Photothermal AFM-IR spectroscopy and imaging: Status, challenges, and trends

AFM-IR is an Atomic Force Microscopy (AFM) based technique that combines the high spatial resolution of AFM with the chemical identification capability of infrared (IR) spectroscopy. The development of this technique is one of the most impressive innovations from the last decade and has led to a remarkable success for this new tabletop nano-spectroscopy tool. There exist now over 250 systems across the world forming a large nano-infrared spectroscopy community, i.e., user base, of at least 2000 people, according to an estimate by Bruker. This community produces about 70% of the publications involving nano-infrared spectroscopy analysis with 65 papers in 2020, which covers a huge diversity of applications as shown by several review papers already published on the AFM-IR technique.^1–9 Moreover, the AFM-IR technique has been widely embraced by industrial companies due to its ease-of-use, low requirements on sample preparation, simplicity, reliability, and direct correlation to FTIR spectra without the need for modeling or data post-processing.

The first AFM-IR experiments have been performed in 2005 at the CLIO (center laser infrarouge d'Orsay) facility center at Université Paris-Sud,¹⁰ by imaging the amide I band distribution inside a single bacterium and by obtaining its IR fingerprint spectra with a resolution of around 100 nm. The idea to use photothermal excitation (i.e., light absorption that results in sample heating) in order to measure an infrared spectrum with an AFM tip has been inspired by the Mirage¹¹ experiment already installed at CLIO and offered to users in 2003. Indeed, the main principle of Mirage is to illuminate the sample with a tunable infrared source at a sample's specific absorption band to induce the photothermal effect and to detect it by probing the sample surface with the reflection of a visible laser. If the sample absorbs, then the air directly on top of the sample gets heated and its refractive index changes, inducing a deviation of the visible laser beam reflection. The spatial resolution of such a setup is limited by the diameter of the visible laser spot.¹² The idea of AFM-IR is not to detect the induced photothermal heat but to detect the thermal expansion with an AFM tip, using the high sensitivity of the AFM system to vertical height changes of the sample (tens of picometers) and its excellent lateral resolution (10–20 nm). This first implementation at CLIO was not a marketable tabletop system as the infrared source was a free-electron laser but the concept to use the photothermal effect to realize a local absorption measurement with a resolution that goes far beyond the diffraction limit was demonstrated. These first results have been the beginning of a constant technical development to increase the performance of this technology.

In early 2010, OPO (optical parametric oscillator) lasers became available with a tunability from 4000 to 1000 cm⁻¹, covering a large part of the mid-infrared spectrum. At that time, the first commercial systems were launched by a start-up company in the US (Anasys Instruments). Despite their attractiveness to infrared users, the performance of the first systems was limited by the laser technology delivering a local full spectrum within 15 min and a single chemical map within 30 min with a sensitivity and resolution around 50–100 nm at 4–10 cm⁻¹ spectral resolution.^13,14 The next evolution of the AFM-IR system was driven by the fast development of infrared QCL lasers (quantum cascade lasers).¹⁵ These lasers cover the IR fingerprint spectral region (∼900–2000 cm⁻¹) with a narrower spectral range per single QCL chip (typically 200–300 cm⁻¹) but achieving a 0.1 cm⁻¹ spectral resolution, a scan speed shorter than 1 s for a 1000 cm⁻¹ spectral range in fast spectra acquisition mode, and a repetition rate of the laser tunable up to 2–3 MHz, unlike the fixed 1–10 kHz of the OPOs. The tunable repetition rate allowed to drive a specific mode of the AFM cantilever resonantly with the sample expansion, producing resonance enhanced signal amplification and leading to a significant increase in sensitivity. In 2014, this sensitivity increase enabled the detection of a self-assembled monolayer, and a chemical imaging with less than 25 nm lateral resolution.¹⁶ These resonance enhanced AFM-IR experiments were still based on AFM contact mode. It restricted the measurements to hard and non-adhesive samples. In AFM imaging, the most convenient mode to overcome such limitations is the tapping mode. In 2018, the first experiments using tapping AFM-IR on soft samples were published showing the great potential of the photothermal technique.^17–19 Despite these technological advances, AFM-IR still continues to evolve and improve as new modes and concepts emerge.

This article is organized as follows: Sec. II describes the principle of AFM-IR, Sec. III reviews the main operating modes, common artifacts, and fields of applications, and Sec. IV presents some emerging trends.

The AFM-IR implementation is relatively straightforward as it uses an infrared tunable pulsed laser to induce a photothermal effect in the sample and an AFM to detect the resulting mechanical expansion via the cantilever tip that is in contact with the sample.

The photothermal effect describes the temperature increase induced by radiation absorption, while the photoacoustic effect is the generation of sound waves (material waves) due to radiation absorption. Photothermal or photoacoustic effects have been extensively used as an alternative technique in spectroscopy as they represent a measurement directly proportional to the absorbance [abs(σ)], thus comparable to classical spectroscopy,^20–22 

where A is related to the thickness of the sample and constant, σ represents the wavenumber, and κ(σ) the imaginary part of the refractive index of the sample [$n~(σ)=n(σ)+iκ(σ)$].

In AFM-IR, the signal measured by the AFM probe is caused by the photothermal effect. The infrared source used is a tunable one. If the selected wavelength corresponds to an absorption band of the sample, part of the excitation beam energy is absorbed by the sample and directly converted into heat. This happens only inside the absorbing region so that the temperature rise is localized. To relax the excess of stress induced by the local increase of temperature, the absorbing region expands and pushes the cantilever tip in contact with the sample surface (Fig. 1). The expression for the thermal expansion cycle u(t) can be written as²³ 

where α is the thermal expansion coefficient of the sample, K includes sample parameters, ΔT is the increase of temperature, u₀ is the maximum thermal expansion, and R(t) expresses the temporal behavior of the expansion.

As described by Fig. 2, the typical thermal expansion cycle is composed of two phases: the first phase is the expansion itself (from 0 to t_p) driven by the laser pulse duration (usually between 10 and 400 ns depending on the laser type). As the laser heats up the sample, its volume increases and surface pushes against the tip. The second phase appears when the laser illumination stops (after t = t_p). At this time, the heat accumulated inside the sample diffuses outside. The sample volume declines exponentially with a characteristic relaxation time (defined as τ_relax in Dazzi et al.)²³ related to the sample size and the thermal properties. In that case, the sample surface pulls the tip.

A relation between the thermal expansion and the local infrared absorbance can be established,²³ 

where G contains all optical and geometrical parameters of the sample, α is the thermal expansion coefficient, I_inc the incident laser energy, ρ the sample density, C_p the sample heat capacity, and t_p the duration of the laser pulse.

Equation (3) confirms that photothermal- and photoacoustic-based techniques can directly probe the imaginary part of the refractive index of the sample which is related to the absorption from Eq. (1). Consequently, signals obtained from AFM-IR can be directly compared to the ones obtained in FTIR spectroscopy, without any data post-treatment.

The thermal expansion cycle is transmitted to the cantilever through the tip. Its typical time scale is usually fast enough (a few μs considering the kind of sample commonly studied)²⁴ to act as a shock for the cantilever (which has a typical response time around 2–20 μs corresponding to the period of the modes) and kicks it. This kick generates oscillations of the cantilever that are detected by the deviation of the laser diode reflection on the AFM 4 quadrant photodiode [Fig. 3(a)].

As the tip is in contact before the thermal expansion, the deflection of the cantilever is the superposition of all the contact resonance modes resulting from the ringdown [Fig. 3(b)] and can be expressed as¹ 

where A contains all optical, thermal, and mechanical parameters of the sample and cantilever, P_n corresponds to the tip–surface transfer coefficient for each mode, g_n represents the deformation of the cantilever mode, f_n the frequency of the mode, Γ the damping, x the axis related to the cantilever length, L the cantilever length, and u₀ the maximum thermal expansion which is proportional to the absorbance [Eq. (3)].

