Solid-state thermometry via ionic–electronic coupling in two-dimensional heterostructures

Abstract

The coupling of ionic and electronic transport in solid-state systems offers new opportunities for realizing compact, energy-efficient sensing technologies, yet practical implementations remain limited. Here we introduce a two-dimensional iontronic platform based on field-effect transistors that integrate monolayer MoS₂ channels with van der Waals bimetallic thiophosphates (ABP₂X₆, A = Li, Cu, Ag and so on; B = In, Sc and so on; and X = S and Se) as ionic gate dielectrics to realize on-chip thermometry. Specifically, we exploit thermally activated ion migration within the gate dielectric leading to conductance modulation in the MoS₂ channel for temperature sensing. We achieve ~1–2 °C resolution, fast electronic readout and subpicojoule energy consumption in an ultracompact footprint (~1 µm²). Beyond thermometry, these results establish bimetallic thiophosphates as a versatile platform for solid-state iontronics and broaden the functional design space of van der Waals heterostructures for sensing, actuation and adaptive electronics.

Main

Precise and energy-efficient sensing is essential for continued scaling and reliable operation of modern electronic systems. Among parameters that demand real-time monitoring, temperature strongly impacts performance, power regulation and device lifetime^1,2,3,4,5. As integration densities increase and architectures shift towards three-dimensional (3D) stacking^{6,7,8,9,10,11} and neuromorphic computing^12,13,14, there is a growing need for thermal sensors that are compact, accurate and fast¹⁵. Whereas early integrated circuits used off-chip sensing, modern multicore processors require dozens of on-chip sensors^16,17, intensifying area and energy constraints. State-of-the-art targets include footprints below 0.005 mm² and active power consumption under 1–10 µW per sensor, with energy per readout ideally remaining below 1–10 nJ to support always-on and edge artificial intelligence applications^{18,19,20,21,22}.

Conventional on-chip temperature sensors, based on diodes^23,24,25, bipolar junction transistors^{26,27,28,29,30}, field-effect transistors (FETs)^31,32,33,34 or ring oscillators^35,36, primarily exploit temperature-dependent intrinsic electronic properties and phenomena such as carrier mobility, diffusion, threshold voltage, oscillation frequency and so on, associated with those of active semiconducting channel materials. Although widely used, they often require sizable footprints, complex calibration circuitry and higher energy, limiting suitability for next-generation dense systems. Notably, gate dielectrics are rarely treated as sensing elements: conventional oxides (SiO₂, Al₂O₃, HfO₂ and so on) offer strong electrostatic control but little intrinsic temperature sensitivity. By contrast, van der Waals (vdW) ionic dielectrics^37,38 contain mobile cations with thermally activated migration, enabling ion-mediated, temperature-dependent modulation of channel potential and reconfigurable device behaviour.

Building on this concept, we introduce a solid-state iontronic sensing modality in two-dimensional (2D) heterostructures. Specifically, we integrate a new class of layered ionic dielectrics, vdW bimetallic thiophosphates (ABP₂X₆, A = Li, Cu, Ag and so on; B = In, Sc and so on; and X = S and Se) with monolayer MoS₂ FETs. Under fixed bias, Li/Cu/Ag ion migration within the dielectric tunes the conductance of MoS₂ channel in a temperature-dependent manner, enabling a miniaturized thermometer with ~1 µm² footprint, subpicojoule energy expenditure, fast electronic readout (down to 100 ns) and ~1–2 °C resolution.

Results

2D vdW heterostructure for solid-state iontronic sensing

To realize solid-state iontronic sensing, we designed dual-gated FETs that integrate a semiconducting monolayer MoS₂ channel^39,40,41,42 with an exfoliated flake of ABP₂X₆ (ref. ⁴³) as the ionic top-gate dielectric, forming a vertical vdW heterostructure as shown in Fig. 1a. The devices are fabricated on a 25 nm Al₂O₃ back-gate dielectric grown by atomic layer deposition over a Ti/Pt (10/40 nm) stack and use Ni/Au as the source-drain contacts. The top gate is defined by e-beam lithography and Ni metallization. A scanning electron microscopy (SEM) image of a representative dual-gated FET with LiInP₂S₆ as the top-gate dielectric is shown in Fig. 1b, and similar SEM images of FETs with LiInP₂Se₆, CuInP₂S₆, AgInP₂S₆ and AgScP₂S₆ are shown in Supplementary Fig. 1. Complete fabrication details are provided in the Methods. This architecture facilitates electrostatic coupling between ion migration in ABP₂X₆ and charge transport in the MoS₂ channel, forming the basis for iontronic temperature sensing. The optical and spectroscopic characterization of the monolayer MoS₂ is presented in Supplementary Fig. 2a,b. Figure 1c presents an atomic force microscopy (AFM) image of a representative LiInP₂S₆ flake, showing smooth surfaces and uniform thickness of approximately 70 nm.

Fig. 1: 2D iontronic thermometer based on ABP₂X₆-gated MoS₂ field effect transistor (FET).

a,b, A schematic (a) and false-coloured angled SEM image (b) of a dual-gated 2D FET incorporating LiInP₂S₆ as the top-gate dielectric, monolayer MoS₂ as the semiconducting channel and 25 nm Al₂O₃ as the back-gate dielectric. c, An AFM image of an exfoliated LiInP₂S₆ flake with a measured thickness of ~70 nm. d, The top view of the ABP₂X₆ crystal structure, where A represents Li, Cu, Ag and so on; B corresponds to In, