Advances in submicron infrared vibrational band chemical imaging

The technique of infrared near-field microscopy with submicron resolution is an important addition to the chemical sciences arsenal in the last few years. Although related to highly successful scanning optical probe microscopies in the visible, infrared near-field microscopy had to overcome several obstacles, which slowed its development. This review illustrates the history as well as the state of the art of this new field, its limitations and perspectives. At present, two main experimental approaches have been successful: the apertureless metal tip approach and the fibre tip aperture approach. The two variants are compared from the point of view of resolution, ease of implementation in the laboratory and image formation mechanisms. The techniques using chemically specific vibrational absorption contrast are emphasized here, in the general context of chemical microscopy, which includes other methods such as chemical force, Raman and fluorescence microscopies. The phenomenon of surface-enhanced infrared absorption is also mentioned in relation to near-field infrared microscopy, with regard to important aspects of image formation and possible improvements. The main advantages of spatial resolution, chemical sensitivity, non-intrusiveness, minute amounts of specimen and the possibility of quantitative analytical measurements make infrared near-field microscopy a powerful tool. We also examine here possible future applications that go beyond the limits of classical vibrational microspectroscopy, as well as directions for additional advances.


Introduction
Mapping the chemical state of a specimen is of great importance in a variety of scienti® c ® elds. Some materials, such as polymer ® lms or living cells, possess a complex hierarchical organization at levels ranging from molecular to macroscopic while others, such as semiconductor devices, are made from rather simple building blocks, spanning sizes of the order of a few nanometres to microns. Studies of such systems more often than not require some types of direct imaging technique, which provide both chemical contrast and spatial resolution.
The technique of infrared (IR) microscopy has developed into a powerful analytical tool because it can perform spatially resolved chemical analyses in situ, is non-intrusive and requires small amounts of sample. The technique combines spatial resolution with the possibility of tuning the IR wavelength across the chemically speci® c vibrational bands of the sample. Typically, a conventional microscope working at IR wavelengths is combined with a Fourier-transform infrared (FTIR) spectrometer to acquire a spectrum at each spatial resolution element of the image. The quantitative character of the method arises from the fact that the IR optical density is linearly proportional to the number of molecules (absorbers ) in the beam path.
The importance of these qualities is re¯ected in the broad variety of applications encompassed by FTIR microscopy investigations, such as materials science, forensics, biological sciences, medical research, conservancy and identi® cation of cultural artefacts, polymer science, electronics and industrial characterization of adhesives, laminates and lubricants. Since this review does not intend to be comprehensive in the area of FTIR microscopy, interested readers are referred to the original literature [1,2].
Many problems studied by FTIR microscopy would gain from better spatial resolution. Several examples of such problems are considered in detail and others are listed in section 4 of this review. The spatial resolutions of available FTIR microscopes range from a few microns to a few tens of microns depending of the wavelength. The fundamental limit that prevents submicron resolution in the IR is the diOE raction limit ; a lens-based imaging apparatus cannot achieve a better spatial resolution than approximatel y half of the wavelength (about k}2). However, in practice, the maximum spatial resolution achieved in IR microspectroscopy is approximatel y 2k, because of throughput de® ciencies and optical aberrations [3]. The closest method of investigation to IR microspectroscopy is Raman scanning microscopy. In this case, a similar but complementary type of molecular information (molecular vibrations ) as in IR microscopy is revealed [4], while the spatial resolution is de® ned by the spatial pro® le of the focused laser beam, which has a wavelength typically situated in the visible or near-IR. In this way, resolutions close to 1 l m are achievable. One problem with Raman microscopy is the fact that it is a nonlinear process with a small cross-section ; therefore the possibility of thermal damage to the sample by the pump beam limits either the spatial resolution or the sensitivity in practical situations [5].
There are a number of other techniques that enable submicron mapping with chemical speci® city, which will be brie¯y enumerated in the next section, but only IR near-® eld microscopy truly transfers all the characteristics of FTIR microscopy into the submicron realm. Near-® eld scanning optical microscopy (NSOM) has broken the diOE raction limit barrier by more than one order of magnitude [6,7] but until recently has been restricted to optical wavelengths. Because of the longer wavelengths, the use of infrared with near-® eld optical microscopy is di cult. The cross-sections of vibrational transitions are generally smaller than the electronic transitions usually used in visible-wavelength NSOM for contrast. The detectivity of IR detectors is also orders of magnitude smaller than in the visible. Finally, the far-® eld to near-® eld transfer of radiation occurs with greater losses in the IR, for the same spatial resolution. Nevertheless, with new IR materials for ® bres, new apertureless probes and better design, the technique of IR-NSOM is now becoming a reality. Although chemical contrast has been demonstrated, IR-NSOM still needs widely tunable bright IR sources in order to acquire a broad wavelength spectrum at each resolution element, or pixel, of an image. This is the last obstacle that prevents the FTIR microspectroscopy capabilities from being fully transferred to near-® eld IR spectroscopy. It is worth mentioning that not only can chemically speci® c IR contrast be achieved using IR-NSOM, but also topographic mapping of the sample surface is obtained.
The subject of this review is focused on rapidly emerging IR-NSOM techniques. Various solutions to the inherent problems related to the utilization of IR wavelengths are described. The topic is still in its initial stage and much of the reviewed material focuses on technical issues, such as the probe fabrication or the resolution limit. However, several applications already available are also reviewed, such as the mapping of buried semiconductor interfaces and the characterization of ion-implanted semiconductor surfaces and III± V semiconductor heterostructures, carrier dynamics in silicon, polymer blend chemical mapping, line dimensions and shrinkage in photochemically modi® ed polymeric resists and polymer laminate systems. Although a review is usually aimed at a description of what has been done in the ® eld, we feel that the truly important contributions from IR-NSOM are yet to come. We also discuss two examples here, which we think would greatly bene® t from a tenfold to hundredfold improvement in spatial resolution with respect to FTIR microscopy, a performance already achieved by several groups working on IR-NSOM. One is the relation between the diOE usion of small molecules and the mesoscopic structure of thin polymer ® lms, a direction that has been undertaken in our laboratory. The other is the study of the deoxyribonucleic acid (DNA) localization in a living cell undergoing its natural division cycle, an important issue related to the progression from normal to precancerous and cancerous cellular states. in the inset [8]. The sample rests on a piezoelectric x, y, z translation stage. A laser beam is re¯ected from the back side of the cantilever on to a photodiode to measure the de¯ection or the torsion of the cantilever when the tip interacts with the surface.

Submicron chemical microscopy techniques
This section deals brie¯y with the principles of various scanning probe microscopies used to acquire spatially resolved chemical information, in addition to IR-NSOM. These include atomic force microscopy (AFM) and chemical force microscopy (CFM), for which comprehensive reviews related to these techniques already exist [5,8]. Since we are concerned here with IR probes, other NSOMs are also described in more detail as an introduction to and because of their close relation to the IR-NSOM technique. All these high-resolution chemical microscopy techniques are evoked here to provide a basis for comparison and to emphasize the contrasting aspects with respect to IR near-® eld microscopy, which is described in a later section.

Atomic force microscopy
In the atomic force microscope a miniature cantilever with a sharp tip (radius, about 5 nm) scans the surface (® gure 1) [9,10]. The interaction between the surface and the tip depends, among other factors, on the nature of the sample and the tip material. These forces can be measured by monitoring the de¯ection of a laser beam due to the bending of the cantilever (contact mode) by the short-range forces between the tip and the surface. Alternatively, the reduction in the free oscillation amplitude when the tip is brought close to the surface is employed (tapping mode). Local sample stiOE ness and adhesion can be quanti® ed in this way [11]. A priori knowledge of the values of these mechanical parameters gives access to the chemical nature of the sample. A variable force range allows diOE erentiation between phase-separated regions lying below the surface. Varying the surface temperature of a polymer sample, for example, has permitted spatial measurements of the onset of the glass transition [12]. Resonant frequency shifts and phase changes of the oscillation were also used to emphasize diOE erences between elastic¯uctuations of topographically smooth samples (see, for example, [11,13]). Although useful in the determination of the micromechanical characteristics of the sample at spatial resolutions down to the atomic scale [14], AFM cannot provide speci® c chemical group information. In the case of polymer blends with phase separation, an elegant solution to the lack of the chemical speci® city problem was demonstrated by Krausch and co-workers [15]. They used diOE erent solvents and polymer swelling to identify unambiguously the diOE erent phases present at the surface.

Chemical force microscopy
To overcome the problem of chemical non-speci® city, Lieber and co-workers [16] introduced in 1994 the concept of CFM, an AFM variant based on chemically functionalized tips. This approach takes advantage of the fact that the absolute force resolution in AFM can be several orders of magnitude smaller than the weakest chemical bond. With tips covered with a well-de® ned layer of molecules, molecular interactions between the chemical groups of the surface and those of the tip can be estimated directly (® gure 1). CFM not only is used to probe forces between diOE erent chemical groups but also can provide an energy surface map on a nanometre scale and determine pK values of surface acid and base groups locally. Through lateral force imaging, diOE erences in friction between the functionalized tip and the sample can be used to map the chemical¯uctuations of the sample [8].
While tremendous progress has been achieved in the last few years in this area, especially in applications concerning biomolecular interactions, there are several important issues which have to be kept in mind when comparing force microscopy with IR, chemically speci® c imaging techniques.
(A) Force microscopy is a surface-sensitive technique since the interaction comes mainly from the topmost layers of the tip and of the sample. (B) A priori knowledge of the sample is necessary and the probe has to be speci® cally prepared to achieve predictable chemical contrast. (C) The chemical force has to be separated from other interactions related to morphology, mechanical properties and capillary forces, which together determine the tip± sample interaction. (D) The access to molecular dynamics is restricted by the relatively low scanning speed. Figure 2. Near-® eld image formation principles illustrated with two apertures in two opaque screens juxtaposed to each other. The second aperture transforms the near-® eld waves excited by the progressive wave on the left of the ® rst aperture, according to the reversibility of the light path, into progressive waves, again. Without the second aperture, the near-® eld waves would be exponentially damped in the direction normal to the ® rst aperture surface. The wavelength is k.

Near-® eld scanning microscopies 2.3.1. Principles of near-® eld imaging
The techniques of NSOMs inherit the power of optical microscopy and combine it with the high spatial resolution of scanning probe microscopy. Synge [17] laid out the principles of near-® eld microscopy in 1928 ; a very small probe (which can be an emitter or a collector of light) is brought into close proximity to the sample surface. The light scattered from the local interaction with the sample is collected in the far ® eld, by a conventional lens system, for example. The probe is scanned across the sample surface, and a raster image of the sample is formed. The near-® eld term (and with it the high spatial resolution) comes from the requirement of having the probe and the sample very close to each other ; otherwise diOE raction would prevent subwavelength resolution from being achieved during image formation. Suppose that the probe is a small light source. A broad spectrum of spatial frequencies is associated with the localized electromagnetic ® eld of the source. In free space, the high spatial frequencies decay exponentially, radially from their source, and only the low spatial frequencies (less than 2}k) of the Fourier spectrum propagate into the far ® eld. When the sample is brought close to the subwavelength source of light, the evanescent ® eld is retransformed to propagating waves through scattering events with the ® ne details of the sample, which become present in the image (® gure 2). Numerous reviews on near-® eld microscopy theory and applications have been published to date [18± 25]. We mention here only those aspects of near-® eld microscopy that provide a minimal basis for the IR-NSOM discussion at the end of this section.
Although many diOE erent experimental con® gurations have been reported in the early years of NSOM [18,26], from the viewpoint of the measurement process they can be classi® ed into two main categories : illuminating probe devices and collecting probe devices. In illuminating probe devices, the electromagnetic ® eld emerges from a small source of light (® gure 3 (a)). The key feature of the most common near-® eld scanning optical microscope is the fabrication of a subwavelength optical aperture at the apex of a tapered transparent optical ® bre, which is metal coated [27].
In collecting probe devices, the sample illumination need not be achieved through a transparent tip. Collection of the light can be achieved through a ® bre tip or even apertureless probes can be used, usually of the same type as an atomic force microscope tip. In the apertureless case, an external ® eld illuminates the sample and the evanescent waves associated with the interface are probed by the tip, which acts here like a scatterer to convert the sample-bound near ® eld into propagating waves. The sample can be externally illuminated, as in ® gure 3 (b), or internally illuminated at angles greater than the critical angle of total re¯ection. This last variant it is also called the photon tunnelling scanning microscope [7].
From a practical point of view, in near-® eld microscopy, one is concerned with the light throughput (signal), the spatial resolution, the contrast mechanisms or image formation and the presence of artefacts in the optical images. We shall focus here on the optical ® bre-based illuminating con® guration and the apertureless collection con® guration.

The optical ® bre aperture near-® eld scanning optical microscope
The light travelling through a tapered optical ® bre will experience re¯ection in the taper region owing to the continuous mismatch of the light modes and the waveguide since the impedance of the waveguide varies continuously with the cross-section [28]. Near the end of the taper, beyond the cut-oOEdiameter, the ® bre does not sustain progressive modes of propagation, and the ® eld is evanescent. The shorter the distance that the light has to propagate through evanescent modes, the larger will be the ® eld that reaches the aperture (® gure 4). It follows that, from the viewpoint of light throughput, wide-taper cone angles are preferable. Also, single-mode ® bres are better than multimode ® bres for a near-® eld scanning optical microscope since they do not have losses associated with the early cut-oOEof higher modes [29,30]. Besides the attenuation due to the cut-oOEof propagating modes of the light guide, one must consider the transmission coe cient of the aperture itself. Bethe [31] in 1944 and Bouwkamp [32] considered the problem of the transmission of a plane electromagnetic wave through a hole in an in® nitely thin perfect conductor screen and provided an analytical solution to the problem. The Bethe± Bouwkamp model for a near-® eld scanning optical microscope aperture is remarkable in the sense that it catches all the main characteristics of the scattering within and outside the hole and therefore the transmission, even when compared with more realistic numerical approaches [33]. According to this model, the aperture transmission coe cient should scale in a ® rst approximation as (a}k)%, where a is the aperture radius, and k is the wavelength [34]. The severe signal limitation imposed by the fourth-power law, as well as the importance of the taper shape in this matter, triggered the interest in research on optimal tip fabrication procedures. Two main ® bre tip fabrication technologies emerged : the ® rst was ® bre pulling [35]. In this procedure, a focused CO # laser heats the ® bre locally, while pulling. A narrow taper and a neck form in the melted region (® gure 5). The neck breaks eventually, for glass ® bres typically when it is below 100 nm in size. The exact value depends on the pulling and heating parameters, which can be controlled to some extent [36,37]. Such a pulled tip with a 50 nm aperture obtained by lateral coating with a metal ® lm typically exhibits 10 ' transmission e ciency.
The other procedure is chemical etching (the Turner [38] method) [39]. The ® bre cladding is ® rst removed and then the ® bre is dipped into an etching¯uid with an organic protective overlayer¯oating on it. Chemical etching usually generates larger cone angles than mechanical pulling. HoOE man et al. [40] and Zeisel et al. [41] showed that adequate choice of organic solvent could control the cone angle between 8 and 41°. Because of their larger cone angle, the etched ® bres by the Turner method provide higher optical throughput than the pulled ® bres ; however, the etching process usually leaves the taper surface rough, and the tip asymmetric. Owing to the surface roughness, pinholes form in the subsequently deposited metal coating, and leakage of light through these pinholes adds a background to the near-® eld imaging by the aperture, thus decreasing the signal-to-noise ratio. Optimization of the etching procedure has been achieved for glass ® bres by the`tube etching ' method [42,43]. In this technique, the acrylate jacket of the ® bre is not removed before dipping the ® bre into etchant. During the process, the acrylate jacket plays the role of a protective container, yielding smooth surfaces after etching. With ® bre tips of this type, transmissions up to 10 $ for an aperture of 70 nm diameter were reported. The signi® cant increase in the instrument throughput achieved by the etching procedure allowed Bukofski and Grober [44] to obtain near-® eld images at video rates. Besides the signi® cant increase in transmission, the tube-etched ® bres are characterized by much better reproducibility than both etched (by the Turner method) and pulled ® bres. Limitations of this method are the acrylate jacket removal step, which threatens the tip integrity and the ® bre material speci® city. Since the truly important variable for light attenuation within the taper is the length of the region of evanescent propagation, another source of throughput improvement is the optimization of the taper shape. Two-step tapered probes with improved throughput (10 $± 10 # transmission for an aperture of 40 nm diameter) were obtained by a hybrid pull± etch method on glass ® bres, while a combination of focusedion-beam milling and chemical etching yielded comparative results from triple-step tapered ® bres [45]. A double-taper structure obtained by a variable pulling velocity is currently employed in our laboratory, on¯uoride tips for IR-NSOM [46].
The throughput of the ® bre tip would not be a problem if there were not an incident light energy threshold associated with the damage of the metal coating. For aluminium-coated ® bres this threshold is about 10 mW (continuous-wave laser). An increase in the damage threshold (fourfold ) was observed when using additional adhesion layers (titanium, chromium, cobalt or nickel) between the glass surface and the aluminium coating [47].
The damage energy threshold is not the only parameter that one must consider during tip fabrication. For the best spatial resolution, the roughness (as well as the oxide formation during the metal coating procedure) has to be minimized [48] by using high evaporation rates at base vacuum pressures below 10 $ Pa (10 & Torr). The grainy structure of the metal coating can also cause the aperture to deviate from a circular shape, reducing the optical imaging capabilities and making the images more di cult to interpret [36]. Veerman et al. [49] and Pilevar et al. [50] introduced the use of focused-ion-beam milling to polish the end of the coated ® bre. They obtained well-de® ned¯at apertures down to 20 nm, having improved polarization, throughput and resolution characteristics (® gure 5). Moreover, in this case the oblique evaporation of aluminium is not required and, therefore, the pinhole occurrence in the metal ® lm in the taper region is reduced. Although a costly procedure, a few groups systematically use it, especially for demanding applications such as polarization-selective¯uorescence NSOM of single molecules [22,51].
The issue of the spatial resolution of the aperture near-® eld scanning optical microscope has been discussed (see for example [22,52]). Numerical approximations to solutions of the Maxwell equations applied to a conical transparent tip coated by a metal layer predict that, close to the aperture, the power density is laterally con® ned to the same area as the aperture itself plus the ® eld penetration depth into the metal coating. Normal to the aperture plane, the power density remains approximately collimated up to distances equal to the aperture radius [53]. At a distance greater than the aperture radius, the emerging light is rapidly diverging, approximatel y following a dipole pattern (® gure 6). It follows that the near-® eld scanning optical microscope resolution depends on the working distance of the microscope, an interesting feature that diOE erentiates near-® eld microscopes from conventional optical microscopes.
In order to obtain the best spatial resolution, the sample and the tip must be brought close to each other, at least at a distance comparable with the aperture radius, that is at least about 50 nm for glass ® bres working in the visible region. This requirement is usually achieved by a close-range (less than 20 nm) mechanical (shear± force) interaction between the tip and the sample. The principles of the feedback loop and implementation are similar to those encountered in AFM [26]. Keeping the interaction strength constant during the scan provides a constant gap between the tip and the surface features. However, the strong evanescent ® eld gradients normal to the surface induce a pure topographic component in the near-® eld signal if the tip has to follow the surface height pro® le. These topographical artefacts are a continuous source of concern especially at very high resolutions [54,55], or on rough surfaces. Fortunately, the constant preoccupation to elucidate the mechanisms of near-® eld image formation showed that, if the constant-ga p mode is replaced by constant-height scanning (the tip scans in a horizontal plane above the surface, without feedback from the sample topographi c features), the images are free of topographic artefacts [56,57]. Since it is not always possible to generate a constantheight image in close proximity to the sample, taking an ensemble of diOE erent constant-gap images of the same area can be used for e cient artefact removal from the optical image. The method is called pseudo-constant-heigh t imaging [58,59]. While e cient in eliminating the artefacts, this method requires, in principle, a long time (hours) for the acquisition of a single image.

The apertureless near-® eld scanning optical microscope
The ® bre-based aperture near-® eld scanning optical microscope has been a popular choice because it provides an intuitive, although arguable, similarity to scanning aperture confocal microscopes, which instead of having a scanning physical aperture above the surface use the real image of an aperture (pinhole) formed through a lens system. However, in recent years, the apertureless near-® eld scanning optical microscope, based on atomic force microscope tips (® gure 3 (b)) emerged as a very promising variant providing the following four distinct advantages with respect to the aperture near-® eld scanning optical microscope.
(a) The spatial resolution of the apertureless microscope depends on the enhancement of the light ® eld in the vicinity of a sharp probe tip. The ® eld enhancement eOE ect is spatially con® ned within a zone of the order of the tip radius, which determines the spatial resolution [60,61]. Radii of the order of 5 nm are now common for commercial atomic force microscope tips, and optical resolutions better than 10 nm, using silicon [62] or metal tips [63], were reported. (b) The major problem of the ® bre taper transmission of the aperture near-® eld scanning optical microscope is overcome in the apertureless near-® eld scanning optical microscope by focusing the light directly from free space on the tip. This is an important feature, especially when using long wavelengths. (c) A simple coupled-dipoles model is su cient in some cases to understand the contrast mechanism [64]. Following Zenhausern et al. [64], if the tip and the sample feature are modelled by spheres of radius a and polarizabilities a t and a f , the polarizability modulation which describes the coupling between the tip and the feature is given by where r is the tip± surface distance. This relation has been successfully used to model the height dependence of the scattered ® eld and allows the calculation of the complex polarizability of the surface feature from the scattered ® eld, when the tip polarizability is known. The latter can in turn be experimentally found on a reference surface. In addition, since the apertureless microscope essentially works in collection, the acquired images should re¯ect the same electromagnetic near-® eld distribution at the surface as in the absence of the tip, in comparison with the illumination aperture near-® eld scanning optical microscope for which, in some instances, the images are more di cult to interpret [65]. (d ) An apertureless microscope is not limited by the wavelength range of the probe transparency as for ® bre-based near-® eld scanning optical microscopes. Moreover, a ® eld enhancement will take place near the sharp apex of an atomic force microscope tip regardless of the nature (insulator or metal) of the tip. This adds¯exibility for the imaging mechanisms, without restricting the resolution. This last feature is especially useful for projected studies of IR near-® eld spectroscopy, which requires a relatively broad band of wavelengths.
The limitations of the apertureless microscope come from the fact that the topographical artefacts associated with the constant-gap scanning mode are, in this case, enhanced. Scattering from zones of the probe other than the tip, combined with the z motion associated with the topographic feedback can obscure the actual near-® eld signal. Then either the apertureless near-® eld scanning optical microscope is limited to¯at samples [66], or supplemental technical precautions have to be taken in order to exclude topographic artefacts while in the constant-gap mode [63,67,68]. Since the apertureless microscope is based on the local ® eld enhancement at the tip, the depth of probing into the sample is reduced too, with respect to an aperture near-® eld scanning optical microscope working in illumination. Therefore, the apertureless near-® eld scanning optical microscope quali® es more as a surface probe. Depth discrimination in near-® eld experiments has been experimentally achieved using the aperture near-® eld scanning optical microscope [69].

Optical near-® eld microscopy in chemistry
In principle, having to take into account evanescent waves for rigorous interpretation of the NSOM images prevents the use of any simple approximation in the set of Maxwell' s equations [20], and therefore the scattering problem is di cult. There are a few cases, however, in which the elaborate theoretical treatment required for a complete understanding of the near-® eld image leaves open the possibility of simpler approaches, allowing the practical use of near-® eld microscopy for the characterization of selected phenomena. A few examples of work in this category for chemically speci® c probing are described below.

Fluorescence near-® eld microscopy
Room-temperatur e¯uorescence NSOM is an extremely sensitive technique allowing single-molecule detection [70]. Often, sample preparation by adding ā uorescent tag to the molecules of interest is necessary. Sometimes the sample's own uorescence can be used, as in the case of the green¯uorescence protein [71]. Although single-molecule detection and spectroscopy with a near-® eld aperture was initially used to probe single-molecule orientation, in most of the single-molecule studies, the use of a local probe is not necessary and confocal (far-® eld) microscopes can be equally employed with greater ease. Moreover, it is known that near-and far-® eld measurements can exhibit diOE erences, owing to the in¯uence of tip metal coating on the emission lifetime and the¯uorescence intensity. A near-® eld probe is desirable only when topographic information from the sample is required at the same time as the optical image, or when the average distance between two¯uorophores is less than the spatial resolution limit of a far-® eld microscope.
Applications of¯uorescence single-molecule detection include¯uorophore photochemistry studies, the motion of individual motor proteins, diOE usional trajectories of labelled molecules in membranes, and conformational protein dynamics through uorescence resonance energy transfer. Although the ® eld emerged less than a decade ago, a few reviews are available on single-molecule detection and applications (see, for example, [51,72,73]). Applications of¯uorescence NSOM to the study of thin organic ® lms, and in particular to the study of molecular semiconductor junctions and emitting conjugated polymers have been reviewed in [74]. Strategies for high-resolution uorescence imaging of living cells and other biological systems have been reviewed in [22].
The non-ideal properties of the¯uorophores is the main limitation of thē uorescence techniques. In some cases, it is di cult to separate out the¯uorophore dynamics from the sample dynamics, at both single-molecule and macroscopic ensemble levels. Photodestruction of the excited-state¯uorophores, induced by the presence of molecular oxygen, is another limitation. Furthermore, the¯uorescence spectra often do not reveal su cient information for chemical identi® cation, other optical spectroscopies being better suited to this purpose.

Raman near-® eld microscopy
One of the optical techniques able to provide su cient chemical information for species identi® cation is Raman near-® eld spectroscopy. Besides the vibrational band-speci® city advantage , in comparison with¯uorescence, Raman microscopy does not have the bleaching problem.
The chemical contrast of Raman spectroscopy arises from the fact that the polarizability of a molecule is modulated by vibrational excitations [75]. The gross selection rule for IR absorption transitions is that the electric dipole moment of a molecule must change when the atoms are displaced relative to one another during the molecular vibration. The selection rule for a Raman transition states that the polarizability should change as the molecule vibrates [76]. The latter is a less restrictive condition than IR and therefore the Raman spectra are more complicated.
Compared with typical IR vibrational cross-sections of about 10 "(± 10 "' cm #, the Raman cross-section is extremely small : 10 $"± 10 #* cm#. The high laser intensities required to overcome the small Raman scattering cross-section can heat and therefore damage the most sensitive samples. The situation changes dramatically when the molecule resides in the vicinity of a rough metal surface or colloidal particle that supports surface electromagnetic waves (plasmons), such as gold or, more often, silver. In this case, surface-enhanced Raman scattering (SERS) occurs and the average enhancements reach 10'. When there is a resonance of the pump wavelength with a molecular electronic transition, another increase of up to 10& is expected, resulting in a total enhancement of 10"". There is a chemical enhancement and a ® eld enhancement. The latter accounts for the most of the increase in scattering. For a review of these enhancement origins see, for example, [77]. Even after this total enhancement, a scattering cross-section of 10 #!± 10 ") cm # seems to be too small to be used in NSOM experiments where the pump beam is limited to less than 10 mW average power by the damage threshold. However, Nie and Emory [78] and Kneipp et al. [79] used near-® eld microscopy to show that the forementioned enhancements are only ensemble-averaged values, and that approximately one out of a thousand silver colloidal particles exhibits intrinsic Raman enhancement factors of 10"%± 10"&. With such enhancements, the signal-to-noise ratio of the SERS spectrum of rhodamine 6G molecules adsorbed on the`hot ' particles was even better than the signal-to-noise ratio of the¯uorescence spectrum. It follows that the Raman-NSOM methods may have promise for chemical imaging, provided that surface enhancement is employed. This requirement of special sample preparation represents a limitation, which is often called upon with regard to the SERS spectroscopy of thin ® lms.
One important aspect of the Nie± Emory and the Kneipp et al. experiments is that they show that individual metal particles have tremendous potential as probes with high sensitivity. The ideal SERS near-® eld scanning optical microscope would be ® t with such a surface-plasmon resonant metal probe [60]. However, a complete understanding of the mechanisms of the enhancement is necessary before one can fabricate SERS probes for the near ® eld that could compete with the`hot ' particles found in silver colloids. EOE orts in this area have been made with the fabrication of a localized plasmon resonance probe [80] or by using small metallized polystyrene particles [81]. However, changes in the intensity and selection rules of the Raman spectral features have been recently observed [82] and seem to complicate the problem further.
Despite the small cross-section of the non-enhanced Raman scattering, evidence for submicron resolution by near-® eld Raman spectroscopy has been provided. Webster et al. [83] studied the residual stresses associated with local plastic deformations of a silicon wafer surface using an aperture near-® eld scanning optical microscope. The lattice deformation induced by a scratch shifts the Raman lines, which provides the way to map the residual stress across the scratch [83]. Near-® eld spectra of diamond ® lms and liquid interfaces were also recorded [25]. Resonant Raman near-® eld spectra where recorded on a polydiacetylene crystal in an earlier work [84]. However, probably because of the large exposure time required, a twodimensional image was not available when non-enhanced Raman scattering was used. Other studies took advantage of the very large polarizability of molecules like KTiOPO % to obtain actual Raman images [82,85]. Advantages of the Raman-NSOM over IR microscopy are the weak Raman scattering cross-section of water, which enables the technique to be used on natural biological samples, and the availability of light sources and much better detectors. In this context, we believe that the Raman near-® eld scanning optical microscope will ful® l its promise for chemical imaging provided that better probes, possibly based on SERS, are devised in the future.

Photothermal microscopy
Point-by-point spatial variations in the IR absorption of a surface can give rise to thermal spatial variations when the sample is exposed to IR radiation. The temperature variations can be measured in turn either by a miniaturized Wollaston wire resistive thermometer or by local thermal expansion. Hammiche et al. [86] set the basis of this technique by incorporating the sample and the scanning thermal probe into a FTIR spectrometer. The probe can be used to measure resistively the rise in the local temperature occurring when the IR wavelength matches at the same time the constructive interference condition of the Michelson interferometer and an absorption band of the sample (® gure 7) [87,88]. Alternatively, heat¯ow images can be acquired using a thermal feedback mechanism, which keeps the temperature of the heated probe constant during the scanning. In this case, the contrast is determined mainly by variations in the local thermal diOE usivity of the sample. Measuring the de¯ection of the Wollaston probe during pressing with a constant force while heating the sample locally enabled Hammiche et al. [87] to measure phase transitions on composite polyethylene terephthalate± resin samples and spatially discriminate between the two materials according to their diOE erent thermomechanical analysis traces. The photothermal approach was used to detect buried layers of polystyrene (50 l m thick) under polyisobutylene. The attainable spatial resolution by these techniques depends upon probe size and the thermal properties of the sample, as well as on the temperature modulation frequency. A quantitative analysis is therefore complicated, because of the thermal step. Estimates for the thermal diOE usion length in polymers seem to limit the resolution to 10 l m and above. Acquisition times of about 1000 s per spectrum are required to obtain a good signal-to-noise ratio on relatively thick polymer ® lms. Because of the long integration times no photothermal images are available at present. A related technique, based on thermal spectral emission was presented by Boudreau et al. [89] who mapped the temperature distribution by collecting in the near-® eld the IR emission from micron-scale conductors traversed by electric currents. The technique might ® nd use as a tool for detecting short-circuits or overheating areas in prototype ultralarge-scale integrated circuits [90].

Near-® eld absorption microspectroscop y at infrared wavelengths
Pioneering work in the area of IR near-® eld microscopy was done by Massey et al. [91] who demonstrated subwavelength resolution using far-IR emission (118 l m) from a methanol vapour laser to study the light coupling between two slits 10 l m wide scanned at diOE erent gaps. Scanned images with subwavelength resolution were reported for the ® rst time at near-IR wavelengths by Isaacson et al. [92]. Metallized hollow pipette probes from aluminosilicate glass were used in this work to map the near-® eld emission of a single quantum well laser at 868 nm. The images of the near-® eld modal pattern were acquired in the constant-height mode. Isaacson et al. gave evidence, with a spatial resolution of about 100 nm, for fabrication-induced strain areas and suggested the utilization of IR-NSOM to deduce directly the material growth pro® le from the measured near-® eld emission pattern. Also in the near-IR, and also on emitting quantum well heterostructures , but this time at 1.54 l m, Ben-Ami et al. [93] utilized a bent ® bre probe to obtain simultaneous AFM and emission maps of the active region of the laser. De Fornel et al. [94] used a photon tunnelling con® guration to obtain near-IR images (k¯1.3 l m) of silicon oxide samples and to study the height dependence of the tunnelling.
These early studies at wavelengths greater than visible wavelengths were not intended to take advantage of the chemical speci® city of the IR radiation. This chemical aspect was put forward for the ® rst time by Piednoir et al. [95], who announced in 1992 the ® rst attempt to build a tuneable near-® eld microscope working in the mid-IR range. The instrument was built around a¯uoride glass ® bre and it worked in the photon tunnelling mode (frustrated total internal re¯ection). However, the ® rst results, showing IR contrast on silica bands deposited on silicon, at 4 l m wavelength, were reported only in 1995 by Piednoir et al. [96]. These workers carried out local mid-IR near-® eld spectroscopy through a¯uoride ® bre in the same experimental con® guration, on a diazoquinone resin (600 nm thick, deposited on silicon). Characteristic bands of the local spectrum, with the tip collecting light from a single spot, reproduced well the spectrum of the bulk material. The tuneable light source was in this case a free-electron laser (10 ) W l m #). Piednoir and Creuzet [97] investigated the spectroscopic imaging capabilities of their system on a sol± gel solution of latex spheres, tuning the wavelength on and oOEthe Si± H absorption band. No clear diOE erence in contrast between the two images could be attributed to the IR spectral signature, owing to the weak extinction coe cient and to the prevalence of topographical artefacts. The spatial resolution in these mid-IR experiments was about 1 l m. CO # lasers were demonstrated to be a suitable light source for IR-NSOM by Nakano and Kawata [98] who used an uncoated ZnSe pyramidal tip in the illumination mode to image polystyrene particles 6 l m diameter.
From these early experiments it soon became obvious that the technique of IR-NSOM is confronted by several problems, speci® c to the long IR wavelengths. These problems include the inherently ine cient transfer between bound and propagating waves, detection problems and light source problems. In the following we review the available solutions to these di culties.

Infrared probes 3.1.1. Aperture ® bre tips
We shall begin our discussion on IR probes with the aperture type, since historically they were the ® rst to be used. The silica glass ® bres are ideal for the visible near-® eld scanning optical microscope since they are¯exible, they are available in a variety of types (single mode, multimode or polarization preserving), and a large amount of work on ® bre pulling and etching has been carried out. Unfortunately, the silica ® bres cannot be used beyond 2.2 l m because of the presence of molecular absorptions (especially OH) in the mid-IR. To the best of our knowledge, there are   four types of ® bres, which, in principle, are candidates for mid-IR applications and are commercially available :¯uoride glass ® bres, chalcogenide glass ® bres, sapphire (single-crystal) ® bres, and hollow silica guides [99]. Their characteristics are summarized in table 1 and ® gure 8. The hollow silica guides and sapphire ® bres have major limitations for IR near-® eld scanning optical microscope ® bre tips. The hollow guides strongly attenuate the radiation for inner diameters less than 200 l m. This prevents their use for near-® eld aperture fabrication. Sapphire cannot be pulled reliably since it is a crystalline material, nor can it be etched and cleaved by the same techniques as for the glass ® bres because it is chemically inert.
Fluoride ® bres from¯uorozirconate glasses can be easily pulled into tips owing to their low glass transition temperatures. They have fewer losses than any other material transmitting beyond 3.0 l m. They were used for tips fabricated by conventional pulling by Piednoir and co-workers [95± 97] in their early experiments and for doubletaper tips by dynamically controlled pulling, by Dragnea et al. [46,100]. The last method gives tips with a radius of curvature of about 100 nm, and an aperture of 200± 300 nm as measured by electron microscopy (® gure 9). The tip transmission at a wavelength of 3.0 l m is 10 %± 10 &, which is remarkable for a wavelength ® ve to six times the visible wavelengths, considering the scaling power law of section 2.3. The Figure 10. Power losses in the taper of a ® bre probe for diOE erent wavelengths, along the same ® bre taper, as a function of the distance prior to the tip. Note the early cut-oOEfor the IR wavelengths. Adapted from calculations in [30].
high throughput allows image acquisition at scanning rates of 30± 50 ms pixel ". The transparency of the¯uoride ® bre at visible and ultraviolet (UV) wavelengths makes optical diagnostics of the aperture particularly easy. In our laboratory a visible laser is focused on the cleaved end of the ® bre, while a strong microscope objective projects the image of the circular aperture on a charge-coupled device camera. If the aperture diameter is less than the resolution limit of the microscope objective at that particular wavelength (350 nm, in our case), the image is that of an Airy disc. If leakages through the coating or aperture asymmetry occur, they are easily captured as distortions in the diOE raction pattern (Airy disc) of the aperture, and the tip has to be discarded. A more elaborate method of far-® eld diagnostics for tapered probes has been proposed by Obermu $ ller and Karrai [101]. Chemical etching of¯uoride ® bres or ion-beam fabrication have not been reported, yet. The¯uoride tips have two disadvantage s : they are brittle and di cult to handle, and they are limited in transmission to about 4.5 l m wavelength, while many molecular vibrational bands are located beyond this wavelength. Chalcogenide ® bres transmit down to 11 l m wavelength, and they are more¯exible. Unger et al. [102] published a thorough study of chalcogenide ® bre etching by the Turner method and reported transmissions of 10 '± 10 & for apertures of 1± 5 l m# area, at 6.01 l m wavelength. The ® bres were completely coated with gold (190 nm thick), and the aperture was subsequently fabricated by polishing the apex with an abrasive surface. Although promising from the point of view of the throughput and reproducibility, the Turner method has its de® ciencies related to the rough etched surface and pinhole formation in the metal coating. However, Talley et al. [103] found that arsenic selenide ® bres etch su ciently smoothly to allow good-quality aluminium coatings and aperture formation.
Chalcogenide tips were also obtained by pulling [104] and, at the time of writing, the pulled tips seem still to be a more popular choice [105,106].