Finally, measurements of the AFM deflection signal through the photodiode during the thermal expansion of the sample give access to the local absorbance. By recording the maximum amplitude of the deflection ringdown or its FFT amplitude at each wavenumber of the laser, it is possible to generate infrared spectra that are comparable to FTIR spectra as the local absorbance is directly transduced by thermal expansion on the AFM tip. Consequently, AFM-IR can be seen as a photoacoustic technique because the measurement by the cantilever is directly linked to a mechanical displacement and not an increase of temperature.

As the thermal expansion is not temporally resolved by common cantilevers, it is difficult to obtain a good understanding of the behavior of the tip on the sample surface during the expansion cycle. To the best of our knowledge, Chae et al. study is the only one that measured the relaxation time of the sample in a comparable setting but with a special resonator characterized by a response time of around 10 ns which is below the thermal expansion duration.²⁴ In that particular case, this probe can follow the surface motion during the relaxation time and perfectly depicts the thermal diffusion while staying in contact with the sample during the entire cycle.

For other cantilevers, the time response is generally three orders of magnitude slower and slower than the thermal expansion duration. As a consequence, with such conventional or classic cantilevers, the thermal expansion cycle is felt like a transient deformation of the cantilever transmitted by the tip and the motion is quite different.

Figure 4 represents a typical chemical mapping of triacylglycerol vesicles inside Streptomyces bacteria obtained with 1 kHz OPO with 10 ns pulse. In this example, the time between pulses is long enough for a full recovery of the thermal equilibrium. The thermal diffusion length estimated by finite element methods using the thermal parameter of polymer for 500 nm vesicles is around 3 μm. Nevertheless, the vesicles’ size detected by AFM-IR is quite smaller—around 100 to 500 nm—demonstrating that the tip detection is based on sample expansion and not on the temperature change. These common AFM-IR results with the 1 kHz OPO laser finally show that the thermal diffusion is not a limiting factor affecting the resolution of AFM-IR images. A comparable study was conducted by Quaroni.²⁵ The particular conditions used in our work here are different from those he studied, as the AFM-IR signal was obtained with a repetition rate corresponding to a resonant mode of the AFM.²⁵ Nevertheless, in Quaroni's publication, the resolution limitation in the case of resonance enhanced mode is still dominated by the mechanical effect of the expansion (related to the sample size) and not by the sample thermal diffusion which is in agreement with the results obtained here.

The AFM-IR setup can be realized in two different configurations of infrared laser illumination with distinct advantages and drawbacks.

Historically, the first illumination configuration is a bottom-up illumination [Fig. 5(a)] where the laser passes through an infrared transparent prism (ZnSe, Zns, and CaF₂). In this case, the sample on top of the prism is illuminated in a total internal reflection geometry with an evanescent wave at the prism/sample interface. This specific configuration presents advantages that can be decisive, depending on the application. The main advantage is the possibility to use any kind of cantilever, specifically uncoated ones, as the infrared light does not interact with it, avoiding spurious absorption within the cantilever itself.²⁶ The main drawback of the bottom-up configuration is the maximum sample thickness that can be probed. As the illumination is based on infrared evanescent waves, the distance from the prism surface to the top sample surface must be smaller than about 1 μm.²⁷ Moreover, the sample should be compatible with the prism material properties, which limits the range of suitable samples. The second possible configuration is a top-down illumination [Fig. 5(b)]. It is a more versatile one as there is no limitation on the sample thickness. In addition, an IR-transparent substrate is not always required. Removing the thickness limitation and the requirement for a specific substrate greatly facilitates the sample preparation, i.e., a larger variety of samples become compatible. The main drawback of this illumination configuration is that the IR laser spot directly irradiates the cantilever, and as a consequence, must be metal coated (e.g., Au or Pt) to shield the cantilever against absorption and provide field enhancement. The necessity to use metal-coated tips limits the usable AFM probes, and polarization effects should be taken into account when employing such tips (see Sec. III A 1). This top-down configuration is the most common AFM-IR setup distributed over the world.

Many papers have already been published to describe and compare the different AFM-IR operating modes and their associated theory. Here, we wish to summarize most of this knowledge and give more details about their advantages and drawbacks.

Initially, AFM-IR measurements were realized in contact mode and employed tunable IR sources with fixed repetition rates, such as a free-electron laser or an OPO.^8,28–30 In this case, the observed IR-induced cantilever ringdown is a damped mode as presented in Fig. 3(b). The sensitivity in contact mode varies with the sample thickness and the cantilever stiffness: usually, a minimum thickness of 50–100 nm is required.³¹ Some cantilevers specifically designed to be extremely sensitive are able to follow the minuscule thermal expansion of the sample allowing to measure a self-assembled monolayer.²⁴ Unfortunately, these cantilevers are not easily available but have already shown incredible results—by quantifying the maximum of thermal expansion (2 pm) and by assessing the relaxation of metal-organic microcrystals.

The development of novel IR sources allowed to push the sensitivity limit in the first improvement of the technique: the resonance enhanced AFM-IR mode.¹⁵ Currently, this mode is the main operating AFM-IR mode besides tapping AFM-IR. The idea is to use an infrared tunable laser with an adjustable repetition rate that is set to exactly match one of the contact resonance mode frequencies of the cantilever. In this condition, the thermal expansion energy is fully transferred to this mode and not spread over all the modes, as is the case with a 1–10 kHz repetition rate initially used (in an OPO, for instance). Then, the AFM-IR deflection signal expression (4) becomes¹ 

where Q_n is the quality factor of the mode and u₀ is the maximum thermal expansion which is proportional to the absorbance.

Here, the expression shows that the measurement of the oscillation amplitude will also give the local absorbance. Comparing this expression with Eq. (4), the damping factor is removed and the amplitude in resonance enhanced AFM-IR is consequently amplified by the quality factor (Q_n/2π) of the selected mode. This aspect clearly represented a breakthrough for the technique. The combination of this mode with local field enhancement due to gold-coated tips has proven to drastically improve the sensitivity, pushing the detection limit to a few nanometer thick samples and monolayers with ∼25 nm spatial resolution.¹⁶ Figure 6 illustrates the sensitivity and resolution obtained with this resonance enhanced mode combined with a gold-coated tip on a bacterium purple membrane.

Compared to the original contact mode AFM-IR, the resonance enhanced mode requires feedback for the laser repetition rate. The frequency of the contact resonance (f_n) is directly related to the mechanical contact between the tip and the sample surface. As a consequence, during the scan, any small changes in the sample stiffness or topography will induce a shift in the frequency of the resonance maximum. If the laser repetition rate cannot track this frequency shift, the measurement will go out of resonance, inducing a variation of the signal which is no longer related to local absorbance changes. To solve this problem, the repetition rate of the laser can be locked either with the resonance amplitude maximum³² (chirp mode) or its phase into a PLL feedback loop (phase-locked loop).¹⁹ 

The main advantage of the PLL compared to the chirp mode is its speed. Furthermore, it allows simultaneous chemical (looking at the amplitude of the deflection signal) and mechanical (looking at the frequency change of the repetition rate) mapping together with the topography image (typical time of 3 min for a 300 × 300 pixels image).

