Direction-resolved nanoscale optical imaging with near-nanometer resolution by emerging infrared torsional force microscopy Authors: Yonatan Gazit, Son T. Le, Aubrey T. Hanbicki, Adam L. Friedman, Min Ouyang Published: 2026 ABSTRACT Resolving nanoscale light-matter interactions requires both high spatial resolution and anisotropic sensitivity to in-plane and out-of-plane optical responses. We introduce torsional force microscopy-infrared microscopy, a new optical imaging technique that combines cantilever torsional dynamics with a nonlinear frequency-mixing scheme to map both in-plane and out-of-plane photothermal signals. Using birefringent mica as a model system, we resolve distinct in-plane and out-of-plane vibrational responses and reconstruct the anisotropic strain distribution of nanobubbles, in excellent agreement with simulation. Furthermore, we demonstrate near-nanometer ( ~ 1 nm) spatial resolution in optical imaging of twisted bilayer graphene, enabling site-resolved spectroscopy within individual moiré cells. Energy-dependent imaging further reveals intra-unit cell optical features, highlighting the role of competing physical processes in a moiré lattice modulating its optical properties. By providing a direct, anisotropy-resolved view of nanoscale optical heterogeneities, this emerging infrared torsional force microscopy establishes a powerful platform for probing, understanding, and ultimately engineering light-matter interactions in complex quantum materials. Introduction Nanoscale optical imaging is a powerful tool for probing light-matter interactions, resolving spatially localized optical responses that cannot be accessed with conventional far-field methods. Integrating atomic force microscopy (AFM) with optics, such as scattering-type scanning near-field optical microscopy (s-SNOM) and infrared nanospectroscopy (AFM-IR), enables exploration of phenomena including phonon polaritons, excitonic resonances, and nanoscale absorption heterogeneity, with broad applications spanning chemical reactions, biology, and materials science1–7. However, existing AFM-based optical techniques are primarily sensitive to a sample’s out-of-plane response, leaving distinct and complementary in-plane properties largely inaccessible. Additionally, their spatial resolution remains constrained by the AFM tip apex. There is thus a pressing need for ambient-condition methods that combine high spatial resolution, directional sensitivity, and broadband optical access. Previous attempts to address these challenges often required specialized environments, complex instrumentation, or sacrifice optical contrast, thereby limiting insight into active light-matter interactions8–10. Herein, we introduce infrared torsional force microscopy (TFM-IR), a new optical imaging technique that extends TFM11,12 to optical measurements by exploiting the cantilever’s torsional resonance under optical excitation, enabling nanometer-scale optical imaging resolution and selective detection of both lateral (in-plane) and vertical (out-of-plane) components of optically induced forces. By harnessing the inherent anisotropy of tip-sample torsional interactions, directional sensitivity can be selectively enhanced through nonlinear frequency mixing between the cantilever’s torsional resonance modes and the laser repetition rate. This capability enables the first ambient-condition nanoscale direction-resolved photothermal mapping without compromising the broadband spectral access and quantitative detection advantages of standard AFM-IR. We demonstrate reconstruction of in-plane and out-of-plane strain fields in a mica nanobubble with excellent agreement with finite element method (FEM) simulation. Energy-dependent nanobubble imaging reveals variation of anisotropic strain distributions. Furthermore, TFM-IR achieves near-nanometer spatial resolution, enabling spatially resolved spectroscopy within a single moiré unit cell and energy-dependent imaging of twisted bilayer graphene for the first time. Together, these results establish TFM-IR as a broadly applicable technique for nanoscale anisotropic optical imaging under ambient conditions. It offers a new dimension of insight into photothermal and optomechanical interactions with unprecedented spatial resolution, opening a new horizon of nanoscale optical study for diverse areas ranging from two-dimensional materials and nanophotonics to strain engineering and quantum materials13. Results Principles of TFM-IR and direction-resolved nanoscale optical imaging Figure 1a illustrates the schematic setup of TFM-IR. Briefly, a pulsed laser source with repetition rate flaser is coupled to the AFM tip in either top- or bottom- illumination configurations. The tip is mounted on a holder equipped with two piezoelectric actuators at the frequency fm in an out-of-phase configuration, enabling controlled torsional excitation of the cantilever with a tunable driving amplitude. The cantilever motion is monitored by a position sensitive quadrant detector (PSD), whose vertical DC deflection provides tip-sample feedback while the horizontal AC deflection is demodulated by employing multiple lock-in amplifiers for monitoring the driven torsional oscillation amplitude and phase as well as for detecting TFM-IR signals arising from different frequency intermodulation schemes, as exemplified in Fig. 1b. Importantly, in TFM-IR the tip-sample feedback and the driven torsional motion are independently controlled through orthogonal mechanical modes. This decoupling allows the optical signal intensity to be tuned and optimized via the torsional drive amplitude without perturbing the tip-sample feedback. It is worth noting that in conventional AFM-IR and sSNOM techniques based on flexural cantilever motion, the tapping amplitude can influence both the feedback and the optical signal readout. As a result, the measured optical response may remain affected by tip-sample contact mechanics, leading to a degree of residual interdependence between the two channels14.Fig. 1Principle and implementation of TFM-IR.a Schematic of the experimental setup. A tunable pulsed light source is focused onto the cantilever tip via an off-axis parabolic mirror. Two piezoelectric actuators driven out of phase excite the cantilever’s torsion motion, which is detected by a PSD. The PSD signal is routed to a lock-in amplifier to extract the optical response. b Nonlinear frequency mixing scheme for probing sample anisotropy. (Top) Frequency relationship among the pulsed laser repetition rate, the cantilever driving frequency of the m-th torsional mode, and the cantilever detection frequency at the n-th non-driven mode. Even- and odd- harmonic frequency mixing between the driving and detection frequencies is matched to the laser repetition rate, enabling detection of in-plane and out-of-plane sample response, respectively. (Middle) Illustration of the laser pulse sequence with repetition rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}$$\end{document}flaser and two sinusoidal driving waveforms at frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{m}$$\end{document}fm with opposite phase for exciting the m-th torsional mode, as shown in a. (Bottom) Decomposition of the effective sample force exerted on the AFM tip (perspective view) as a function of torsion angle θθ. c Typical TFM-IR image (top) and corresponding line profile averaged over 10 pixels along the profile’s path perpendicular to the flake edge (black curve, bottom). The red curve shows a fit using a sum of two exponentially decaying sinusoidal functions (see Supplementary Note 1 and Methods). Data