Apertureless tips
The power losses in the taper of a ® bre probe increase exponentially with the wavelength [30] (® gure 10). This limitation represents the bottleneck for the utilization of IR metal-coated probes at long wavelengths. To overcome this obstacle, Lahrech et al. [107] followed by Knoll and Keilmann [108], replaced the ® bre tip by a bent, electrochemically etched tungsten apertureless tip [107,108], while using a CO # (10.6 l m) laser focused at grazing incidence on the tip± sample interaction region (® gure 11). The tip-to-surface distance was modulated (about 100 nm) as in noncontact (tapping ) AFM. The image represents the ac amplitude of the scattered IR radiation.
Although true optical contrast was present in these ® rst images, it was di cult to assert what fraction of the total contrast was due to topographical artefacts. One must keep in mind in this type of experiment that the scattering is not coming only from the near ® eld, but also from remote areas of the probe. For particles small compared with the wavelength of light, the scattering e ciency factor (which is de® ned as the ratio between the scattering cross-section and the geometrical area) is proportional to (a}k)%, where a is the radius of the spherical particle and k is the wavelength [109]. Consider the probe as composed of two parts : a 100 nm sphere at the apex and a 10 l m sphere for the rest of the probe. In the absence of a surface, the ratio between waves scattered by the tip apex, which would probe the near ® eld and by the rest of the probe is 10 ). However, enhancement of the ® eld between the tip apex and the surface may increase the near-® eld contribution to the scattered light by a factor of 10± 1000, depending on the tip geometry and material [60,62]. It follows that the small near-® eld contribution has to be extracted from the background either by using a perfectly¯at sample with optical spatial¯uctuations only [110] or by putting in greater evidence the optical contribution, for instance by using a spectral contrast mechanism that compares images at two diOE erent wavelengths [46,111]. Another solution would be to apply the constant-height mode scanning or pseudo-constant-heigh t mode scanning [59], depending on the surface roughness. This possibility has not yet been implemented. The eOE ect of the scattered light from the tip shank can be reduced by applying a heterodyne detection in which a low-frequency modulation of the tip or sample position is superimposed on the high frequency of the cantilever [63,67,68]. The sensitivity was further improved by Zenhausern et al. [63] who used an interferometric set-up to detect small refractive index¯uctuations on the surface of a cover-glass sample.
There are not su cient experimental results at this time to compare safely the e ciency (sensitivity and resolution) of the ® bre-based and apertureless microscopes working at long wavelengths. However, the trend seems to be in favour of the apertureless con® guration, primarily because of the surface ® eld-enhancement eOE ect advantage of the apertureless tips and of the taper attenuation problem of the ® brebased IR near-® eld scanning optical microscope, with the added advantage of broad wavelength acceptance of the apertureless method.

Other probes for infrared techniques
In addition to the above mainstream probes, a number of diOE erent other probe approaches are worth mentioning. Grober et al. [112] suggested the fabrication of a planar antenna at the end of a cylindrical waveguide using lithographic techniques. They demonstrated on a scaled system, at radio frequencies, that the coupling to the near-® eld has an e ciency of about 2 % for a spatial con® nement (resolution ) of about k}10.
Active probes may also be fabricated by integration of vertical cavity surfaceemitting lasers into AFM cantilevers. Recently, a near-® eld source of 980 nm wavelength on a gallium arsenide cantilever was reported [113].
Quartel and Dainty [114] fabricated polycrystalline silver halide pyramidal tips by sharpening the ® bre with a sharp blade. With this instrument they sampled, in the photon tunnelling mode, the evanescent ® eld distribution above the surface of an anisotropically etched silicon water. In the same category of pyramidal probes, there are ion-milled slit-shaped apertures on metal-coated silicon tips for an IR near-® eld scanning optical microscope [115,116].
The concept, put forward by Keilmann [117] in 1991 that coaxial near-® eld probes should give a better throughput since they have no cut-oOElike tapered coated ® bres, was realized by fabrication of hollow metal tips on a silicon cantilever and ion-beam deposition of a rod-shaped inner conductor [118]. However, no NSOM images taken with these tips are available to date. Extremely high calculated e ciencies were obtained for the coaxial tips by ® nite-diOE erence numerical integration. According to Leinhos et al. [118], a coaxial tip with a 200 nm aperture transmits 5 % of the incident radiation (10 l m wavelength).
An exotic approach, which avoids the di culties related to both ® bre and apertureless probes but is limited in spatial resolution, is the remote generation of a transient ® eld probe in a visible-pump± IR-probe experiment on a semiconductor surface, which can play the role of a substrate for a thin-® lm sample [119]. The transient probe has the size of the diOE raction-limited spot of the pump beam and can be scanned the same way as a beam in a scanning confocal microscope. The resolution is therefore limited to a few hundreds of nanometres. Special preparation of the sample and high pump intensities are also of concern.
3.2. Infrared detection IR detectors, viewing ambient temperature surroundings, have typically 10%± 10& less detectivity per photon than the average photomultiplier working at visible wavelengths. This limits the sensitivity of an IR near-® eld scanning optical microscope to relatively weak signal diOE erences, as for example those due to vibrational absorptions in thin organic samples. For comparison, the absorption coe cient of a direct-gap semiconductor at wavelengths above the gap is of the order of 10% cm ", while the absorption coe cient of the CH stretch band in poly(methyl methacrylate ) (PMMA) is about 50 cm ".
Unfortunately, there are very few examples at this time of sensitivity limits in the IR NSOM area, since most of the studies focused on spatial resolution performance. Because there is a trade-oOEbetween resolution and sensitivity, which is obvious in the case of the aperture microscope but also present in the apertureless case [120], the highest resolutions were reported on samples with the highest contrast.
The experimental set-up in our laboratory, which is based on an aperture¯uoride ® bre microscope, has a spectral range from 2 to 5 l m, and a sensitivity of 0.05 % in transmission, with 300 nm spatial resolution at 3 l m wavelength (® gure 12) [121]. We use two InSb detectors with an active area of 0.015 mm# and a cooled ® eld of view of 10°. The small active area has two advantages : ® rst, the spectral detectivity is greater for small area detectors and, second, the detector itself serves as a pinhole in a confocal detection set-up, which minimizes the stray light background [122]. The cold ® eld of view reduces the thermal background radiation. The calculated detectivity is in this case 0.9¬ 10"# cm Hz" /# W " (at 1 kHz). The electronic bandwidth of the detector is from 0.1 Hz to 1.5 kHz. A reference and a signal detector remove the Ar + laser noise. Without subtraction, the laser noise would limit the sensitivity to about 5 % . With these characteristics, and a transmitted IR power of about 100 nW, the detectivity of our set-up is limited by the pre-ampli® er noise.