As discussed above, top-down illumination calls for metal-coated AFM cantilevers to minimize interactions with the IR beam. Metal coating (e.g., with Au or Pt) also provides local field enhancement which leads to an increase of the IR signal and achieves the best sensitivity obtained in AFM-IR.¹⁶ This effect is well known and already used in other tip based near-field spectroscopic techniques such as tip enhanced Raman spectroscopy.³³ While this local field enhancement is a key factor to increase AFM-IR sensitivity, this implementation may cause distortions in spectra³⁴ suggested to originate from the influence of the gold coating of the tip on the polarization of the local field. This initial hypothesis has recently been confirmed with an extensive study of the polarization effect made by our team.²⁶ We compared the local IR spectra obtained on different protein fibers with the two different illumination setups (i.e., bottom-up vs top-down) associated with silicon tip and not silicon-coated tip or gold-coated tips. It was possible to clearly describe the effect of the gold-coated tip on the local polarization and the potential effects on the protein spectra based on whether the object does or does not carry an intrinsic polarization. Therefore, one should be careful with the use of gold-coated tips depending on the nature of the studied object. While the local field enhancement is a real improvement for the technique sensitivity, it can induce spectral changes that render the measured spectra, no longer comparable with classic FTIR spectra. As the use of gold-coated tips is mandatory for top-down illumination, it also means that the bottom-up illumination setup should be preferred for polarization studies or specific objects with intrinsic polarization such as amyloid fibrils.

Many experimental results have shown that the region of interest could even be detected if buried sub-surface under a different material.^7,35–38 The probing depth of the technique is something that is not currently well-calibrated and difficult to estimate. It strongly depends on the mechanical nature of the measured sample but usually, it is around a few hundred nanometers. Consequently, the AFM-IR technique is an in-depth probing technique and does not focus solely on the surface layer. If the user wants to exclusively detect materials on the top surface such as contaminations, it will only be possible if the top species exhibit a specific absorption band different from all of the bulk samples below.³⁹ The lateral resolution is also associated with the probing depth: the deeper the absorbing object is located, the more the expansion will spread out around the absorbing region. The work of Quaroni has extensively analyzed the AFM-IR resolution and sensitivity as a function of several parameters, such as the laser repetition rate and pulse duration.²⁵ 

He has demonstrated that the thermal wave does not limit the resolution and that the spatial resolution of AFM-IR was one order of magnitude better than the one calculated from thermal diffusion [Fig. 7(d)]. Based on the results obtained by Quaroni, we can notice that the thermal length (L_th) and the thermal expansion length (L_exp) are not following the same power law even if the modulation frequency is the key factor for both,

where f is the repetition rate of the laser that matches the chosen contact resonance mode.

To deal with the in-depth effect, a limited solution is to use a higher contact resonance mode to decrease the thermal expansion length as demonstrated by Quaroni.²⁵ 

Figures 7(a) and 7(b) illustrate how the repetition rate has an impact on the contrast/resolution of the infrared mapping. We demonstrate this by imaging the absorption of a PMMA bead.

Figure 7(c) illustrates that using high frequency modes is a possible solution to improve the lateral resolution by decreasing the expansion length, and consequently, the probed volume. The estimated resolution at 1750 kHz from Fig. 7(d) is around 80 nm.

AFM-IR in contact mode is a very sensitive nano-spectroscopy tool. However, in some instances, the measured signal does not entirely originate from the tip–sample interaction. Under some experimental conditions, the obtained IR spectra and maps can be disturbed by a signal caused by the acoustic pressure that the laser spot induces. Note that the average diameter of the focused IR spot can be 50–60 μm, which needs to be compared to a typical 450 μm long and 50 μm wide contact mode cantilever. The first experimental results that have demonstrated the existence of acoustic waves has been reported by Chae et al. using a nanophotonic transducer.²⁴ This acoustic wave is created by the thermal expansion of the absorbing sample regions that surround the tip. The wave interacts with the cantilever and creates additional excitation. As a consequence, the acoustic contribution can be directly added to the AFM-IR deflection signal,

where P_n^tip corresponds to the transduction coefficient tip–surface and P_n^acoustic describes the corresponding coefficient for the interaction between the acoustic wave and the cantilever.

If the excitation by the acoustic wave represents a significant share compared to the tip contribution, then recorded spectra may be contaminated in part by a chemical signal coming from the surrounding absorbing region. Figure 8 represents experimental results that demonstrate the existence of the acoustic wave. Individual PMMA beads are distinguished in the chemical image [Fig. 8(a)]. Imaging the same area out of contact with the sample, the acoustic mapping [Fig. 8(b)] clearly reveals the full shape of the PMMA beads bundle (30 μm wide). All PMMA beads inside the bundle act like an emitter. We note that if the PMMA bead is isolated, the acoustic wave is negligible as observed for the beads located in the top left of the map. Note that the data in Fig. 8 were recorded for different resonance frequencies and laser powers, exaggerating the acoustic drive for clarity.

The described acoustic effect can be considered as an important source of spectral misinterpretation if it is not evaluated and taken into account. Unfortunately, it is not possible to directly subtract the acoustic signal from the AFM-IR signal in contact, as the resonance frequency and transmission coefficient are different (P_tip ≠ P_acoustic). Since the acoustic effect comes from the less-localized laser spot absorption, the sample preparation could be a way to avoid it during spectroscopic studies. In the presented example in Fig. 8, this could be achieved by reducing the concentration of absorbing beads inside the laser spot to only one. Another possible approach is to minimize the size of the laser focus.

Tapping AFM mode is a dynamic mode that has been successfully used to image soft or non-adhesive samples when the contact mode failed. The AFM-IR contact mode has the same limitation even in resonance enhanced configuration and is unable to study soft samples or just small nanoparticles.

Tapping AFM-IR has been recently implemented and is based on non-linear interaction between the AFM tip and the sample surface.^17,19 The detection of the infrared signal occurs at the exact sum or difference frequency between the tapping frequency at which the AFM operates and the repetition rate of the laser.⁴⁰ For signal amplification by the Q-factor, this detection mode is chosen to overlap with a cantilever resonance. Usually, the employed cantilever modes are the first and second eigenmodes. If both modes were equivalent for the AFM tapping operation, i.e., they were characterized by the same Q-factor and stiffness, the largest IR signal would be obtained when the driving mode (for AFM feedback) is the second tapping mode and the first one is used to detect the infrared signal.⁴¹ In practice, for hard tapping cantilevers, the driving mode and infrared detection are around 1600 and 250 kHz, respectively, leading to a laser repetition rate of 1350 kHz. For soft tapping cantilevers, the driving mode and infrared detection occur around 450 and 70 kHz, respectively, resulting in a 380 kHz laser pulse rate. As discussed in the previous paragraph, the lateral resolution is better for a higher laser repetition rate, which means that hard tapping cantilevers that intrinsically exhibit higher resonances are to be preferred if compatible with the sample properties. Using this configuration, the amplitude of the detected infrared signal can be extracted from⁴⁰ 

where B contains parameters of the tapping mode and its set point, χ_s is the non-linear term of elasticity, t_tap is the time of contact, f_drive the frequency of the driving mode, f_IR the frequency of the detection mode, Q_IR the quality factor of the detection mode, and u₀ is the thermal expansion.

The non-linear detection has no influence on the spectroscopic acquisition: infrared spectra obtained by this mode match FTIR spectra as well as in contact mode.¹⁷ However, some differences can be observed in the imaging mode. Tapping AFM-IR generally provides higher spatial resolution than resonance enhanced AFM-IR at the same laser repetition rate frequency. The non-linear effect is confined to the top surface region where the tip modulates the sample at the driving frequency of the tip, which reduces the probed volume. As the operating mode is tapping, the time of contact of the tip is short, limiting the transduction of the thermal expansion equation (8). These aspects minimize the thermal stress spreading and lead to an increase in the resolution. Unfortunately, the mechanical or topographic changes occurring during the imaging scan will have a strong impact on the measured infrared signal. The cantilever modes in tapping are defined in the air when there is no interaction with the surface. When the AFM system engages the tip on the surface, the amplitude of the driving mode decreases but the resonance also shifts to higher values. Each time the tip interacts differently with the surface, visible in the phase variation, the resonance frequencies of the first mode and second mode will vary. Under such conditions, a fixed laser repetition rate does not match the difference frequency of the involved modes anymore, resulting in an IR signal decrease. As an example, Fig. 9 represents how an edge effect can disturb the chemical mapping in the case of the analysis of a spherical object, namely, a polymeric nanoparticle. Here, the inhomogeneous IR signal is caused by a topography (and hence, phase) variation at the top edge of the particle.