acquired at 1455 cm–1. Scale bar, 1 µm. Principle and implementation of TFM-IR. a Schematic of the experimental setup. A tunable pulsed light source is focused onto the cantilever tip via an off-axis parabolic mirror. Two piezoelectric actuators driven out of phase excite the cantilever’s torsion motion, which is detected by a PSD. The PSD signal is routed to a lock-in amplifier to extract the optical response. b Nonlinear frequency mixing scheme for probing sample anisotropy. (Top) Frequency relationship among the pulsed laser repetition rate, the cantilever driving frequency of the m-th torsional mode, and the cantilever detection frequency at the n-th non-driven mode. Even- and odd- harmonic frequency mixing between the driving and detection frequencies is matched to the laser repetition rate, enabling detection of in-plane and out-of-plane sample response, respectively. (Middle) Illustration of the laser pulse sequence with repetition rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}$$\end{document}flaser and two sinusoidal driving waveforms at frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{m}$$\end{document}fm with opposite phase for exciting the m-th torsional mode, as shown in a. (Bottom) Decomposition of the effective sample force exerted on the AFM tip (perspective view) as a function of torsion angle θθ. c Typical TFM-IR image (top) and corresponding line profile averaged over 10 pixels along the profile’s path perpendicular to the flake edge (black curve, bottom). The red curve shows a fit using a sum of two exponentially decaying sinusoidal functions (see Supplementary Note 1 and Methods). Data acquired at 1455 cm–1. Scale bar, 1 µm. To demonstrate the feasibility of TFM-IR and compare its performance with existing AFM based optical imaging techniques, Fig. 1c presents a typical TFM-IR image of a 64 nm- thick hexagonal boron nitride (hBN) flake with natural boron abundance exfoliated on a ZnSe substrate. hBN is a polar dielectric material that support surface phonon polaritons (PhPs) through strong photon-phonon coupling within the lower (760–820 cm–1) and upper (1365–1610 cm–1) Reststrahlen bands15,16. Prior studies using either sSNOM or AFM-IR have established that PhPs can be launched by either metallic AFM tips or hBN edges, creating oscillatory optical contrasts across the flake surface17–20. The TFM-IR image in Fig. 1c, acquired at 1455 cm−1 within the upper Reststrahlen band, exhibits oscillatory fringes running parallel to the crystal edges. Line profiles taken perpendicular to the edge (Fig. 1c) show that the fringes are strongest near the flake boundary and gradually decay with increasing distance. A distinctive feature consistently observed in our TFM-IR measurements is the presence of beating patterns in the oscillatory signal, with two distinct periodicities. The shorter-wavelength component decays more rapidly from the edge. Fits to the line profiles (Supplementary Note 1) confirms that the periodicities differ by roughly a factor of two, corresponding to the full and half wavelengths of the hBN PhPs. This beating is caused by the interference of edge-launched and tip-launched, edge-reflected PhPs17–22. For comparison, conventional tapping AFM-IR only revealed edge-launched PhPs, corresponding to the full periodicity of PhPs (Supplementary Fig. 1). Furthermore, we systematically investigated the effects of torsional driving amplitude and feedback conditions (Supplementary Fig. 2). The TFM-IR signal scales strongly with the torsional drive, but is nearly independent of feedback settings, further supporting the working principle outlined in Fig. 1a and 1b. To elucidate the underlying mechanism of TFM-IR, we analytically studied the torsional dynamics of an AFM cantilever (Supplementary Fig. 3) and tip-sample interactions under optical excitation, with detailed discussion provided in Supplementary Note 2. We found that the torsional eigenmodes of an AFM cantilever exhibited a linear dispersion of eigenfrequencies, in contrast to the quadratic dispersion of flexural modes23. Under tip-sample interaction, the dynamics of the n-th torsional mode can be modeled as a damped-driven harmonic oscillator governed by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ddot{\theta }}_{n}-\gamma {\dot{\theta }}_{n}+{\omega }_{n}^{2}{\theta }_{n}={F}_{{eff}}\left(t\right)$$\end{document}θ¨n−γθ°n+ωn2θn=Fefft, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta }_{n}$$\end{document}θn is the torsional angle, γγ is the damping coefficient, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{n}$$\end{document}ωn is eigenfrequency, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{eff}}\left(t\right)$$\end{document}Fefft is the effective force arising from tip-sample interaction that drives torsional motion. We specifically evaluated the experimental configuration in Fig. 1, where the cantilever is driven at the m-th mode with frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{f}_{m}=\omega }_{m}/2\pi$$\end{document}fm=ωm/2π and amplitude \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{m}$$\end{document}Am, while the sample is periodically excited by a pulsed laser at repetition rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}$$\end{document}flaser. This optical excitation induces a modulated tip-sample force, for example, via the photothermal effect, which can be decomposed into in-plane and out-of-plane components (Fig. 1b) determined by the anisotropic nature of sample. Accordingly, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{eff}}\left(t\right)$$\end{document}Fefft can be expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{F}_{{eff}}\left(t\right) =F}_{{in}}\left({f}_{{laser}},t\right)\cos \left[\,{A}_{m}\cos \left({2\pi f}_{m}t\right)\right] +{F}_{{out}}\left({f}_{{laser}} ,t\right)\sin \left[{A}_{m}\cos \left({2\pi f}_{m}t\right)\right]$$\end{document}Fefft=Finflaser,tcosAmcos2πfmt+Foutflaser,tsinAmcos2πfmt, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{in}}$$\end{document}Fin and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{out}}$$\end{document}Fout are the laser-modulated in-plane and out-of-plane forces, respectively. While the orthogonality of modes typically suppresses coupling between the driven m-th mode and other non-driven n-th modes, we show in the Supplementary Note 2 that the trigonometric decomposition of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{eff}}\left(t\right)$$\end{document}Fefft enables intermodulation between the optical excitation and the driven m-th mode, and this nonlinear frequency coupling can resonantly excite non-driven torsional modes at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{f}_{n}=\omega }_{n}/2\pi$$\end{document}fn=ωn/2π. Specifically, under conditions of even (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}={f}_{n}\pm 2k{f}_{m}$$\end{document}flaser=fn±2kfm) or odd (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}={f}_{n}\pm \left(2k-1\right){f}_{m}$$\end{document}flaser=fn±2k−1fm) harmonic frequency mixing (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=1,\,2,\,3,\,\ldots$$\end{document}k=1,2,3,…), the in-plane or out-of-plane optically induced forces produce a detectable resonance response at the non-driven n-th torsional mode. A key implication of this mechanism is that by judiciously tuning the laser repetition rate to meet the even or odd harmonic mixing condition, TFM-IR can selectively probe in-plane or out-of-plane sample anisotropy as shown in Fig. 1b. This unique capability makes TFM-IR an indispensable and general approach for exploring anisotropic properties, such