Infrared light sources
High-spectral-brightnes s collimated sources are needed in both the aperture and the apertureless IR near-® eld scanning optical microscope because of either the small acceptance of the IR ® bre or because of the minimum diOE raction-limited spot requirement. The IR output of a glowbar source, which is commonly used in FTIR instruments, is distributed over a broad range of wavelengths but emits in all directions. Even for classical FTIR microscopy applications, the conservation of e; tendue limits the attainable spatial resolution, and better sources such as laser diodes can replace, in some cases, incoherent thermal sources [3]. To date, laser diodes, colour-centre lasers, CO # lasers and free-electron lasers have been used in IR-NSOM experiments. The main limitation of the ® rst three of these lasers is their reduced tuneability of 10± 15 % from the central wavelength. Only the excellent characteristics of the free-electron laser, with its spectral range between 2 and 20 l m [123,124], short pulses (picosecond ) and high average power (0.1± 1 W) could support, at present, the realization of an IR near-® eld microspectrometer that would surpass the spatial resolution of a FTIR microscope but cover the same spectral range. Unfortunately, Figure 12. Schematic diagram of a diOE erential IR-NSOM apparatus for absorption microspectroscopy. The reference and the sample signals from the pinhole detectors R and S are fed to the diOE erential input of a lock-in ampli® er, locked on the chopper frequency (1 kHz). The numerical aperture of the IR objective is important in determining the contrast mechanism, for example, by limiting the angular range to discriminate pure absorption contrast from index of refraction contrast mechanisms. The spatially ® ltered He± Ne laser beam is used as a pilot beam for alignment, as well as for confocal re¯ection microscopy (the signal detector is replaced in this case with a silicon photodiode) [100].
free-electron laser facilities, such as synchrotron radiation facilities, are relatively few and costly. A cheaper alternative is a tabletop picosecond optical parametric oscillator (OPO), which has a modest tuneability and a lower power output. IR OPOs are better than IR laser diodes or F-centre lasers (FCLs), in terms of both spectral range and output power. Figure 13 gives a comparison of various IR sources discussed here.

Infrared near-® eld scanning optical microscopy applications
We review in the following sections a number of applications of IR-NSOM in semiconductor surface characterization, polymer blend characterization, thin-® lm polymer photolithograph y and single-living-cell IR microspectroscopy. We shall end this section with several examples of extensions of the IR near-® eld to terahertz and microwave applications.

Semiconductor characterization
Spatially resolved measurements of the lifetimes of excess carriers on silicon surfaces provide a way to image surface and subsurface defects. When the nature of the defects is known, the time-resolved mapping of the non-equilibrium carrier population density gives insight into the interplay between carrier recombination, transport and  Nechay et al. [127] pushed the time resolution limit down to 250 fs with a near-IR (850 nm wavelength) pump± probe experiment on patterned GaAs}Al x Ga " x As heterostructures fabricated by ion implantation. They found that the lateral carrier diOE usion across the ion implantation features plays an important role in carrier relaxation. Implanted semiconductor patterns were also imaged at longer IR wavelengths (10.6 l m ; CO # laser) in a demonstration of pure optical contrast achieved by re¯ection apertureless IR-NSOM experiments [110] (® gure 15).

Polymers
The ® rst IR near-® eld attempt to address molecular processes by taking advantage of the`® ngerprint ' region of vibrational molecular spectra was a study of the poly(ethylene glycol) diOE usion at the interface of a polymer laminate [128]. Although it was not clear what the actual spatial resolution was that was involved in the diOE usion coe cient determination, this early work, employing a modi® ed FTIR microscope, clearly announced the promise of IR-NSOM for the polymer science.
DiOE usion in micropatterned polymers has great interest from two points of view. First, diOE usion, together with other physical properties of a polymer, changes when the polymer becomes spatially con® ned [129]. It is of fundamental interest, therefore, to apply methods capable of local investigations of the properties aOE ected by structure. Second, diOE usion is important in many polymer applications in which heterogeneous phases are present, from medicine to plastics technology. One important technological example where diOE usion plays a role is latent image formation in chemically ampli® ed resists for deep-UV lithography [130]. The chemically ampli® ed lithographic technique involves the following generic steps : (1) A polymeric resist ® lm composed of a polymeric host and a photoacid generator (PAG) is cast on the semiconductor wafer. (2) A photochemical image is transferred to the resist through short-wavelength (deep-UV) exposure of selected areas via projection. Photolysis of the PAG within the exposed areas creates a latent image of the pattern. (3) The image is intensi® ed (chemically ampli® ed) by a thermal step (postexposure bake) in which the acid is involved in an autocatalytic reaction that results in the deprotection of the polymer host to a solvent (positive resists). (4) The deprotected polymer is removed by dissolution, the bare semiconductor substrate is selectively etched, and the remaining polymer is removed.
Several factors are believed to play a role in the spatial resolution of chemically ampli® ed resists : the diOE usion of the acid during the post-exposure baking step, the catalytic chain length, and the random¯uctuations in the width of a resist feature termed`line-edge roughness ' [131,132]. The latent image characterization requires methods that have chemical speci® city and spatial resolution, which also have the ability to probe beyond the sample surface. Absorption IR NSOM is ideal for this purpose.
A ® rst study of deep-UV patterned and post-exposure-bake d poly(tert-buty l methacrylate) (PTBMA) ® lms containing triphenylsulphonium hexa¯uoroantimonat e PAG by IR-NSOM showed that, if the IR wavelength of the FCL is tuned on and oOE , the OH stretch absorption band (3 l m) of the poly(methyl acrylate acid) (PMAA), which is the deprotected form of PTBMA, chemical images of the latent pattern can be generated [46]. Initial near-® eld IR images contained contributions not only from absorption but also from the real indices of refraction, which were diOE erent for PTBMA and PMAA. The refractive index contribution was subsequently reduced, by eliminating the highest angular parts from the light cone of the transmitted light (® gure 16) [100]. The maximum optical resolution obtained in this work was about 280 nm, at 3 l m in the IR, while the topographi c (shear-force ) images showed a resolution of about 100 nm. In addition to these spectral diOE erences, the same contrast was obtained in the constant-ga p mode and in the constant-height mode and therefore the presence of image artefacts due to topography was ruled out. The surface topography was the result of shrinkage (50± 80 nm) of exposed regions during the postexposure bake step. The exposure dose dependence of the acid-catalysed chemistry was studied on line± spaced patterned samples. At low exposure doses, negligible shrinkage of the exposed zones occurs, but IR contrast persists (® gure 17) [121]. It was also found that the deprotection yield is diOE erent for narrow-line± spaced patterned  samples than for blank exposed samples. It is believed that the main reason for this is edge interference eOE ects occurring in the deep-UV exposure step, when metal proximity masks are used to project the pattern [100]. IR near-® eld spectroscopy made a signi® cant step forward with a recent demonstration of a near-® eld microscope based on a broadly tunable infrared light source producing ultrafast pulses with a full width at half maximum (FWHM) bandwidth of 150 cm " and tunable between 4000 cm " and 833 cm " (2.5 l m to 12 l m) [133]. In this work, Michaels et al. used a single-mode, chemically etched uoride-glass optical ® bre and an infrared focal plane array detector to acquire broad band near-® eld spectra of the aliphatic C± H stretching region of a polystyrene ® lm, in a single spatial location.
A breakthrough for the chemical mapping of polymer blends at submicron resolution was achieved by Knoll and Keilmann [108] who demonstrated IR vibrational band contrast at around 10 l m wavelength on polystyrene particles embedded in PMMA (® gure 18). The measured contrast value was approximately one order of magnitude greater than that calculated using the eOE ective polarizability model. This result was interpreted as evidence for surface-enhance d IR absorption [111]. If the interpretation of Knoll and Keilmann is correct, then their result opens the way to the realization of a new generation of apertureless near-® eld scanning optical microscopes, which would take advantage of the surface plasmon resonances of the near-® eld probe to enhance by a 1000 times the absorption sensitivity [134].