Despite the advantages of tapping AFM-IR over resonance enhanced the contact mode in terms of the spatial resolution and breadth of compatible samples, tapping AFM-IR must be used with caution to avoid any misinterpretation. The best way to avoid these artifacts is to implement a PLL to track the infrared phase signal, similar to the resonance enhanced AFM-IR.

A third AFM-IR spectroscopy technique is peak force IR which is based on the peak force tapping (PFT) operational AFM mode.⁴² In PFT, the probe–sample distance is modulated sinusoidally (with either the tip or sample moving) at a frequency typically in the 1–4 kHz range, far below the cantilever resonance. In each cycle, the probe approaches the sample surface, see Fig. 10(a), and snaps into an intermittent contact [around 160 μs in Fig. 10(a)] during which the force (assessed via the cantilever deflection) increases up to a defined maximum. This “peak force” (at ∼195 μs) is kept constant in each PFT cycle and used as AFM feedback. After reaching this maximum force, the tip is withdrawn from the surface and snaps off the sample at the adhesion force (at ∼240 μs). Peak force and amplitude of the sinusoidal modulation are user adjustable, the first limiting the maximum tip–sample contact force, the latter determining the duration of intermittent contact. The process of force feedback allows for the imaging of samples at precise, user-defined forces down to 10 pN, while the lateral shear forces from contact mode are minimized. This preserves tip and sample integrity and renders PFT especially suitable for the imaging of soft and delicate samples, such as polymers⁴³ or biomaterials like DNA,⁴⁴ and living cells.⁴⁵ In addition, PFT inherently acquires force curves at kHz rates from which nanomechanical properties such as modulus, adhesion, or dissipation are automatically extracted in PF-QNM (quantitative nanomechanical mapping). This ability to correlate topography and nanomechanical data can be furthermore augmented with simultaneous electrical measurements such as Kelvin-probe force microscopy (KPFM) and Tunneling AFM (TUNA).

Similar to contact mode and tapping mode, PFT can be supplemented with IR laser excitation to enable NanoIR spectroscopy and chemical imaging termed PFIR. Invented in 2017 in Xiaoji Xu's group,^46,47 IR laser pulses illuminate the tip–sample region for every other PFT cycle during tip–sample contact. Photothermal expansion of the sample kicks the probe and results in an oscillation of the cantilever, in general, at several contact resonances. The transient deflection signal of the cantilever is then dictated by Eq. (4) as for the contact mode AFM-IR. Figure 10(a) displays the PFIR deflection signal with IR-induced oscillations (red curve) in every other PFT cycle when IR pulses are present. The subtraction of consecutive PFT cycles, one with [red curve in panel (a)], the other one without IR radiation (black curve), removes the slowly varying cantilever deflection of the force curve and exposes the pure IR-induced cantilever oscillation, shown in Fig. 10(b). An FFT [Fig. 10(c)] and subsequent spectral integration around one or several contact resonance peaks serve as the IR response signal of the sample. Note that synchronization of the laser pulses with the PFT modulation frequency allows for coherent averaging of several PFT cycles in the time domain before FFT calculation, which improves the signal-to-noise ratio. Lock-in detection in principle is also possible, especially for multi-pulse excitation. Such multi-pulse excitation with a pulse train, as shown in Fig. 10(a), employs multiple laser pulses per PFT cycle as opposed to a single pulse and benefits from signal amplification for increased signal-to-noise ratio proportional to the square root of the number of pulses.⁴⁸ Recently, further improvement in sensitivity originated from subjecting every PFT cycle to IR pulsing while the slowly varying curvature of the PFT cantilever deflection is subtracted based on a polynomial fit.^49,50 The duty cycle is then doubled and can still be enhanced by extending the pulse train over the entire contact time.⁵¹ For such resonance enhanced operation with pulsing at a cantilever contact resonance, the IR-induced deflection signal in PFIR may be described by Eq. (5), in the same way as for resonance enhanced AFM-IR. In addition to the cantilever oscillation, the photothermal expansion may lead to an offset/step in the vertical deflection signal from the volume expansion under the tip [up to a few 100 pm,⁴⁶ see offset at ∼195 μs in Fig. 10(b)] between green and red lines from fits), but most commonly, the IR-induced cantilever oscillation is evaluated exclusively.

PFIR is conceptually very similar to contact mode IR. The spatial resolution is, however, much better in the sub-10 nm range in air^46,52,53 or ∼10 nm in liquid.⁴⁹ Compared to the contact mode, the absence of lateral forces in PFT during scanning results in a smaller contact area for probe–sample interaction which ensures already superior AFM imaging quality. Another factor is the precise force control in every PFT cycle that allows imaging with small forces of typically ∼1–15 nN in PFIR, again leading to a small contact area. On the other hand, compared to tapping AFM-IR, PFIR lacks the non-linear mixing effect that reduces the probed volume. And usually, the IR laser repetition rates in PFIR for multi-pulse excitation are lower in the 200–1000 kHz range compared to 1200–1500 kHz for tapping AFM-IR—so that the thermal diffusion length in PFIR should be slightly longer causing spatial averaging under the tip. Nevertheless, the current PFIR spatial resolution data for a comparable tip radius matches or exceeds tapping AFM-IR, and additional work is needed to gain a better understanding.

In order to optimize PFT imaging and PFIR signal-to-noise, the peak force and modulation amplitude are important parameters. Typically, the peak force is in the 1–15 nN range. The PFIR signal was found to increase with the PFT peak force at the expense of spatial resolution, which is a consequence of a larger contact area for higher setpoints.⁴⁶ The PFT amplitude determines the contact time with respect to the entire PFT cycle time, hence, the duty cycle for PFIR. The lower the amplitude, the longer the contact time and the higher the S/N, but on the other hand, the higher the risk for the probe to be unable to pull off the surface due to adhesion. Typical amplitudes are in the 30–150 nm range with a duty cycle of ∼10%–25%. PFIR probes have a typical tip radius of 20–35 nm and conductive coatings such as Pt or Au. Since PFIR is often combined with QNM to extract nanomechanical data, the cantilever stiffness should be matched to the expected modulus range if quantitative data is desired,⁴² ranging from ∼0.4 N/m for soft 0.1 MPa biomaterial⁴⁹ to 40 N/m for several GPa in shale source rock.⁵² 

Similar to contact mode and tapping mode AFM-IR, PFIR is sensitive to contact resonance shifts from changes of the mechanical properties under the tip. Such artifacts have likely been observed in the literature⁵⁰ and are expected to depend on how the PFIR signal is obtained. Using a single IR pulse in a PFT cycle and extracting the IR signal as integrated FFT amplitude over a wide frequency window around the contact resonance peak is less sensitive to small shifts of the resonance within the chosen integration boundaries. However, using multi-pulse excitation with pulse trains at a frequency tuned to the contact resonance for signal amplification is more susceptible to induce shifts. Figure 10(d) exemplifies the problem based on a PFIR scan across the interface of a PMMA bead embedded in an epoxy matrix. Pulse trains at 787 kHz were used. The IR absorption at the PMMA carbonyl resonance of 1730 cm⁻¹ displays a vanishing IR signal in the highlighted, dashed areas within PMMA. In comparison, implementing frequency tracking in Fig. 10(e) with a phase-locked loop (PLL) to automatically maximize the IR signal at every imaging pixel reveals a more homogeneous absorption signal for otherwise same scan and sample conditions. Specifically, the apparent low absorption regions highlighted in panel (d) originate from a drop of the mechanical resonance down to ∼730 kHz compared to the ∼810 kHz of the surroundings and does not stem from a local change in the chemistry of the sample. This example highlights the improvements from a frequency tracking PLL in PFIR to minimize mechanical artifacts.

