkey instruments for this type of imaging. Despite this seemingly mature status there is very active methodological research in this field. Grasping the motivation for these efforts requires some basic understanding of conventional (linear) Raman microscopy.
Raman microscopy in general makes use of the inelastic light scattering by molecular vibrations. In linear Raman microscopy and spectroscopy one illuminates the sample with monochromatic light of frequency νL. Nowadays, usually a laser provides this light. In the course of the inelastic scattering incoming photons are annihilated and new photons with frequencies νS,i are generated. The frequencies of the generated photons are given by νL -νS,i = νR,i (fig. 1a). Hereby, νR,i are the frequencies of the Raman active vibrational modes of the molecule studied. These frequencies depend on the masses of the atoms constituting the molecule and the force constants of the chemical bonds connecting them. The photon frequencies νS,i and the respective signal strengths, in other words the Raman spectrum, are therefore very informative about the chemical constituency of matter. This is a property Raman spectroscopy shares with infrared (IR) spectroscopy. To exploit this for microscopy the laser light is focused onto the sample and scattered light is collected by an objective. After spectral dispersion the Raman scattered light is registered by a multi-channel detector.
By raster-scanning the region of interest a three dimensional (two spatial and one spectral coordinate) data set is obtained. Algorithms transform these data sets into false color images representing concentrations of chemical entities. These images are often referred to as chemical maps.
“Drawing” these maps obviously requires acquisition of many Raman spectra, for instance 10 000 for a 100×100 spatial raster. Small acquisition times per spectrum are crucial in this context. Assuming an acquisition time of one second per Raman spectrum – an interval which is very tolerable in spectroscopy – the 100×100 Raman image takes around three hours to record. Unfortunately, the Raman effect is not very cooperative when it comes to rapid acquisition. The Raman signal strength depends on power of the incoming laser light and the Raman cross section. For reasons of spatial resolution the laser beam is tightly focused to arrive at a diameter of around one micrometer. This limits the incoming laser power to around 0.1-1 W since otherwise the intensities (power per area) at the focus exceeds typical damage thresholds. The Raman cross section is an effective area parametrizing the interaction strength of light and matter with regards to the Raman effect. Typical values are in the order of 10-28 cm2 or smaller . For comparison, cross sections for light absorption, responsible for the color of matter, are around 10-17 cm2 . The small Raman cross sections and the limits in the laser power translate into typical acquisition time of 0.1-1s per spectrum in linear Raman imaging (see discussion in ref. ).
Non-linear Raman Spectroscopy
Many groups work on mitigating that by using non-linear Raman effects. Non-linear Raman comes in many “flavors”, the most prominent being coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS). CARS [7-8] as well as SRS [9-10] were introduced into microscopy. Both methods have in common that they require two incoming laser beams with different frequencies. Photons of the two beams have to coincide at the sample which is why commonly pulsed lasers are employed. Provided that the frequency difference of the two beams equals the frequency of a Raman active mode, νR,i, a Raman transition is driven by the two beams. Thereby, the effective Raman cross section is increased. The required matching of the frequency difference renders the acquisition of spectra more challenging than in linear Raman spectroscopy. It can either be achieved by scanning the frequency of one of the two lasers or by a multiplex approach. In non-linear Raman there is a tradeoff between spectral coverage and acquisition speed .
Femtosecond Stimulated Raman Microscopy
Femtosecond stimulated Raman microscopy (FSRM)  introduced by our group is a multiplex approach (see fig. 1) which aims at faster acquisition while retaining the spectral coverage of linear Raman microscopy. It utilizes a spectrally broad and weak femtosecond pulse (Raman probe, frequency components νPR) and an intense spectrally rather narrow picosecond pulse (frequency νPU). The two pulses interact at the focus of a scanning microscope. Stimulated Raman interaction generates dips or peaks in the Raman probe spectrum provided that the condition |νPR-νPU|=νR,i is fulfilled. Dips are observed for νPR-νPU>0, peaks for νPR-νPU<0. The former condition is realized in our instrument. The ratio of the Raman probe spectrum in presence and absence of the Raman pump light yields the Raman spectrum of the sample. The spectral width of the Raman probe pulse should be comparable to the largest Raman shifts amounting to ~ 3000 cm-1 (in vibrational spectroscopy wavenumbers ν/c are usually employed), so most Raman bands can be addressed simultaneously. Furthermore, because of the referencing mentioned above, the Raman probe light should be very stable in terms of amplitude and spectral shape. Since the Raman pump pulse defines the spectral resolution of the instruments, its spectral width ought to match typical Raman linewidths in solution (1-10 cm-1).
The light source we have developed for FSRM comprises a commercial titanium sapphire based femtosecond laser (fig. 2) . It emits pulses of a few femtoseconds and therefore features the required spectral width. The pulses emitted by the laser serve – after attenuation – as Raman probe light. The avoidance of non-linear spectral broadening ensures a very stable probe light. A homebuilt fiber amplifier produces the Raman pump light. The amplifier is seeded by a small fraction of the laser output which ensures perfect synchronization of pump and probe light. The probe light leaving the scanning microscope is directed towards a spectrograph equipped with a 512 elements detector [13-14]. The maximal read-out rate of the detector is 20 kHz. The pump light is on/off modulated at a rate of 10 KHz. Thus, the shortest possible acquisition time per spectrum is 0.1 ms.