as strain and related phenomena, with ultra-high spatial and energy resolution that are otherwise challenging to access8–10. Resolving anisotropic strain at the nanoscale: experiment and simulation To highlight the unique capability of TFM-IR in probing nanoscale anisotropic properties, we selected natural muscovite mica as an example, which is a well-known birefringent material that exhibits characteristic in-plane and out-of-plane refractive indices in the spectra range of 900−1100 cm–1 (ref. 24,25). Conventional methods for studying birefringence or optical anisotropy, such as ellipsometry and polarization-dependent Fourier transform infrared spectroscopy, are fundamentally limited in spatial resolution by far-field diffraction. More recently, techniques that combine polarized light with scanning probes, including sSNOM and AFM-IR, have been developed to overcome this limitation, yet often require careful consideration of unavoidable tip-induced perturbations to the incident polarization state or specialized instrumentation adapted to particular materials8,9,26–28. Figure 2a presents two typical TFM-IR spectra collected from the same location but with different laser repetition rates on a 200 nm- thick mica flake exfoliated onto a ZnSe substrate. The spectra were obtained by driving and detecting the cantilever at the second and first torsional modes (m = 2; n = 1), while tuning the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}$$\end{document}flaser to either \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}={f}_{{out}}={f}_{2}-{f}_{1}$$\end{document}flaser=fout=f2−f1 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}={f}_{{in}}={2f}_{2}-{f}_{1}$$\end{document}flaser=fin=2f2−f1. Accordingly, we refer to subsequent measurements under these two configurations as the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- spectra/maps.Fig. 2Anisotropy-sensitive mapping of nanoscale strain in mica nanobubbles.a (Left) Two typical TFM-IR spectra acquired from the same flat mica region (blue dot in the inset). Blue and red circles are experimental \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- data, respectively. Blue and red curves are global fits using four characteristic peaks centered at ~963 cm−1, 995 cm−1, 1022 cm–1 and 1050 cm–1 (Supplementary Fig. 4). The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- spectra was normalized to the intensity of the 995 cm–1 peak. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- spectra was normalized to the intensity of the 1050 cm–1 peak. Inset, three-dimensional height profile of a typical mica nanobubble. Scale bar, 1 µm. (Right top) and (Right bottom) are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- TFM-IR images acquired at 1045 cm–1 from the mica nanobubble in the inset, respectively. Scale bar, 1 µm. b (Left top) and (Left bottom) are reconstructed in-plane and out-of-plane maps obtained from the images in a after crosstalk correction. Scale bar, 1 µm. (Right top) and (Right bottom) are computed in-plane and out-of-plan strain displacement distributions for the nanobubble in a by FEM, respectively. c Radial profiles by azimuthal averaging along constant radii from the nanobubble center to the edge, extracted from maps in b. Solid curves: experimental in-plane (red) and out-of-plane (blue) profiles. Dashed curves: FEM simulated in-plane (red) and out-of-plane (blue) profiles. For comparison with experimental data, simulated profiles are normalized by a factor of 440. Anisotropy-sensitive mapping of nanoscale strain in mica nanobubbles. a (Left) Two typical TFM-IR spectra acquired from the same flat mica region (blue dot in the inset). Blue and red circles are experimental \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- data, respectively. Blue and red curves are global fits using four characteristic peaks centered at ~963 cm−1, 995 cm−1, 1022 cm–1 and 1050 cm–1 (Supplementary Fig. 4). The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- spectra was normalized to the intensity of the 995 cm–1 peak. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- spectra was normalized to the intensity of the 1050 cm–1 peak. Inset, three-dimensional height profile of a typical mica nanobubble. Scale bar, 1 µm. (Right top) and (Right bottom) are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- TFM-IR images acquired at 1045 cm–1 from the mica nanobubble in the inset, respectively. Scale bar, 1 µm. b (Left top) and (Left bottom) are reconstructed in-plane and out-of-plane maps obtained from the images in a after crosstalk correction. Scale bar, 1 µm. (Right top) and (Right bottom) are computed in-plane and out-of-plan strain displacement distributions for the nanobubble in a by FEM, respectively. c Radial profiles by azimuthal averaging along constant radii from the nanobubble center to the edge, extracted from maps in b. Solid curves: experimental in-plane (red) and out-of-plane (blue) profiles. Dashed curves: FEM simulated in-plane (red) and out-of-plane (blue) profiles. For comparison with experimental data, simulated profiles are normalized by a factor of 440. These two spectra are correlated but exhibit clear distinction. First, both can be consistently decomposed into four peaks centered at ~963 cm−1, 995 cm–1, 1022 cm–1 and 1050 cm–1 (Supplementary Fig. 4). These observed features are consistent with established lattice dynamics and spectroscopic analyses of muscovite mica24,25,29,30. Among them, the ~995 cm–1 peak is attributed to an apical (out-of-plane) Si-O \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{1}^{1}$$\end{document}a11 stretching mode involving the apical oxygens of the SiO4 tetrahedra. The other three peaks correspond to basal (in-plane) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}_{1}^{1}$$\end{document}e11 vibrations within the tetrahedral sheets. Variations in bond geometry and local cation substitutions split the in-plane Si-O stretching manifold into multiple components24,25: the ~963 cm–1 feature is associated with weaker, more localized Si-O-Si stretching involving bridging oxygens shared between adjacent tetrahedra; the ~1022 cm–1 peak corresponds to asymmetric Si-O stretching in mixed Si-O-Si and Si-O-Al environments; and the ~1050 cm–1 peak reflects the stronger, more collective asymmetric Si-O stretching vibrations along the tetrahedral-sheet backbone. It is well established that natural muscovite mica is chemically complex, and compositional variability (e.g., Fe or Cu staining, Al/Si substitutions, or variations in octahedral site occupancy) can alter local force constants and shift vibrational frequencies by several to a few tens of wavenumbers for the same vibrational modes30–33. Importantly, this variability does not alter the relative ordering of out-of-plane versus in-plane modes, which is reproducible across studies. Our observation of four distinct and reproducible features, together with their relative positions, is therefore consistent with the established distinction between out-of-plane ( ~ 995 cm−1) and in-plane ( ~ 963, 1022 and 1050 cm–1) vibrational modes in layered aluminosilicates, supporting our peak identification; Second, the relative intensities of these peaks differ between the two spectra. In the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- spectrum, the out-of-plane 995 cm–1 peak dominates, whereas in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- spectrum, the three in-plane peaks are strongly enhanced proportionally. This observation is consistent with the TFM-IR scheme in Fig. 1b, where different odd- and even- harmonic frequency