In conclusion, PFIR represents an interesting AFM-IR spectroscopy technique that has been applied to a diverse range of samples^52–55 including liquid environment based on the bottom-up configuration.^48,50 The advantages of PFIR lie in precise force control for the imaging of delicate samples like cells in fluid,^45,56,57 and the simultaneous access to additional nanomechanical and nanoelectrical information.⁵⁸ To overcome possible mechanical artifacts in imaging, a PLL-based frequency tracking mechanism has been demonstrated. In the future, the sensitivity of the PFIR technique may be further improved, e.g., by extending the tip–sample contact time per PFT cycle, by higher PFT modulation frequencies, or via probe optimization for higher contact resonances and larger IR-induced deflection changes.

Since the first publication on the proof of concept,¹⁰ different review papers have covered the diversity of applications of the AFM-IR technique.^1–9 During the last years, the number of publications in which AFM-IR is involved has continuously risen (Fig. 11). Thus, in parallel with technical improvements, there is an evolution of the applications as well as in the complexity of the studied samples. In this section, selected examples using AFM-IR are cited to illustrate those evolutions from the first publication to now, 16 years later.

First, for the two historical fields of applications, polymer science and biology, a change in the use of the technique gradually emerges. In both fields, it shifts from the “proof of concept in terms of capabilities”^59,60 to a standard tool: one technique among a set of others to characterize a sample.^29,61–65 Furthermore, applications become more and more complex and diverse. In polymer science, we move from bulk polymer studies^60,66 to organic/metallic interfacial interactions,^67–71 nanodomain focused studies,^72,73 and polymeric nanoparticles.^17,18 In biology related applications, the first studies were performed on isolated cells^35,74 and microorganisms,^7,13,28,59 as it was the optimal sample size to work with, from where research moved on to extremely small samples, such as protein fibers^26,75–77 and large scale samples such as tissue and bones.^{64,65,78–84}

This evolution illustrates the versatility of the technique and its potential to become a standard lab tool. As a consequence, correlative methodologies with other well-known techniques are becoming more and more important for comprehensive sample characterization. Different possibilities exist such as correlative measurements with TEM,⁷⁷ fluorescence microscopy,^38,85,86 non-linear microscopy,⁸⁷ Raman microscopy,⁸⁸ MEB/EDX,⁸⁹ or other AFM techniques themselves such as acoustic AFM,³¹ QNM,⁵² or KPFM.⁵⁸ 

Nowadays, the AFM-IR technology continues to enter new fields of applications, far from the classical polymer and biology domains of photothermal techniques. On the one hand, despite the low thermal expansion coefficients of solids, AFM-IR helps to address more and more complex problems in solid-state physics. It is used in energy research to study conducting polymers^29,90–92 and solar cell components,^54,93–97 and delivers important insight in the field of semiconductors,^98,99 plasmonic effects,^100–102 or 2D materials analysis.^48,103–106 On the other hand, hybrid samples are analyzed using the technique such as metal-organic framework based samples^24,107,108 or samples from cultural heritage domains,^87,109–111 geology (organic inclusions in whole rocks and bitumen),^52,112–115 and extraterrestrial materials, such as interplanetary dust particles (micro-)meteorites.^89,116

During the last decade of the evolution of AFM-IR, we clearly observe a transition from a proof of concept to a tool for infrared nano-spectroscopy. Driven by technical improvements such as resonance enhanced and tapping AFM-IR modes, the technique has been involved in a huge variety of applications. Here, we propose three new promising developments that probably push the technique for the next decade.

AFM-IR measurements in the past years were mainly focused on soft matter samples, such as polymers or biological specimens. The capability to study mineral phases still largely needs to be explored. Recently, some research works concentrated on geological materials. All of them focused mainly on organic inclusions inside the samples.¹¹² First measurements on inorganic samples were conducted by Katzenmeyer et al. on silica nanoparticles inside a PMMA matrix¹¹⁷ and on metal-organic frameworks¹⁰⁷ and by Cattinari et al.⁶² on silica nanoparticles inside a natural rubber matrix. These publications demonstrate the possibility to acquire an AFM-IR signal on samples with low thermal expansion coefficients, paving the way to measure complex materials such as organic crystal and mineral phases.

Among the first attempts to analyze a mineral phase with AFM-IR was the one by Kebukawa et al.¹¹⁶ Studying Belt and Murchison meteorites, they were able to measure a photothermal expansion signal from silicates. Contrary to soft matter samples usually studied with AFM-IR, minerals harbor a thermal expansion that can not only be anisotropic but also their thermal expansion coefficient is at least one to two orders of magnitude below those of polymers. Consequently, if it is possible to obtain a photothermal signal from mineral phases, it is not as straightforward to compare it with classic FTIR measurements, as it is the case for organic phases. This first result by Kebukawa et al.¹¹⁶ was confirmed by a study from our group in which we measured the Si–O absorption from silicate nanoparticles, highlighting the possibility to obtain reliable AFM-IR signals on inorganic phases.⁶² 

Other interesting results from our group were obtained with the AFM-IR technique on organic crystals such as cysteine. Even if some discrepancies can be observed, such as peak shifts and peak intensity variations, the AFM-IR spectra are broadly comparable with those from conventional FTIR reflectance spectroscopy. While in terms of chemical structure, these organic crystals cannot be directly compared to minerals; our results still demonstrate the possibility to acquire reliable signals from crystal structures¹¹⁸ (Fig. 12).

To address the AFM-IR capabilities for mineral studies, we analyze results obtained on two different standard mineral samples with different intrinsic properties. The first one is a smectite powder that was pressed onto a CaF₂ surface. The second one consists of a polished section of a forsterite monocrystal, which was polished along its main crystalline orientation. Smectite is a clay mineral that belongs to the phyllosilicates family, sheet silicate minerals with parallel sheets of silicate tetrahedron. It crystallizes in the monoclinic crystal system. Forsterite is the magnesium-rich end-member of the olivine silicates class. It crystallizes in the orthorhombic crystal system.

In Fig. 13, we compare FTIR microscopy and AFM-IR data on the same samples of smectite and forsterite. The AFM-IR acquisition is limited to the 900–1175 cm⁻¹ range, corresponding to the spectral coverage of the QCL chip used in this study. It overlaps with the Si–O stretching vibrational mode absorptions in both samples.

The measurements in Fig. 13 reveal significant differences in the agreement between AFM-IR and conventional FTIR data, depending on the studied mineral. In the case of the smectite powder, the spectra recorded in AFM-IR and FTIR microscopy are very similar, the main difference being a small spectral shift whose magnitude is comparable to the one already observed with organic crystals in Fig. 12. In the case of the Forsterite monocrystal, while specific and reproducible AFM-IR signals are obtained, some band positions correspond to the conventional FTIR microscopy ones (980 and 1075 cm⁻¹ bands), others have no direct correspondence to the expected Si–O absorption bands. Moreover, band ratios drastically differ from one technique to the other. As a consequence, these measurements show discrepancies depending on which crystalline mineral is investigated.

