Millikelvin Microwave Impedance Microscopy

• Millikelvin MIM is a near-field scanning probe technique that measures local complex admittance, combining conductivity and dielectric contrast at mK temperatures.
• The method uses evanescent microwave fields confined to a nanoscale tip-sample junction, enabling imaging of subtle variations in quantum materials.
• Implementations face challenges in calibration, thermal noise, and TLS effects, which are managed through advanced impedance matching and cryogenic engineering.

Millikelvin microwave impedance microscopy (MIM) is a near-field, GHz-frequency scanning probe technique that measures the local complex impedance or admittance between a sharp metallic tip and a sample while operating in the milliKelvin regime inside a dilution refrigerator. In the simplest lumped-element description, the tip–sample junction is written as $$Y(\mathbf{r}) = G(\mathbf{r}) + i\omega C(\mathbf{r})$$, so the microscope probes local conductivity, permittivity, multilayer geometry, and tip–sample distance through position-dependent changes in the microwave reflection coefficient $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$ (Cao et al., 2023).

1. Measurement concept and observables

Millikelvin MIM is a near-field microscopy rather than a far-field imaging method. The tip is brought within tens of nanometers of the sample, the microwave fields are evanescent and confined to a sub-wavelength region around the tip apex, and the spatial resolution is therefore determined by the tip size and tip–sample separation rather than by the free-space microwave wavelength (Cao et al., 2023). In the quasi-electrostatic regime, the local response is commonly represented by the tip–sample admittance $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$, where $G_{ts}$ tracks dissipative response and $C_{ts}$ tracks capacitive or dielectric response (Wu et al., 2018).

The experimentally reported MIM channels are normally interpreted as quadratures of the perturbation of the reflected microwave field. In quantitative tuning-fork-based MIM, the in-phase component is labeled MIM-Re and is proportional to $\delta G_{ts}$, while the quadrature component is labeled MIM-Im and is proportional to $\omega\,\delta C_{ts}$ (Wu et al., 2018). In dilution-refrigerator implementations, the same language is retained: MIM-Re is used as a dissipative channel and MIM-Im as a reactive or capacitive channel, although the precise conversion from lock-in voltage to admittance depends on the matching network and calibration (Jiang et al., 2023).

A persistent misconception is that MIM is only a conductivity microscope. The reported implementations measure the local complex admittance, not only the local DC conductance, and are therefore sensitive to both conductive and dielectric response. The same near-field coupling can be used to image metallic, insulating, and buried structures, and in several systems the capacitive channel is the primary contrast mechanism (Cao et al., 2023).

2. Cryogenic implementations and instrument architectures

The dilution-refrigerator realization reported in 2023 integrates a tuning-fork-based atomic force microscope with microwave reflectometry and is capable of reaching temperatures down to 70 mK during imaging and magnetic fields up to 9 T (Cao et al., 2023). In that system, an etched tungsten tip with apex radius $\lesssim 200~\mathrm{nm}$ is glued to a quartz tuning fork, the microwave probe tone is delivered through an impedance-matching network resonant near 1.8 GHz, the tip power is approximately $-33~\mathrm{dBm}$, and the reflected signal is amplified by a cryogenic amplifier at the 3 K stage and demodulated in an AC mode referenced to the tuning-fork resonance near $32.768~\mathrm{kHz}$ (Cao et al., 2023).

A second dilution-refrigerator platform reported the implementation of a scanning MIM with a base temperature of $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$0, vibration noise as low as 1 nm, and a 30 $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$1 scan window at $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$2 (Jiang et al., 2023). That instrument used Attocube coarse positioners and a fine scanner, a sharply etched tungsten tip of $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$3 apex radius mounted on a quartz tuning fork, a $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$4 hand-formable cable segment with a series capacitor $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$5 and a parallel bias resistor $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$6 for impedance matching, operation around 2.8 GHz, and a 30 dB cryogenic low-noise amplifier close to the tip (Jiang et al., 2023). Scan speed was limited to $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$7 to minimize hysteresis and friction heating (Jiang et al., 2023).