mixing pathways selectively probe out-of-plane and in-plane characteristics of a sample, respectively. To further validate this mechanism, we performed a control experiment by systematically varying the driving amplitude of the m = 2 mode while recording TFM-IR spectra under otherwise identical conditions. Comparing the evolution of the 995 cm–1 out-of-plane peak intensity from the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- spectra (Supplementary Fig. 5a) with that of the 1050 cm–1 in-plane peak from the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- spectra (Supplementary Fig. 5b), we find a clear contrast: the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- peak scales linearly with the driving amplitude \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{m}$$\end{document}Am, whereas the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- peak follows a quadratic dependence, which is predicted by TFM-IR theory in Supplementary Note 2. Taken together, these results provide strong experimental validation of our TFM-IR technique and demonstrate excellent agreement with theoretical predictions, highlighting its potential to enable new applications in diverse materials systems through tailored nonlinear frequency mixing schemes. For our exfoliated mica samples, we observed scattered nanobubbles of various sizes (Supplementary Fig. 6a). Figure 2a shows a typical three-dimensional AFM height profile of a mica nanobubble, along with its corresponding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- TFM-IR maps acquired at 1045 cm–1. While its height profile shows a smooth dome-like morphology with a maximum at its center, the TFM-IR images manifest a strikingly different annular spatial pattern characterized by a distinct optical intensity drop at the nanobubble center. In general, absorption of IR radiation induces local photothermal expansion, which exerts a force and torque on the AFM tip. The stark disparity between the TFM-IR image and the topographic profile indicates that the TFM-IR image provides a direct and unique fingerprint of the nanobubble’s inhomogeneous strain distribution arising naturally from its geometry34 (Supplementary Note 3). Importantly, the unique anisotropic sensitivity of TFM-IR enables direct mappings of both in-plane and out-of-plane strain. Analogous to spectroscopy, these images can be associated with in-plane and out-of-plane strain responses, respectively. Indeed, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- maps in Fig. 2a exhibit distinct features, although further analysis is required to fully unravel the underlying physics discussed below. Although the results in Fig. 2a confirm the ability of TFM-IR to preferentially detect in-plane or out-of-plane vibrational modes by the nonlinear frequency mixing, the appearance of all four characteristic peaks in both the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- maps indicates crosstalk between these two detection channels. For example, the out-of-plane vibrational signal at 995 cm–1 does not vanish in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- spectra, and vice versa. Such arises from a combination of physical and instrumental factors, as discussed in Supplementary Note 4. We characterized and corrected the crosstalk by a linear instrument response matrix, which can be calibrated by comparing the relative intensities of characteristic in-plane and out-of-plane peaks in the two channels. We extracted a crosstalk ratio (R) from the calibration, enabling a linear unmixing correction to reconstruct accurate, anisotropy-resolved photothermal maps and spectra (see Supplementary Note 4 for details). For the mica nanobubble in Fig. 2a, we selected the 995 cm–1 out-of-plane peak and the 1050 cm−1 in-plane peak as reference marker. The crosstalk ratios Rin and Rout were 0.814 and 0.538, respectively (Supplementary Note 4). Using these values, we reconstructed both the in-plane and out-of-plane spectra (Supplementary Fig. 7) as well as the two-dimensional maps (Fig. 2b) derived from the raw \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{in}}$$\end{document}fin- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{out}}$$\end{document}fout- data in Fig. 2a. Radial line profiles extracted from reconstructed maps along the nanobubble radius are also presented in Fig. 2c for direct comparison. After crosstalk correction, the spectral weights of the in-plane and out-of-plane components increase significantly in their corresponding spectra, by approximately fourfold and twofold, respectively. Furthermore, the reconstructed maps reveal pronounced and distinct features between the in-plane and out-of-plane. First, both maps show a bright annular maximum surrounding the nanobubble, but with a radial offset from the geometric center. However, the annular maximum in the out-of-plane map lies closer to the nanobubble apex ( ~ 0.3 µm from the center), whereas in the in-plane map it shifts farther outward toward the perimeter ( ~ 0.7 µm) with a shoulder peak near ~0.2 µm. Second, in the in-plane map, the lowest intensity occurs outside the bright ring, decaying toward the edge, producing a hollow-ring feature. In contrast, the out-of-plane map displays negative signals at the apex and adjacent to the bright annulus, resulting in a dipolar-like contrast pattern; This reflects the alternating compressive and tensile regions expected from the dome-like geometry. Third, the bright ring feature in the in-plane map spans a relatively broader radial extent, reflecting strain localization near the nanobubble edge where lattice distortion is maximized. The out-of-plane bright ring is narrower and accompanied by surrounding negative regions, indicating more localized strain relaxation in the out-of-plane direction. These results demonstrate that TFM-IR separately resolves in-plane stretching and out-of-plane curvature-induced strain, capturing complementary strain components generated by the nanobubble geometry. To substantiate this interpretation, we performed a FEM simulation of the strain- displacement distribution by directly importing the measured height profile of the nanobubble from Fig. 2a and reconstructing it on a flat mica disk (Supplementary Fig. 8). Figures 2b and 2c show the simulated maps and their radial profiles alongside the experimental data, respectively. The FEM results closely reproduce the key experimental observations: the out-of-plane strain exhibits a negative central region surrounded by a positive annulus, producing a dipolar-like contrast, while the in-plane strain remains positive and extends more broadly outward with a characteristic hollow-ring feature. Radial profiles extracted from both experiment and simulation further demonstrate excellent quantitative agreement, capturing the relative positions of the maxima, including the out-of-plane peak located closer to the center and the in-plane peak shifted outward to ~0.7 µm with a secondary peak near ~0.2 µm, as well as their relative widths. The strong agreement between experiment and simulation validates the conclusion that TFM-IR images directly reflect the anisotropic mechanical strain generated by the nanobubble geometry, enabled by the nonlinear frequency mixing scheme in Fig. 1b. Notably, the experimental maps exhibit broader linewidth and less sharply defined boundaries compared to FEM predictions, which is likely due to experimental broadening effects from finite tip size and thermal diffusion. The simulated strain distributions of the ideal nanobubble also appear more symmetric and uniform, whereas the experimental maps