Multiple effects can explain these differences. Crystals are highly oriented objects, and hence, these AFM-IR measurements face a similar issue as another previously discussed representative of such an object, the amyloid fiber (see Sec. III A 1). Based on the results observed on the amyloid fiber, it is expected that some polarization related effects occur that could explain differences in band position and relative band intensities. As we are dealing with crystals, the form of the sample will have an influence on the results. In the case of smectite, we analyzed a fine powder. One can, thus, expect that the recorded spectrum represents an average over many different orientations of the polydisperse smectite crystals. Such averaging also explains the broad shape of the spectrum in Fig. 13(a). By contrast, in the case of the Forsterite sample, that is a monocrystal, a single orientation is probed, also leading to sharper peaks [Fig. 13(b)]. This is, however, not enough to explain such huge spectral differences observed in the case of the Forsterite crystal. Another parameter that will affect the result is the optical behavior of the sample. Mineral optical properties are not comparable to the ones of most organic samples usually studied with the AFM-IR technique. In the case of top-down illumination, which is the most commonly used configuration and the one employed here, these optical properties can have an influence. Even though the AFM-IR signal is generated by the imaginary part of the sample's refractive index, the optical transfer processes through the sample should be taken into account. Here, the IR beam is focused on the surface with a given incidence angle and optical effects, such as reflection, induce distortions in the resulting spectrum. Complementary experiments are ongoing to discriminate which factors are predominant and rationalize the observed spectral distortions.

In conclusion, the presented preliminary data on minerals highlight the capabilities of the AFM-IR technique to study mineral phases. However, compared to most polymers, a deeper understanding of the nature, form, and orientation of the mineral is required, to be able to compare the results with classic FTIR spectroscopy. Additional experiments are necessary to address and constrain the origin of the spectral divergences observed in some instances, and to define the driving parameters to allow a more direct AFM-IR analysis for a broader kind of minerals.

The liquid environment AFM is routinely used for biophysics research in fields such as protein auto-assembly and living cell stress.¹¹⁹ In these mere AFM studies, the obtained information mainly consists of the dimension of the sample (down to nanometer resolution) or their mechanical properties. The possibility to simultaneously collect infrared spectra at the sub-micrometer scale is the most exciting and challenging endeavor of these last decades. Some encouraging results materialized but each time the contribution of the water absorption is strongly present or dominates^74,120 or it is required to replace water with heavy water (D₂O) which has a much smaller absorption coefficient.¹²¹ The main problem of AFM-IR operation in a fluid is the strong absorption of liquid water in the mid-infrared range. To drastically reduce this contribution, the best configuration for working in fluid seems to be the bottom-up illumination [Fig. 5(a)]. Indeed, in this configuration, the evanescent field is first exciting the sample in contact with the prism surface prior to the liquid, whereas in top-down illumination, the IR beam is passing through a significant part of the water medium (millimeter long) before illuminating the sample. However, even in the bottom-up configuration, the evanescent wave at the prism top surface interacts with sufficient water around the tip to provoke a non-negligible acoustic wave hitting the cantilever, which creates a background signal limiting the AFM-IR capabilities. In the following sections, we propose a way to deal with the water absorption in order to obtain AFM-IR results similar to measurements in the air without special data treatment or modification of the system. We restrict our analysis here to resonance enhanced and tapping AFM-IR. We remark that the third IR nano-spectroscopy mode, Peak force IR, has recently established its ability for fluid measurements.^49,51 Since its concept shares similarities with resonance enhanced AFM-IR, we expect similar challenges for PFIR as for this operational mode.

First measurements with resonance enhanced mode and bottom-up illumination have been conducted by Jin et al.¹²¹ in heavy water (D₂O) and shortly after by Ramer et al.¹²⁰ in H₂O and D₂O. In the latter case, in standard water, the thickness of the measured samples exceeded a few hundred nanometers, which is one to two orders of magnitude below the achievable sensitivity in an equivalent setup in air. To evaluate the loss of sensitivity in liquid, we reproduce a comparable experiment. Here, our sample is a commercial varnish polymer that has been spread on a ZnSe (Zinc Selenide) prism in many droplets of different sizes within distilled water. The AFM-IR system is the nanoIR1 (first commercialized version) with the bottom-up illumination. The varnish possesses two specific bands in the 1800–1530 cm⁻¹ spectral range; an ester carbonyl at 1750 cm⁻¹ and a carbonyl at 1630 cm⁻¹. Liquid water has a large absorption band centered at 1650 cm⁻¹. The chosen cantilever is a gold-coated contact cantilever with 0.1 N/m stiffness typically used in AFM-IR measurements in air. The contact mode resonance is locked at the third mode at 350 kHz. The chemical image at 1750 cm⁻¹ and corresponding spectra on varnish droplets are represented in Fig. 14.

The AFM topography displays droplets of different sizes on the surface, ranging in thickness from micrometers to a hundred nanometers. The corresponding chemical mapping at the ester carbonyl absorption [Fig. 14(b)] clearly reveals the large droplet (1.3 μm thick) centered in the image, and with a much weaker response, the two other droplets with a thickness of ∼400 nm [marker Fig. 14(b)]. Droplets with a thickness under ∼400 nm are not detected. Spectra on the different varnish beads [Fig. 14(c)] confirm what is apparent in the chemical mapping. The spectrum taken on the substrate away from any varnish shows a large, wide band around 1650 cm⁻¹ which can be assigned to liquid water [position D in Figs. 14(a) and 14(c)]. The spectrum obtained on the largest droplet [position A in Figs. 14(a) and 14(c)] reveals a distinct infrared fingerprint of varnish that is not disturbed by the presence of water. For the 400 nm thin droplet [B, Figs. 14(a) and 14(c)], the spectrum still exhibits an ester band at 1750 cm⁻¹ but also a large band centered at 1650 cm⁻¹. The recorded spectrum represents a mix between the varnish signal and the acoustic signal [Fig. 8, Eq. (7)] coming from water absorption. The droplet with a thickness of 350 nm does not exhibit any varnish-related absorption band in its spectrum, but rather the background from the aqueous environment [position C, Figs. 14(a) and 14(c)]. In this experiment, we determined that the minimum thickness for detection of the varnish is above 400 nm, which is two orders of magnitude lower than the threshold in an equivalent AFM-IR measurement in air. This is also the same order of magnitude that was observed in Ramer et al.¹²⁰ All the droplets below this minimum thickness are not detected because the force of the acoustic wave from the water absorption dominates the excitation of the cantilever when compared to the sample thermal expansion excitation. This situation is a perfect illustration of artifacts and issues due to the acoustic wave [P_n^tip << P_n^acoustic, Eq. (7)]. The only way to increase the sensitivity is to suppress the acoustic wave signal in order to only detect the tip–surface interaction effect.

Tapping AFM-IR has been briefly described in paragraph in Sec. III A 2. The equation of motion of the cantilever mode used for the infrared detection can be written as a damped harmonic oscillator,⁴⁰ 

where z_IR is the amplitude of the detection tapping mode, f_IR the detection tapping frequency, m* the effective mass of the detection tapping mode, F_tip the force induced by the thermal expansion, and F_acoustic the force applied on the cantilever by the acoustic wave.

The force induced by the thermal expansion under the tip is related to the Hertz contact and can be expressed as using Taylor expansion,⁴⁰ 

where k_z is the contact stiffness, A_drive the amplitude of the driving tapping mode, f_drive the driving tapping frequency, D the tip–surface distance, u(t) the thermal expansion driven by the laser repetition rate f_laser, and χ_s the non-linear term of contact elasticity.

The time behavior of the acoustic force follows the thermal expansion as the propagation speed is ∼1500 m/s in water and the distance between the sample surface and cantilever is around 10 μm. The acoustic force can be written as

where F_ac is the resulting force of the acoustic pressure, u₀ the maximum thermal expansion, and u(t) the thermal expansion driven by the frequency f_laser.