Cryogenic MIM architecture predates dilution-fridge operation. A helium-cryostat instrument placed a shielded cantilever probe and Attocube scanning stage inside a variable-temperature insert covering 2–300 K and 9 T, used a $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$8 impedance-matching section, and exploited distance modulation at 400 Hz to monitor tip–sample distance and to align the microwave quadratures so that MIM-R vanished in the purely capacitive regime (Kundhikanjana et al., 2010). That implementation established the basic cryogenic workflow of stepwise approach, plane fitting, open-loop scanning on flat samples, and semi-quantitative finite-element interpretation (Kundhikanjana et al., 2010).

An alternative low-temperature branch replaces the AFM-style probe with an STM geometry. RF reflectometry was added to a millikelvin STM through a 300 MHz series LCR tank circuit using a hand-wound NbTi inductor $$\Gamma = (Z_{\mathrm{load}}-Z_0)/(Z_{\mathrm{load}}+Z_0)$$9, with the tip–sample capacitance used as a tuning parameter for impedance matching (Marbey et al., 16 Dec 2025). Using a niobium STM tip in a superconductor–insulator–normal metal tunnel junction, that system measured RF losses and electron temperature through the evolution of coherence peaks, and imaged island sizes down to a total surface area of $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$0 (Marbey et al., 16 Dec 2025). Although this geometry is not standard MIM, it is operationally very close to millikelvin microwave impedance imaging.

3. Calibration, sensitivity, and microwave engineering

Quantitative MIM depends on a joint electromagnetic and circuit calibration. Tuning-fork-based quantitative MIM established a two-step framework in which finite-element analysis computes $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$1 for a given tip geometry and sample stack, and transmission-line analysis plus measured $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$2 yields the electronics conversion factor $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$3 linking $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$4 to measured voltage (Wu et al., 2018). In that work, a tip with fitted radius $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$5 was calibrated on bulk dielectrics, the MIM-Im AC signal increased monotonically with $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$6, and the same workflow extracted local sheet conductance $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$7 in $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$8 devices (Wu et al., 2018).

A distinct instrumentation development replaced specialized microwave probes and cancellation circuits with monolithic silicon cantilever probes and a cancellation-free, self-referenced homodyne architecture (Shan et al., 2024). That system demonstrated thermal Johnson-noise-limited, drift-free MIM operation with 15 nm spatial resolution, minimal topography crosstalk, and a capacitance sensitivity of $$Y_{ts}(\omega) = G_{ts}(\omega) + i\omega C_{ts}(\omega)$$9 (Shan et al., 2024). The central relation used there was

$G_{ts}$0

with $G_{ts}$1, $G_{ts}$2, and $G_{ts}$3 for a purely capacitive perturbation (Shan et al., 2024). The reported architecture is not itself a millikelvin demonstration, but it directly targets cryogenic deployment because the Johnson-noise-limited regime shifts to lower input power as temperature is reduced (Shan et al., 2024).

Millikelvin microwave calibration of the line components is similarly decisive. A 30 mK short-open-load calibration of qubit drive-line components reported calibrated 1-port scattering parameters for attenuators and coaxial cables, with cryogenic return losses at 5 GHz of $G_{ts}$4 dB for a 20 dB attenuator, $G_{ts}$5 dB for a 10 dB attenuator, $G_{ts}$6 dB for a 230-mm silver-plated cupronickel coaxial cable, and $G_{ts}$7 dB for a 230-mm NbTi coaxial cable (Simbierowicz et al., 2021). The same work extracted cryogenic insertion losses of $G_{ts}$8 dB and $G_{ts}$9 dB for the two cable types at 5 GHz (Simbierowicz et al., 2021). For MMIM, these numbers set realistic bounds on mismatch, attenuation, and the degree of de-embedding required between room-temperature electronics and the mK probe.

Ultra-low-temperature thermometry and instability control remain nontrivial. A microwave optomechanics platform in a nuclear adiabatic demagnetization cryostat achieved thermometry from below $C_{ts}$0 to about 1 K using a blue-detuned parametric instability, but reported an unstable intrinsic driving force below typically 100 mK, with reliable experiments prevented below typically 10–30 mK (Zhou et al., 2019). Most of that unstable feature could be annihilated with a DC bias, and the mechanism was speculatively linked to atomic-scale two-level systems (Zhou et al., 2019). This is not a scanning MIM result, but it is directly relevant to the millikelvin microwave environment in which a microscope probe operates.