display slight asymmetries and irregular brightness, suggesting imperfections in nanobubble geometry under realistic conditions or local material inhomogeneities. These discrepancies highlight the importance of experimental probes such as TFM-IR to disentangle in-plane and out-of-plane strain components with nanoscale sensitivity, offering a powerful tool for mapping local strain landscape in quantum and van der Waals materials. The in-plane and out-of-plane TFM-IR images in Fig. 2 reveal the intrinsic anisotropic strain distribution of a nanobubble, establishing TFM-IR as a fingerprinting tool for a sample’s anisotropic thermo-mechanical response. Qualitatively, this behavior arises because wavelength-dependent absorption and subsequent thermalization generates both vertical and lateral expansion fields (Supplementary Note 3). Changing the optical excitation wavelength alters their relative contributions, leading to a wavelength-dependent redistribution of strain. Figure 3 shows the evolution of reconstructed in-plane and out-of-plane TFM-IR maps as a function of laser wavelength. Across the probed spectral window, both reconstructions consistently display annulus strain features centered on the nanobubble, confirming that the underlying mechanical strain distribution is robust and primarily geometry-driven rather than tied to a single vibrational resonance. Nevertheless, finer details, such as spatial extent, symmetry and intensity distribution, depend strongly on excitation frequency. For example, the absolute amplitude of both in-plane and out-of-plane maps varies with wavenumber, likely reflecting variation in IR absorption strength and photoelastic coupling to distinct vibrational modes (Supplementary Note 3). At certain wavenumbers (e.g., 990 cm−1 and 1020 cm−1), the in-plane annulus appears sharper and narrower, whereas at others the response is broader and less defined. This suggests that the effective interaction volume and strain readout are modulated by phonon-mode dependent resonant enhancement. Interestingly, the out-of-plane channel occasionally exhibits strong azimuthal asymmetry across the nanobubble (e.g., at 995 cm−1), particularly in the region adjacent to the positive annulus rather than as a direct modulation of the positive band itself. This points to local polarization sensitivity tied to specific out-of-plane resonance. Overall, observed spectral variations in intensity, sharpness and symmetry underscore how strain couples differently to distinct vibrational modes. By scanning across wavenumber, TFM-IR not only reconstructs nanoscale strain maps but also provides access to frequency-dependent photoelastic tensor elements, disentangling in-plane and out-of-plane strain sensitivity.Fig. 3Wavenumber-dependent in-plane and out-of-plane strain maps of a mica nanobubble.Top and bottom panels show reconstructed in-plane and out-of-plane maps, respectively, acquired at the indicated wavenumbers: a 910 cm–1. b 964 cm–1. c 990 cm−1. d 995 cm–1. e 1020 cm–1. f 1090 cm–1. All data were obtained from the same nanobubble presented in Fig. 2. Maps were reconstructed using the same crosstalk correction procedure applied for the 1045 cm–1 measurements in Fig. 2. The evolution of the strain patterns with wavenumber highlights the wavelength-dependent photothermal response and anisotropic strain distribution. Scale bar, 1 µm. Wavenumber-dependent in-plane and out-of-plane strain maps of a mica nanobubble. Top and bottom panels show reconstructed in-plane and out-of-plane maps, respectively, acquired at the indicated wavenumbers: a 910 cm–1. b 964 cm–1. c 990 cm−1. d 995 cm–1. e 1020 cm–1. f 1090 cm–1. All data were obtained from the same nanobubble presented in Fig. 2. Maps were reconstructed using the same crosstalk correction procedure applied for the 1045 cm–1 measurements in Fig. 2. The evolution of the strain patterns with wavenumber highlights the wavelength-dependent photothermal response and anisotropic strain distribution. Scale bar, 1 µm. We note that, although the mica nanobubble in Fig. 2a is employed as an example to demonstrate the unique capability of TFM-IR to resolve nanoscale anisotropy, our conclusions are not specific to a single structure. Supplementary Fig. 6b, acquired simultaneously with Supplementary Fig. 6a at 1045 cm−1, shows that all the mica nanobubbles exhibit similar optical features, including a reduction of optical intensity at the nanobubble center, confirming the underlying mechanical strain mechanism for a nanobubble structure. Furthermore, Supplementary Fig. 9 provides data from another nanobubble with different height and width, obtained using the same in-plane and out-of-plane imaging and reconstruction methods. The corresponding results reveal consistent fundamental in-plane and out-of-plane radial distributions, as in Figs. 2 and 3, with additional fine-scale asymmetries and inhomogeneous features attributed to the specific geometry of each nanobubble. The observation of reproducible features across different nanobubbles both validates the underlying physics and our data analysis approach and highlights the generality of TFM-IR as a sensitive tool for probing nanoscale anisotropy. By combining spectral tunability with anisotropic spatial resolution TFM-IR is uniquely suited to capture robust, geometry-induced strain morphology while simultaneously revealing phonon resonance dependent enhancements or suppressions of specific strain components. Although more comprehensive simulation combining optical absorption and strain-field calculations will be needed to fully elucidate the mechanisms behind the observed spectral evolution, Figs. 2 and 3 together establish that TFM-IR offers a powerful new route to map strain fields and probe anisotropic coupling pathways at the nanoscale. Near-nanometer spatial resolution in TFM-IR imaging In addition to enabling nanoscale anisotropy-sensitive imaging, TFM-IR builds on the intrinsic high-resolution capability of TFM, which has achieved atomic lattice resolution under ambient conditions11. However, extending this level of resolution to optical response measurements has not yet been achieved35. Moiré lattices, which are formed when atomically thin crystals are stacked with a slight twist or lattice mismatch, are a class of potentially useful quantum material interactions36–39. These long-wavelength superlattices dramatically modify electronic bands, phonons, and exciton states, creating varied phenomena such as correlated insulating states, unconventional superconductivity, and tunable optical responses40–46. It is worth noting that, although most vibration modes in the mid-IR regime are IR inactive in far-field optical spectroscopy of pristine monolayer graphene due to symmetry constraints and strict optical selection rules, both experimental and theoretical studies have shown that under symmetry-breaking conditions such as twisting, stacking, straining or electrostatic gating in bilayer and multilayer graphene, phonon dispersion can be modified and enhanced with manifestation of folded phonons47–53. Moreover, tip-based near-field experimental techniques can further relax the selection rules for phonons originally at finite momentum, thereby enabling mid-IR optical activity in bilayer and multilayer graphene systems, even for modes that are otherwise forbidden in pure monolayer graphene3,54,55. Resolving these moiré patterns with optical information at atomic or near-nanometer resolution is therefore impactful, as it facilitates direct mapping of spatially varying optical absorption, exciton localization, and strain-modulated interactions within