The cantilever mode f_IR is only excited if the resulting force applied to the cantilever represents a non-linear term, namely, the sum or difference frequency between the driving frequency f_drive and the laser repetition rate f_laser.¹⁷ The acoustic term Eq. (11) is purely linear with modulation at f_laser, and hence, leading to a vanishing contribution. Only the double product of the second order bracket in F_tip Eq. (10) between the driving mode (f_drive) and the thermal expansion (f_laser) will lead to a combination of these frequencies that can excite the cantilever resonance at f_IR. In other words, any cantilever excitation from the acoustic wave occurs at a frequency f_laser that is not a resonant frequency of the cantilever, in contrast to IR detection that does occur at a cantilever resonance that implies signal generation. In conclusion, looking at the theoretical concept and the associated equations, the non-detection of the acoustic wave is inherent to tapping AFM-IR and it should be possible to suppress any acoustic signal such as the one originating from liquid water absorption. Note that the acoustic effect described in Sec. III A 1 in the air is also absent in tapping AFM-IR for the same reason.

This theoretical approach can be easily validated with an experimental measurement. The same sample, as in the previous paragraph in Sec. IV B 1, has been studied to test the performance of the tapping AFM-IR mode. The tapping cantilever is a high frequency gold-coated one (300 kHz in air) giving tapping frequencies in the liquid around 125 kHz and 250–400 kHz for the first and second modes, respectively. The first mode is about the same for the tested cantilevers but the second tapping mode seems to be more sensitive to the way of how the probe is kept in the cantilever holder. Here, we use the first mode for AFM feedback and the second to acquire IR data because, in this example, driving at the second tapping mode was not stable in the liquid.

The results obtained on the varnish droplets are presented in Fig. 15.

In this experiment, the chemical mapping clearly resolves 100 nm thick small droplets, even if some noisy lines appear during the scan. In general, the infrared signal in fluid seems noisier than in air. This could come from the mechanical support of the probe and the non-optimal cantilever holder of the used system. The AFM-IR spectrum at position D in Fig. 15(a) corresponds to the surface of the prism and shows no water band [Fig. 15(c)] in contrast to the contact mode results in Fig. 14, proving the aforementioned theory of the tapping AFM-IR advantage in the fluid. The other spectra in Fig. 15(c) are associated with droplets of 330 and 100 nm thickness for positions A and B/C, respectively. They not only present the characteristic ester band at 1750 cm⁻¹ but also a band at 1650 cm⁻¹. The latter could stem from a carbonyl resonance of the varnish or from the diffusion of water molecules into the droplet.

Nevertheless, spectra and mapping of 100 nm thick droplets, without any artifact from the surrounding water, demonstrate for the very first time that tapping AFM-IR is the perfect operating mode to perform infrared nano-spectroscopy in liquid without the need for technically advanced additional setups or complex data treatment. This simple way to realize AFM-IR in a fluid environment represents for us a major breakthrough that opens the door for common biological studies at the nanoscale. We emphasize that this is possible only because AFM-IR is related to mechanical and not optical detection.

A quantitative approach for the probing depth of the AFM-IR technique is still one of the open questions. In particular, it strongly depends on the mechanical nature of the sample. There is a lack of experiments to estimate the influence of such factors, but in general, a depth up to several micrometers is expected depending on the operating mode parameters. To limit the influence of the probing depth, and hence of sub-surface material on the analysis, surface sensitive AFM-IR has been recently developed in our laboratory^122,123 to localize the AFM-IR signal sensitivity to the top surface of the sample, independently of its properties. The principle of this novel mode is based on the non-linear interactions of acoustic waves and is inspired by the acoustic AFM.¹²⁴ 

The surface sensitivity mode operates in the contact mode. The laser illuminates the sample at a repetition rate (f_laser) far from the contact resonance with the highest possible laser frequency. The cantilever is modulated by a piezo, e.g., via the existing tapping piezo, at a high frequency (f₁). Both frequencies f_laser and f₁ are chosen such as to match their sum or difference frequency with one of the cantilever contact resonances (f₂). Similar to the tapping AFM-IR operation principle, IR detection at f₂ only happens because of the non-linear interaction between the evanescent acoustic wave created by the modulation of the tip at f₁ and the thermal expansion of the absorbing sample driven by the laser at repetition rate f_laser (Fig. 16). Since the new mode only utilizes an already existing cantilever drive piezo, as in tapping AFM-IR, switching from classic to surface sensitive AFM-IR is seamless without hardware modification, which allows analysis of the same area with both modes.

As the displacement field induced by the tip indentation usually reaches from the top surface of the sample to about a depth of 10–30 nm (depending on the sample material and the applied force), the vertical extent analyzed by the surface sensitivity mode is much smaller than for classic AFM-IR acquisition in contact mode, which probes at least hundreds of nanometers vertically. We note that the tapping AFM-IR mode operates based on the same concept and can be considered as a particular case of the surface sensitivity mode because the frequency of the laser is always fixed to the difference of the cantilever resonances and cannot be tuned freely.

The equation of motion of the cantilever can be taken from Eq. (9) but where f₂ is the frequency of the detected contact resonance mode and F_int is the interaction force between the tip and the surface,

The interaction force is the contribution of the surface modulation and the thermal expansion and can be expressed as

where δ₀ is the static indentation related to the setpoint of the contact mode, A₁ the modulation amplitude of the surface, f₁ the frequency of modulation of the surface, and u(t) the thermal expansion.

As explained in Sec. IV B 2, the only terms that do not vanish from Eq. (13) are those from the double product of the second order bracket leading to the sum and difference of the surface modulation frequency f₁ and the laser frequency f_laser.

Finally, by applying the Fourier transform on Eq. (12) and after simplification, the expression of the infrared signal z_n can be written as

Here, χ_s is the non-linear term of contact elasticity, Q_n the quality factor of the contact resonance mode, k_c the cantilever spring constant, A₁ the amplitude of the surface modulation, f_laser the laser repetition rate, t_p the duration of the laser pulse, and u₀ the maximum of the thermal expansion (proportional to the local absorbance).

Important properties of this novel operational mode can be deduced from Eq. (14). Identical to tapping AFM-IR, the infrared signal is proportional to the local absorbance and gives spectra directly comparable to FTIR. The infrared signal detected by surface sensitive AFM-IR is strongly related to the quality factor of the chosen contact resonance mode. Usually, the first modes are used because of their good quality factor. The pulse duration should be carefully chosen: if it is too long, thermal expansion spreading will be integrated, and if it is too short, the resulting thermal expansion will be too low to obtain a workable signal. Finally, the modulation amplitude A₁ should not exceed the static indentation in order not to pull off the tip from the surface.

In Fig. 17, we show a comparison of the resonance enhanced AFM-IR and the surface sensitivity modes on a polymer sample. The sample is composed of PMMA beads (PolyMethylMethAcrylate) inside an epoxy matrix. Some of the beads are under the surface and are covered by a layer of epoxy [red dot in the topography image of Fig. 17(a)].

The chemical mapping at 1730 cm⁻¹ in the surface sensitive AFM-IR [Fig. 17(c)] seems cleaner and of higher contrast than the resonance enhanced contact mode [Fig. 17(b)] and the edges of the beads are sharper. The smaller probed volume and the higher laser repetition rate used in the surface sensitivity mode can explain this. But the most outstanding point is the non-detection of the buried bead that is clearly detected in resonance enhanced AFM-IR [red circle in Fig. 17(c)]. This observation is supported in the spectral data as well. The contact mode spectrum on the buried bead [red dot in Fig. 17(a)] is a mix of pure epoxy [blue curve in Fig. 17(d)] and pure PMMA [green curve in Fig. 17(d)] demonstrating that the contact infrared signal probably is averaged over the entire thickness of the sample (300 nm). In contrast, the surface sensitivity mode spectrum of the buried bead does not show any ester band of PMMA [red curve in Fig. 17(e)] and is exactly the same as the pure epoxy spectrum [blue curve in Fig. 17(e)] proving that the signal is coming only from the top surface. Note that there are subtle spectral differences in peak shapes that require more work to explain their origin.