4. Materials, substrates, and loss mechanisms

The microwave environment of a millikelvin microscope is limited not only by the sample but also by the resonator substrate, interfaces, and superconducting film. A dedicated study of niobium $C_{ts}$1 coplanar waveguide resonators on substrates suitable for epitaxial growth of high-$C_{ts}$2 superconductors measured temperature and power dependent microwave loss down to about $C_{ts}$3 and around 5 GHz (Arzeo et al., 2014). The resonators had identical geometry on all substrates, with center conductor width $C_{ts}$4, gap $C_{ts}$5, and 200 nm Nb thickness, allowing direct comparison of dielectric and conductor losses (Arzeo et al., 2014).

Within the two-level-system model, the dielectric loss was written as

$C_{ts}$6

with

$C_{ts}$7

and a related dispersive frequency shift derived from Kramers–Kronig relations (Arzeo et al., 2014). In this framework, LSAT showed the best dielectric performance among the three studied substrates, with $C_{ts}$8 and $C_{ts}$9, whereas MgO had $\delta G_{ts}$0 and $\delta G_{ts}$1, and LaAlO$\delta G_{ts}$2 exhibited small $\delta G_{ts}$3 but a larger $\delta G_{ts}$4 dominated by conductor losses likely associated with film inhomogeneity from twin domains (Arzeo et al., 2014).

These results matter directly for MMIM because the tip, matching network, and on-chip resonant structures are often fabricated on exactly the kinds of oxide or dielectric platforms discussed in the resonator study. A microscope intended for few-photon or low-power operation at millikelvin temperatures inherits the same hierarchy of loss mechanisms: substrate TLS, metal–dielectric interfaces, conductor inhomogeneity, and residual background loss. The data suggest that substrate choice can determine whether the dominant millikelvin limitation is a clean TLS-dominated dielectric response, as in LSAT and MgO, or a conductor-loss-dominated response, as in LaAlO$\delta G_{ts}$5 (Arzeo et al., 2014).

5. Imaging quantum materials at millikelvin temperatures

Early millikelvin demonstrations established both benchmark contrast and genuinely low-temperature quantum transport imaging. In a dry dilution refrigerator, MIM at 70–90 mK imaged the conductivity contrast between graphite and silicon dioxide, with finite-element simulations showing that the imaginary part of the MIM signal decreases monotonically with increasing resistivity and saturates in both conductive and insulating limits (Cao et al., 2023). The same instrument visualized edge conduction in thin-film Dirac semimetal $\delta G_{ts}$6: at $\delta G_{ts}$7, deep in a quantum Hall plateau, the MIM image showed strongly enhanced signal at the sample edges decaying into the bulk, whereas at $\delta G_{ts}$8, between plateaus, the signal was uniform across the strip (Cao et al., 2023).

A dilution-refrigerator-based scanning MIM with a base temperature of $\delta G_{ts}$9 visualized quantum anomalous Hall states in Cr- and V-doped $\omega\,\delta C_{ts}$0 thin films on mica (Jiang et al., 2023). At $\omega\,\delta C_{ts}$1, the MIM-Im line profiles across the Hall bar exhibited two prominent edge peaks in the $\omega\,\delta C_{ts}$2 state and again in the $\omega\,\delta C_{ts}$3 state, while near the coercive fields the edge peaks vanished and the sample became more uniformly conductive (Jiang et al., 2023). Between the two $\omega\,\delta C_{ts}$4 peaks in transport, a dip in the center MIM-Im signal indicated the axion-insulator regime (Jiang et al., 2023). The observed edge-mode width of $\omega\,\delta C_{ts}$5 exceeded the expected chiral-channel width, with the paper attributing the broadening to finite tip size, residual bulk conductivity, and limited contrast (Jiang et al., 2023).