a single moiré cell56. Experimentally, achieving this has been highly challenging. For example, sSNOM studies of twisted bilayer graphene typically achieve spatial resolutions of a few tens to hundreds of nanometers, with the tip apex size imposing a fundamental limitation54–59. Figure 4a presents simultaneously acquired TFM amplitude and TFM-IR images of a twisted bilayer graphene sample at 1583 cm−1, along with their line profiles. The torsional amplitude reflects the cantilever’s torsional resonance at the driving frequency, with variations arising from tip-sample damping forces determined by local friction, stiffness, and van der Waals interactions11,60. The TFM amplitude image clearly shows the moiré’s honeycomb lattice, and fitting the amplitude profile yields a lattice periodicity of 6.1 nm, corresponding to a 2.4° twist angle.Fig. 4TFM-IR imaging and site-resolved spectroscopy of a twisted bilayer graphene moiré lattice.a Simultaneously acquired TFM (left) and TFM-IR (right) images of the same moiré lattice, with corresponding line-profiles (black curves) along the white line in the images. The TFM-IR image was acquired at 1583 cm–1. Red curve in the TFM line-profile shows a fit used to determine the twisting angle (see Methods). Scale bar, 10 nm. b TFM-IR spectra acquired from 15 distinct locations within a single moiré unit cell (inset). Colored symbols in the spectra correspond to their spatial positions indicated in the inset TFM image. The spectra cluster into three groups with distinct amplitudes, represented by blue, red and black. Inset, TFM image of a single moiré unit cell showing the measurement locations. Scale bar in inset, 5 nm. TFM-IR imaging and site-resolved spectroscopy of a twisted bilayer graphene moiré lattice. a Simultaneously acquired TFM (left) and TFM-IR (right) images of the same moiré lattice, with corresponding line-profiles (black curves) along the white line in the images. The TFM-IR image was acquired at 1583 cm–1. Red curve in the TFM line-profile shows a fit used to determine the twisting angle (see Methods). Scale bar, 10 nm. b TFM-IR spectra acquired from 15 distinct locations within a single moiré unit cell (inset). Colored symbols in the spectra correspond to their spatial positions indicated in the inset TFM image. The spectra cluster into three groups with distinct amplitudes, represented by blue, red and black. Inset, TFM image of a single moiré unit cell showing the measurement locations. Scale bar in inset, 5 nm. The complementary TFM-IR image at 1583 cm−1 reveals dramatically different spatial features. Despite this, the overall periodicity of the optical image still matches the honeycomb lattice well. We performed a control experiment by tuning the optical excitation to an off-resonance wavelength while keeping all other imaging conditions identical. In this case, the previously observed optical features disappeared (Supplementary Fig. 10), confirming that the TFM-IR signals in Fig. 4a are not caused by mechanical crosstalk from TFM, but instead originate from a genuine optical response. Notably, without applying any image filters or processing, the full-width at half-maximum (FWHM) of the smallest features in the line profile of Fig. 4a is ~ 1 nm, which is an order of magnitude smaller than the typical radius of tip apex ( ~ 25 nm) that was utilized for acquiring the data. The underlying mechanism of the observed super-resolution in TFM-IR imaging remains to be fully elucidated and will require systematic studies and modeling, but it is likely linked to the unique tip-sample interactions that govern the intrinsic resolution of TFM11. The spatial resolution in TFM-IR is qualitatively determined by the detection mechanism, which transduces localized optical responses into torsional motion of the cantilever. Unlike conventional sSNOM and AFM-IR techniques, where the measured optical signal is effectively averaged over a finite tip-sample interaction volume and is thus limited by the tip apex size and thermal diffusion, the TFM-IR signal originates from cantilever’s torsional motion driven by lateral force asymmetries (i.e., torque) within the tip-sample contact region (see Supplementary Note 2). This detection mechanism is inherently sensitive to spatial variations and gradients in the photothermal response, rather than its absolute magnitude. Such gradient-sensitive detection suppresses slowly varying background contributions while enhancing localized variations, thereby increasing sensitivity to high-spatial frequency components and enabling a spatial resolution much smaller than the nominal tip-sample interaction size. The exceptional spatial resolution in TFM-IR imaging as demonstrated in Fig. 4a immediately opens new opportunities for spatially resolved optical studies. To explore the intra-cell variation in greater detail, we performed spatially resolved spectroscopy at 15 different positions spanning from the edges to the center of a single moiré unit cell (Fig. 4b). Remarkably, the spectra segregate cleanly into three distinct clusters according to their spectral amplitude. Specifically, the six edges of the moiré honeycomb lattice group into three pairs of neighboring edges, with each pair exhibiting similar amplitude, while the cell center also shows the largest overall amplitude. This clear grouping highlights a pronounced site dependence of optical response within a single moiré unit. Qualitatively, such spatial variation likely arises from differences in local stacking configurations (e.g., AA vs. AB/BA) across the moiré pattern, which are known to modulate interlayer coupling, electronic density of states, and light-matter interactions41,61. The emergence of distinct amplitude groups within a single cell may reflect the discrete set of symmetry-related local stacking environments, where paired edges correspond to equivalent regions of the moiré potential landscape, and the center corresponds to a high-symmetry stacking that maximizes optical response. Our observation of inhomogeneous optical response within a single moiré cell is significant. It demonstrates that the optical properties of moiré materials can vary dramatically even at the sub-unit cell scale, underscoring the importance of nanoscale optical probes in uncovering hidden intra-cell heterogeneity. Beyond the super-resolution capability of TFM-IR imaging, which enables fine optical details within a single moiré cell to be resolved, energy-dependent high-resolution imaging can also be performed. Figure 5 presents four additional TFM-IR images acquired at different wavenumbers, complementing the data set in Fig. 4a for the same area. TFM images acquired simultaneously with the TFM-IR data in Fig. 5 remain unchanged and consistent with those features in Fig. 4a. In contrast, the TFM-IR images exhibit a clear evolution of optical patterns with wavenumber, leading to several key observations: First, the periodicity of the moiré lattice in Fig. 4a is preserved across all images. The hexagonal symmetry is maintained globally, yet within each unit cell the intensity pattern breaks the six-fold rotational symmetry; Second, for each wavenumber, the hexagonal unit cell hosts recurring localized features or motifs whose relative positions within the cell remain consistent across the lattice. These motifs are coherent over multiple unit cells and not random noise or isolated local anomalies; Third, there are two types of local optical motifs, dot-like hotspots with two-fold symmetry and triangular patch-like features with three-fold symmetry, whose occurrence depends on the wavenumber across the dataset; Fourth, the