Infrared near-® eld scanning optical microscopy of liŠ ing cells
There is much activity in medical research on the IR spectroscopy of entire cells and tissues. Techniques have been demonstrated to be able to follow the large variations in the DNA spectral features of cells as they undergo their natural division cycle [135]. These results suggest that in some cases the spectral changes between normal and neoplastic (cancerous ) tissue are due to an enhanced signature of DNA in neoplastic tissue [136]. The complexity of the IR spectrum of biological samples makes the wavelength assignments the main di culty for FTIR microscopy utilization in biosciences. However, this problem can be overcome, at least in some cases, owing to improvements in spectra analysis, for example by using the near-IR and mid-IR spectra and the generalized canonical correlation analysis [137]. Moreover, there are well-de® ned bands associated with diOE erent components of the cell, such as the amide bands for the nucleic acids, the C± O and C± C stretch bands of glycogen from the muscle cells and liver, or the C± O stretch band of the phospholipids making up the cell membrane lipid bilayer.
At ® rst sight, the omnipresence of water in biological samples presents a di cult obstacle to overcome for IR microspectroscopy, but a proximity probe such as an IR near-® eld scanning optical microscope tip reduces the water layer between the probe and the sample and is capable of collecting data even in aqueous solution, as has been demonstrated by Hong et al. [125]. These workers used a tuneable free-electron laser beam and an uncoated¯uoride tip to measure the thin extensions, called lamellopodia, of human ® broblasts, cells that are involved in repair processes in the body. The reason for an uncoated ® bre tip (¯uoride) was to avoid the strong interaction between the metal coating and the cellular membrane. However, the spatial resolution did not signi® cantly deteriorate because of the strong IR absorption in the aqueous medium outside the ® bre, at the wavelengths in use. The images were acquired in the constantheight mode to avoid topographic artefacts. Tuning the laser from the amide band (6.25 l m) to the lipid band (5.76 l m) gave evidence for the high contrast associated with the enhanced presence of lipids in the lamellopodia (® gure 19). The measured absorption was 250± 500 times greater than expected from a single lipid bilayer. According to Hong et al. [125], the enhanced IR absorption found in these near-® eld studies lends support to the idea that lipid molecules are involved in substantial membrane modi® cations associated with cell motility.

Terahertz near-® eld imaging
Far-IR radiation, or the terahertz (THz) region, became of considerable interest for spectroscopy in the recent years [138], and THz imaging was already used very early to map doping pro® les of semiconductor wafers or the water content of biological samples [139]. The fact the radiation is pulsed, being generated in an optoelectronic emitter by a subpicosecond laser pulse, brings to the fore the interesting capabilities for time-domain spectroscopy. Keilmann' s [140] work is the precursor of THz near-® eld studies, with images generated on semiconductor samples. However, the ® rst images that convincingly proved better than k}4 resolution (k¯220 l m) were presented by Hunsche et al. [141] in 1998, with a very similar apparatus to a ® bre-based aperture near-® eld scanning optical microscope. The hollow tapered waveguide had at the end a circular (100 l m) or elliptical aperture (50 l m¬ 80 l m). The spectrum of transmitted pulses exhibited the high-pass ® lter characteristics of the subwavelength aperture. The transmission e ciency was 1}130. For larger apertures, but lower frequencies, the spectrum did not exhibit a sharp cut-oOE signature. A diOE erent approach for the near-® eld probe, namely a tip with a high index of refraction on top of a subwavelength aperture, recently enabled Mitrofanov et al. [142] to push the spatial resolution limits of microscopy to 60 l m, with a coupling e ciency of 10 $.

Perspectives
The selected studies mentioned above give an overview of several directions where IR NSOM could make a breakthrough. There are still instrumental issues related to the probe fabrication and IR source brightness and tuneability.
In applications requiring the aperture ® bre probe, for instance when a greater probe depth into the sample is necessary, the tube-etch method should be extended to IR-transmitting ® bres for better throughputs . The sensitivity can be increased by using heterodyne (interferometric ) techniques, as has already been demonstrated in the visible range. Apertureless microscopes that take advantage of the surface plasmon resonance of the metal probe may also bring an important enhancement in sensitivity, and this can open the door to surface-enhanced IR absorption microspectroscopy. Besides free-electron lasers, OPOs have a wide range of IR wavelengths accessible. Low-cost low-power alternatives would be lead salt laser diodes. However, their wavelength range limitation makes them no match for the previous two sources and for the spectral range of a classical FTIR. The future applications would have only to bene® t from better-designe d samples and from numerical simulations for contrast assignment. This way, near-® eld analytical IR microscopy will be the reliable approach still awaited for many problems of various ® elds of science and technology.
One direction of study, which would clearly bene® t from the state-of-the-ar t features of IR-NSOM, is the study of water diOE usion in polymer blends and thin polymer ® lms. We recently undertook this direction in our laboratory with the construction of an IR near-® eld scanning optical microscope enclosed in a vacuum chamber. Specially designed polymer ® lm samples and controlled humidity will help us to study at submicron levels issues such as the microscopic transition between linear diOE usion (Fickian) and nonlinear diOE usion (case II), or to follow in real time the diOE usion front for bound and unbound water molecules, since they have diOE erent IR absorption bands. The same technique can be used to study the link between swelling and diOE usion on polymer blends with microphase separation. The IR-NSOM information will be used to explain the build-up of macroscopic properties of polymer blends from microscopic properties.
Another important direction concerns the IR spectroscopy of cells and tissues. The ability of IR-NSOM to analyse minute amounts of tissue for diOE erent chemical compositions without the use of stains or speci® c probes could be of great importance for the early diagnostic of neoplastic tissue. Diem et al. [135] reported the spectral variations of DNA during the cell division cycle. It is supposed that, during the diOE erent phases of the division, the DNA passes from the dense packing in nucleosomes to a more diOE use distribution, which allows observation by FTIR spectroscopy. Based on this hypothesis, Diem et al. built a relation between the increase in the DNA IR spectral contributions of cells and the progress of the cell from normal to pre-cancerous and cancerous states. Mapping the DNA concentration in the cell could check the hypothesis of DNA intermittent delocalization. The key cell components are the nucleus (about 3 l m) and the nucleosomes (about 800 nm), both accessible from the point of view of today' s best instruments.