Millikelvin MIM has also been used to separate distinct edge mechanisms in graphene. At $\omega\,\delta C_{ts}$6 and $\omega\,\delta C_{ts}$7, local conductivity imaging of monolayer graphene showed that at the charge-neutrality point the edge conductivity dropped to zero more slowly than in the bulk as magnetic field increased, consistent with the calculated spatial profile of the charge gap in the canted antiferromagnetic phase (Yan et al., 20 Feb 2026). For $\omega\,\delta C_{ts}$8 integer quantum Hall states, the edge response was fundamentally different: dissipationless chiral edge channels produced MIM-Im peaks at the physical boundary, while MIM-Re peaks were shifted inward and associated with finite-$\omega\,\delta C_{ts}$9 transition regions (Yan et al., 20 Feb 2026). In that work, the apparent edge width was $\lesssim 200~\mathrm{nm}$0 with a standard tip and $\lesssim 200~\mathrm{nm}$1 with a sharper tip, while the modeled conductive strips themselves had widths on the order of the magnetic length (Yan et al., 20 Feb 2026).

These millikelvin results are continuous with earlier cryogenic MIM studies that imaged the metal–insulator transition in doped silicon, where MIM-R peaked at the spatial front where $\lesssim 200~\mathrm{nm}$2 and MIM-C saturated in metallic regions (Kundhikanjana et al., 2010). The progression from cryogenic semiconductor imaging to milliKelvin imaging of quantum Hall, quantum anomalous Hall, and correlated edge states is therefore methodological rather than discontinuous.

6. Theory, interpretation, and open issues

A common simplification is to interpret MIM entirely through local lumped-element circuits. For ordinary metals and insulators that approximation is often adequate, but a general linear-response formulation writes the tip–sample admittance as

$\lesssim 200~\mathrm{nm}$3

with

$\lesssim 200~\mathrm{nm}$4

so that the microscope probes a Coulomb-kernel-weighted convolution of the density response function $\lesssim 200~\mathrm{nm}$5 rather than only a local scalar conductivity (Wang et al., 2023). For Chern insulators, this theory predicts an enhanced edge response due to collective edge magnetoplasmon excitations, with resonance frequencies that depend quantitatively on the Chern number and on the sample circumference (Wang et al., 2023). It also predicts that MIM can be used to extract the topological invariant, distinguish topological from trivial edge states, and analyze dissipation along the boundary (Wang et al., 2023).

The theory is especially relevant in the millikelvin regime, where linear response is well controlled, dissipation is reduced, and edge states are sharply defined (Wang et al., 2023). This implies that not every bright edge feature should be read as a purely local conductivity map. In some systems, particularly Chern insulators and quantum Hall devices, the measured signal can be dominated by nonlocal collective modes rather than by a local lumped resistance or capacitance alone (Wang et al., 2023).

Open technical issues remain equally important. Dry dilution refrigerators introduce pulse-tube vibration near 1.4 Hz and harmonics, and the very high quality factor of tuning forks at millikelvin temperatures makes height feedback more sensitive but also more vulnerable to ringing and vibration-induced tip–sample distance modulation (Cao et al., 2023). TLS noise, dielectric loss, and stochastic drive events below roughly 100 mK remain a recurring concern across microwave platforms (Zhou et al., 2019). At the same time, the field is expanding toward topological insulators and Chern insulators, correlated 2D materials and moiré systems, superconductors and RF cavities, and quantum devices and mesoscopic physics (Cao et al., 2023). A cancellation-free, monolithic-silicon architecture with self-referenced homodyne detection suggests a route toward simpler broadband and nonlinear operation, and therefore toward more accessible millikelvin MIM platforms (Shan et al., 2024).

Taken together, the literature defines millikelvin microwave impedance microscopy as a convergence of near-field GHz electrodynamics, cryogenic scanning-probe engineering, quantitative microwave calibration, and low-loss materials design. Its central observable is the local complex admittance; its practical performance is set by vibration, matching, thermal load, and TLS physics; and its scientific reach now extends from calibration benchmarks and substrate loss studies to direct visualization of quantum Hall, quantum anomalous Hall, Dirac-semimetal, and graphene edge phenomena at millikelvin temperatures (Cao et al., 2023).