intensities of the local motifs vary with wavenumber, with some conditions showing sharper, more pronounced hotspots and others appearing more diffuse with reduced amplitude. Importantly, the wavelength-dependent evolution of the TFM-IR signal, both in overall intensity and in the spatial redistribution of contrast, together with its clear decoupling from the unchanged mechanical TFM channels across all measurements, strongly suggests that the observed nanoscale optical features arise from genuine optical (photothermal) responses that vary with excitation energy. This behavior is inconsistent with mechanical crosstalk or from a static, mechanically mediated readout of tip-sample forces.Fig. 5Wavenumber-dependent TFM-IR imaging of a twisted bilayer graphene moiré lattice.a 1546 cm–1. b 1554 cm–1. c 1594 cm–1. d 1597 cm–1. All images were acquired from the same area shown in Fig. 4 under identical conditions, except for the excitation wavenumber. Their simultaneously acquired TFM images show no discernible change compared to Fig. 4a, confirming that the tip condition remained stable across all measurements. Scale bar, 10 nm. Wavenumber-dependent TFM-IR imaging of a twisted bilayer graphene moiré lattice. a 1546 cm–1. b 1554 cm–1. c 1594 cm–1. d 1597 cm–1. All images were acquired from the same area shown in Fig. 4 under identical conditions, except for the excitation wavenumber. Their simultaneously acquired TFM images show no discernible change compared to Fig. 4a, confirming that the tip condition remained stable across all measurements. Scale bar, 10 nm. Several mechanisms have been theoretically proposed to account for energy-dependent nanoscale optical heterogeneity, arising from the interplay of stacking configuration, strain and interlayer electronic coupling. The aforementioned local stacking configuration can lead to spatial variations in interlayer hybridization, symmetry, and local electronic structure, which in turn modulate local optical absorption and electron-phonon coupling, with supporting evidence from the site-resolved spectroscopy in Fig. 4b. Importantly, such variations can alter both the spectral weight and selection rules of optical transitions, resulting in spatially varying infrared responses that evolve with excitation energy62,63. In addition, strain fields induced by lattice relaxation and interlayer interactions are non-uniformly distributed across the moiré unit cell, often concentrating near domain boundaries or specific stacking regions. These strain variations can locally shift phonon frequencies and modify electron-phonon coupling strength, thereby changing the resonance conditions for infrared absorption64,65. As a result, different regions within a moiré unit cell can exhibit distinct spectral responses, giving rise to energy-dependent contrast in TFM-IR imaging. Furthermore, the interlayer electronic coupling varies spatially across the moiré lattice, leading to direct modulation of the local band structure, including the emergence of flat bands and spatially localized electronic states. These variations can impact optical transitions, even in the mid-infrared regime, where dipole-allowed transitions are typically weak. In particular, symmetry breaking and interlayer hybridization can activate otherwise weak or forbidden transition through indirect processes such as phonon-assisted absorption or coupling to localized electronic states, contributing to spatially heterogeneous optical responses, which are inherently energy dependent63,66. Importantly, these mechanisms are strongly intertwined at the microscopic level and collectively govern a spatially and spectrally varying optical response, providing the basis for the wavelength-dependent phenomena observed in Fig. 5. The preservation of moiré periodicity across all excitation wavelengths in the TFM-IR images further suggests that the observed responses arise from competing processes that vary within a single unit cell, while the emergency of energy-dependent local optical motifs and symmetries reflect domain-specific responses. For example, dot-like hotspots with two-fold symmetry in Fig. 5a and 5c may be associated with domain junctions where three stacking registries meet in a honeycomb lattice, whereas triangular patches in Fig. 5b and 5d may indicate an extended sub-domain response arising from competition between two inequivalent sub-domains (AB vs. BA) with distinct complex susceptibilities65,67–70. A comprehensive theoretical framework that integrates strain-field distributions, stacking-dependent electronic structure, and optical response will be necessary to fully elucidate the sub-moiré-scale optical features and their energy dependence observed in Figs. 4 and 5. In particular, capturing the interplay of interlayer hybridization, symmetry breaking, and electron-phonon coupling across the moiré unit cell likely calls for multi-scale approaches that bridge continuum descriptions with atomistic modeling. The experimental capabilities demonstrated here, combining ultra-high spatial resolution with spectral sensitivity in TFM-IR imaging, provide a new opportunity to directly probe these coupled effects, thereby offering a platform to guide and benchmark future theoretical developments. Discussion We have established TFM-IR imaging as a versatile, ambient condition platform for nanoscale optical characterization. By harnessing a cantilever’s torsional driving and detection in a nonlinear frequency mixing scheme, this approach enables mapping of in-plane and out-of-plane photothermal response, granting direct access to nanoscale anisotropies (e.g., local strain), a capability largely inaccessible to conventional sSNOM and AFM-IR methods1. Furthermore, achieving near-nanometer ( ~ 1 nm) spatial resolution, we demonstrate in bilayer graphene moiré lattices that site-resolved spectroscopy within a single moiré unit cell reveals distinct, spatially dependent optical responses. Moreover, energy-dependent imaging coupled with ultra-high spatial resolution uncovers sub-unit cell optical variations, shedding light on the complex interplay of different physics including local stacking, strain, and electronic structure in moiré lattice. Looking forward, TFM-IR offers an unprecedented window into site-specific and anisotropic light-matter interactions across a broad range of quantum and van der Waals materials71,72. The ability to directly correlate nanoscale optical response with local structural and electronic heterogeneity at near-atomic scales provides a pathway to not only map but also ultimately engineer site-specific functionalities73,74. Beyond fundamental insights, the combination of anisotropy-resolved detection, quantitative spectral fingerprinting, and nanometer spatial precision positions TFM-IR as a promising characterization platform for advanced nanomanufacturing and semiconductor metrology, where accurate monitoring of strain, interface chemistry, and materials anisotropy is critical for device reliability and performance75. While the present implementation of TFM-IR operates in contact mode, the generic decoupling between tip-sample feedback and torsional-based optical detection through orthogonal mechanical modes immediately suggests a natural extension to tapping or intermittent-contact imaging12. Such developments could further broaden their applicability across diverse materials systems. By providing direction-resolved, broadband optical information under ambient conditions, TFM-IR bridges the gap between laboratory-scale nanoscience and industrial-scale process control, offering powerful framework for guiding the