At least for a neat substance, benzonitrile here, it is possible to record a complete Raman spectrum featuring a signal-to-noise ratio of around 10 within this short period (fig. 3) . Also imaging with this short acquisition time per spectrum is feasible. This was demonstrated with a blend of two immiscible polymers (poly(styrene-co-acrylonitrile), SAN and poly(methyl methacrylate), PMMA). Based on spectra obtained within only 0.1 ms a false color image can be generated. From this image it becomes obvious that the islands of SAN are surrounded by a continuous phase of PMMA. This chemical information cannot be obtained with ordinary light microscopy.
At present the detector limits the performance of our instrument. With an “ideal” detector , which is not out research technologically, a threefold shortening of the acquisition while retaining the signal to noise ratio should be feasible. The light powers of pump and probe used here to arrive at this noise level are considered compatible with biological samples. Since in bio-samples the concentrations of scatterers will be much lower than in the polymer blend studied here we do not expect to obtain images with acquisition times of 0.1 ms. Figuring out where FSRM stands in terms of bio-imaging is the next thing on our agenda.
 Hideaki Kano, Hiroki Segawa, Philippe Leproux, and Vincent Couderc: Linear and nonlinear Raman microspectroscopy: History, instrumentation, and applications, Optical Review, 21(6): 752-761 (2014) doi: 10.1007/s10043-014-0123-9
 Ingrid De Wolf: Micro-Raman spectroscopy to study local mechanical stress in silicon integrated circuits, Semiconductor Science and Technology, 11(2): 139 (1996) doi: 10.1088/0268-1242/11/2/001
 Max Diem, Antonella Mazur, Kathleen Lenau, Jen Schubert, Ben Bird, Milo Miljković, Christoph Krafft, and Jürgen Popp: Molecular pathology via IR and Raman spectral imaging, Journal of Biophotonics, 6(11-12): 855-886 (2013) doi: 10.1002/jbio.201300131
 Bernhard Schrader: Infrared and Raman Spectroscopy: Methods and Applications, Wiley-VCH Verlag GmbH (1995) doi: 10.1002/9783527615438
 William W. Parson: Modern Optical Spectroscopy, Springer-Verlag Berlin Heidelberg (2007) doi: 10.1007/978-3-540-37542-5
 Chi Zhang, Delong Zhang, and Ji-Xin Cheng: Coherent Raman Scattering Microscopy in Biology and Medicine, Annual Review of Biomedical Engineering, 17(1): 415-445 doi: 10.1146/annurev-bioeng-071114-040554
 J. Reintjes, M. D. Duncan, and T. J. Manuccia: Picosecond Coherent Anti-Stokes Raman Scattering (CARS) Microscope, Proceedings of the Society of Photo-Optical Instrumentation Engineers, 322: 87-92 (1982) doi: 10.1117/12.933212
 Andreas Zumbusch, Gary R. Holtom, and X. Sunney Xie: Three-Dimensional Vibrational Imaging by Coherent Anti-Stokes Raman Scattering, Physical Review Letters, 82(20): 4142-4145 (1999) doi: 10.1103/PhysRevLett.82.4142
 E. Ploetz, S. Laimgruber, S. Berner, W. Zinth, and P. Gilch: Femtosecond stimulated Raman microscopy, Applied Physics B, 87(3): 389-393 (2007) doi: 10.1007/s00340-007-2630-x
 Christian W. Freudiger, Wei Min, Brian G. Saar, Sijia Lu, Gary R. Holtom, Chengwei He, Jason C. Tsai, Jing X. Kang, and X. Sunney Xie: Label-Free Biomedical Imaging with High Sensitivity by Stimulated Raman Scattering Microscopy, Science, 322(5909): 1857-1861 (2008) doi: 10.1126/science.1165758
 Charles H. Camp Jr, and Marcus T. Cicerone: Chemically sensitive bioimaging with coherent Raman scattering, Nature Photonics, 9(5): 295-305 (2015) doi: 10.1038/nphoton.2015.60
 E. Ploetz, B. Marx, T. Klein, R. Huber, and P. Gilch: A 75 MHz Light Source for Femtosecond Stimulated Raman Microscopy, Optics Express, 17(21): 18612-18620 (2009) doi: 10.1364/OE.17.018612
 B. Marx, L. Czerwinski, R. Light, M. Somekh, and P. Gilch: Multichannel detectors for femtosecond stimulated Raman microscopy - ideal and real ones, Journal of Raman Spectroscopy, 45(7): 521-527 (2014) doi: 10.1002/jrs.4528
 L. Czerwinski, J. Nixdorf, G. Di Florio, and P. Gilch: Broadband Stimulated Raman Microscopy with 0.1 ms Pixel Acquisition Time, Optics Letters, 41(13): 3021-3024 (2016) doi: 10.1364/OL.41.003021
Jakob Nixdorf1 and Peter Gilch1
1Institut für Physikalische Chemie, Heinrich-Heine-Universität Düsseldorf, Düsseldorf, Germany