To confirm that the surface sensitive AFM-IR mode is not only a demonstration of a better lateral resolution originating purely from high repetition rate laser illumination, Fig. 18 shows the comparison between the surface sensitivity mode and resonance enhanced AFM-IR mode measured on a buried PMMA bead in an epoxy matrix with equivalent excitation and detection frequencies. The surface sensitivity mode image [Fig. 18(d)] provides higher contrast and resolution than the corresponding resonance enhanced AFM-IR map [Fig. 18(b)], both taken at similar detection frequencies. The buried PMMA bead (marked by the black arrow) is not detected by the surface sensitive mode [Fig. 18(d)] while the resonance enhanced AFM-IR image reveals an absorption with half the amplitude of the beads on the surface. Figure 18(c) represents the resonance enhanced AFM-IR chemical mapping with a laser repetition rate similar to the one of the surface sensitivity mode in Fig. 18(d). Even though the repetition rate is very high for the resonance enhanced AFM-IR mode, the buried bead is still detected with a smaller size and lower amplitude than at 550 kHz. These experiments demonstrate that the physical process behind the surface sensitive detection is not only based on the high laser repetition rate, but also relies on the non-linear interaction of the tip modulation and the acoustic wave generated by the thermal expansion. This combination results in a smaller probing depth than for the resonance enhanced AFM-IR mode, with a high lateral resolution, contrast, and top-layer sensitivity.

To estimate the probing depth of this new acquisition mode, a PU (polyurethane) spin coated film has been used with a thickness that increases linearly from the edge (Fig. 19).

The top image in Fig. 19 presents the topography and the corresponding cross section [Fig. 19(a)]. Figure. 19(b) shows the resonance enhanced AFM-IR mapping at 1725 cm⁻¹ and the corresponding cross section. In that case, the absorption signal over the cross section increases roughly linearly with the sample thickness as expected in contact AFM-IR.²⁷ In this example, the contact mode still integrates over the full thickness even at 200 nm. The third image [Fig. 19(c)] displays the corresponding surface sensitivity mode mapping and its cross section. Here, the absorption signal quickly saturates even when the thickness increases further. We can observe the same behavior for tapping AFM-IR [Figs. 19(d) and 19(e)] demonstrating that the physical detection process is related to non-linear mechanical interaction between tip and sample. By comparing the cross section for the topography in Fig. 19(a) with the onset of the saturation in Fig. 19(c), we can estimate the probing depth in surface sensitive mode at 25 nm, which is much smaller than the one for the contact mode measurement in Fig. 19(b) (more than 200 nm). Looking at the tapping AFM-IR mapping [Fig. 19(e)] we find from the same analysis a probing depth of around 50 nm.

These three different studies summarized in Figs. 17–19 have evaluated the high performance of surface sensitive AFM-IR and demonstrated that the probing depth for this mode is about 20–30 nm, much better than resonance enhanced mode and better than tapping AFM-IR. This particular ability is possible only because the technique is based on acoustic wave mixing.

This new AFM-IR mode will open up numerous new areas of applications, for example, in the detection of surface contaminants or the analysis of top-layer films. The possibility to switch from the surface sensitivity mode to the volume integrating resonance enhanced contact mode—on the same location without modification of the system and in principle even during scanning—allows us to consecutively probe the top surface and deeper into the sample for a better understanding of the sample's chemical composition or to discard potential contaminations of top layers.

The AFM-IR technique is a relatively young, only 16 years old, technique. Remarkable improvements of its capabilities during the last decade have opened up a wide range of applications in numerous different research areas. The principal AFM imaging modes, namely, the contact, tapping, and peak force tapping modes have been implemented in AFM-IR leading to a versatile laboratory tool. With the recent addition of peak force tapping, correlative AFM-IR, nanomechanical and nanoelectrical measurements have emerged to address and characterize a wider variety of sample properties. Different measurement environments have been developed, allowing the AFM-IR technique to overcome some limitations of classical IR spectroscopy such as the possibility to function in a fluid environment. Compared to the other principal nanoscale vibrational spectroscopy techniques, scattering scanning near-field optical microscopy (sSNOM) and tip enhanced Raman scattering microscopy (TERS), the photothermal approach has a major advantage: it does not rely on the field enhancement or plasmonic effect of the tip, even if it can improve the IR detection. The development of IR laser technology has been a major driving force for the AFM-IR technique improvement. The wide tunability of the wavenumber enabled new applications and an adjustable repetition rate proved key to achieve the great sensitivity of the technique. Spatial resolution and sensitivity accomplished in AFM-IR now routinely allow us to detect molecular monolayers.

This review has shown how the AFM-IR community has expanded quickly over the years. This phenomenon is probably due to the reliability of the technique, its direct correlation with FTIR spectra, its low complexity, and its ease-of-use not requiring complex data treatment or extensive modeling.

The emerging trends described within this publication will dramatically increase the capabilities and attractiveness of AFM-IR. The possibility to measure IR spectra of minerals has been discussed but these materials need to be studied carefully and rigorously because of potential artifacts created by the polarization or illumination. However, the first results are clearly encouraging and open a new domain of applications for AFM-IR.

Measuring in liquid is perhaps the biggest advancement in AFM-IR. Resonance enhanced AFM-IR appears to suffer from acoustic wave background of the surrounding absorbing water, and we expect similar challenges for the peak force IR despite its impressive published results in liquid. However, with its operation based on non-linear mixing tapping, AFM-IR suppresses such acoustic wave backgrounds. It now enables to study any kind of biological sample including living samples in their native fluid environment without any system modification or additional signal processing. This direct IR measurement in water opens the door to new kinds of experiments, for instance, where AFM-IR follows the penetration of nanoparticles into Eukaryotic cells without the need for an additional label or probe molecules or the live study of the production of bio-polymers by bacteria.

The possibility to restrict the probed volume to a depth of only 20–30 nm for pure surface sensitivity in IR mapping and spectroscopy is a recently developed feature adding another advantage to AFM-IR. Contact AFM-IR can probe a depth of around several micrometers under the surface, supporting the detection of sub-surface buried objects. The new surface sensitive AFM-IR mode allows probing only the top surface without the contribution of bulk absorption beneath the 20–30 nm surface layer. Using these two different modes at the same location will provide precious information about the chemical distribution of the sample in the out-of-plane direction. For example, in cell studies, IR spectra averaged over the full sample thickness and from the cell membrane alone will become distinguishable. The new technique will also help in detecting the chemical nature of contaminants on an infrared absorbing surface, like the detection of blooming of plastic films used for food storage.

These new trends show how the AFM-IR technique is constantly advancing to add new capabilities and to render photothermal-based spectroscopy—one of the most exciting and productive approaches for instrumentation and innovative developments in nano-spectroscopy. The great potential of the technique will probably allow measurements in vacuum in the near future and realize the detection of a single molecular vibration.

The authors would like to thank Anirban Roy, Qichi Hu, Xiaoji Xu, Sharise Redmond, Henry Mittel, Shuiqing Hu, Weijie Wang, and Chanmin Su for valuable discussions. Lucie Delauche and Cécile Engrand are acknowledged for their contribution to samples preparation for the mineral study. The authors acknowledge Digital Surf for Mountains software support. This work was supported by the Paris Ile-de-France Region-DIM “Matériaux anciens et patrimoniaux.”

The authors declare the following competing financial interest(s): Martin Wagner is employed by Bruker Nano Surfaces Division, a manufacturer of instrumentation for AFM-based infrared spectroscopy. Alexandre Dazzi is a coinventor of AFM-IR patents and surface sensitive licensed to Bruker Nano Surfaces Division. The other authors state no conflict of interest.

The data that support the findings of this study are available from the corresponding author upon reasonable request.