design, optimization, and quality assurance of next generation nanoelectronic, nanophotonic and quantum devices, where local optical properties and functionalities are dictated by sub- unit cell variations37,76. Methods Materials The hBN flakes (PT110A) used for PhPs imaging were obtained from Momentive Performance Materials. The mica used in this study is natural muscovite mica, iron- and copper-stained with exceptional density, purchased from Asheville-Schoonmaker Mica Co. Both hBN and mica were mechanically exfoliated 3–5 times using Nitto SPV224 PCV vinyl surface protection tape and subsequently transferred onto optically polished, uncoated ZnSe substrates, which were cleaned using First-Contact Polymer Solution. The twisted bilayer graphene sample was prepared using a modified tear-and-stack method (Supplementary Fig. 11)77–79. During fabrication, a PDMS/PPC stamp mounted on a glass slide was first used to pick up a ~ 100 nm thick hBN flake, followed by the first graphene layer. The second graphene layer was then picked up after a controlled rotation of approximately 2.4°, thereby defining the twist angle between the two graphene layers. The hBN flake was intentionally misaligned relative to the graphene layers to minimize substrate-induced moiré effects. The assembled stack was then manually delaminated from the PDMS, flipped so that the bilayer graphene faced upward, and transferred onto a polished, uncoated ZnSe substrate, which were cleaned using First-Contact Polymer Solution. Finally, the sample was annealed at 250 °C for 2 h to remove polymer residues. All samples were mounted on a home-built sample stage using double-sided carbon tape. Moiré imaging was performed with PPP-NCHPt tips (from NanoAndMore USA Corp). Mica and hBN imaging were done with PPP-CONTSCAu tips (from NanoAndMore USA Corp). Cantilever contact force was calculated by engaging the tip on a clean Si wafer, performing a single height ramp, and fitting the contact curve’s slope to get the deflection sensitivity in mV per nanometer moved vertically (mV/nm). PPP tips come with the manufacturer’s measured spring-constant (N/m) for each tip. Dividing the second by the first gives a direct conversion between the deflection setpoint and the tip-sample force. TFM-IR measurements TFM-IR measurements were performed on a Bruker Dimension Icon AFM equipped with custom-designed optical path and a 3D-printed sample stage that enables optical excitation from either above or below the sample. For moiré imaging, once the tip was engaged with the sample, the driving amplitude and deflection setpoint were incrementally optimized to achieve the desired contrast. Optimal imaging conditions were typically obtained with contact forces in the range of 150–250 nN and driving amplitudes of 20–30 mV. For hBN and mica measurements, the driving amplitude and contact force were set to 20 mV and 50 nN, respectively, and subsequently adjusted to optimize the TFM-IR signal. The cantilever’s driving frequency was controlled via an external program interfaced with a RIGOL DG1022 dual-output function generator. One output channel provided a sinusoidal signal at the specified lock-in reference frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{lock}-{in}}$$\end{document}flock−in, corresponding to a torsional resonance distinct from that used for TFM imaging, and was sent to a Stanford Research Systems’ SR844 RF Lock-In amplifier. The second output channel generated TTL pulses at frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{laser}}$$\end{document}flaser to trigger the IR source. The horizontal signal from the PSD was routed through the Bruker’s Signal Access Module (SAM) to the SR844, and the demodulated output was fed into the NanoScope 6 controller to generate TFM-IR images simultaneously with other imaging channels in real-time. During systematic studies involving variation of optical parameters (e.g., excitation wavelength or power), the simultaneously acquired TFM and height images were continuously monitored to ensure consistency and to exclude artifacts arising from changes in imaging conditions. Three tunable pulsed IR sources were used. Two were complimentary Daylight Solutions’ MIRCat quantum cascade lasers (QCLs). The first covered 860−1479 cm−1 (6.9–11.6 µm), and the second spanned 1480–2222 cm−1 (4.5–6.6 µm), both with linewidths <1 cm−1. These two QCLs have a maximum repetition rate of 2 MHz and a maximum duty cycle of 20%. The third source was a Stuttgart Instruments’ optical parametric oscillator (OPO)-based 40 MHz laser, tunable from 700–18,000 nm. The OPO output was modulated in burst mode using an acoustic-optical modulator, enabling customization of waveform, duty cycle, and repetition rate up to 3 MHz. Typical laser output power for TFM-IR measurements in current study was 5–20 mW. To maintain consistent beam alignment during wavelength tuning, broadband optical components were used throughout the optical path. Specifically, the laser beam was collimated using a pair of 50 mm focal-length, 3−12 µm antireflection-coated lenses (Edmund Optics #87-947), and directed to the AFM tip via a series of broadband silver mirrors (Newport 10D20ER.2). A broadband gold-coated fast steering mirror (Optics In Motion OIM5001) was used for fine alignment, followed by focusing with a 15 mm focal-length gold-coated OAP (Thorlabs MPD00M9-M01), ensuring stable and consistent alignment across all wavelengths. AFM data (.spm files) were processed in Python using the pySPM module. Moiré images were corrected using line-alignment, scar filtering, and second-order polynomial background subtraction. Mica and hBN images were processed using plane-fitting, scar-filtering, and local Gaussian averaging. Crosstalk subtraction was also performed in Python using pySPM. The hBN line profiles in Fig. 1c were fitted using a sum of two exponentially decaying sinusoidal functions:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.87+0.2\cdot \sin \left[\frac{\left(x-996.1\right)2\pi }{361.9}+2.81\right]\cdot {e}^{\frac{-\left(x-996.1\right)}{1968}}+0.96 \cdot \sin \left[\frac{\left(x-996.1\right)2\pi }{188.3}+ 0.46\right]\cdot {e}^{\frac{-\left(x-996.1\right)}{473}}\,$$\end{document}1.87+0.2⋅sinx−996.12π361.9+2.81⋅e−x−996.11968+0.96⋅sinx−996.12π188.3+0.46⋅e−x−996.1473. The TFM moiré line profile in Fig. 4a was fitted using an even-power sinusoidal function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left(\sin 1.99x\right)}^{8}\,$$\end{document}sin1.99x8 to extract the cell-size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{\lambda }}}}$$\end{document}λ. This was used to determine the twist angle according to the equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi=\arccos \left[1-\frac{1}{2}{\left(\frac{a}{\lambda }\right)}^{2}\right]$$\end{document}ϕ=arccos1−12aλ2, where a=0.246a=0.246 nm is the lattice constant of graphene. FEM simulations Nanobubble simulations were performed with COMSOL Multiphysics 5.0 with the stationary Structural Mechanics module. COMSOL’s built-in parameters for mica, including Young’s modulus, Poisson’s ratio, and density were utilized in all calculations. The nanobubble geometry was defined by importing an experimentally measured height profile (Supplementary Fig. 8) as a preformation on a flat disk of mica, with a constant internal pressure applied within the nanobubble. An extra fine mesh was employed to ensure numerical accuracy for all the computations. Strain maps were obtained by calculating the resulting in-plane and out-of-plane displacement fields at the nanobubble surface. Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file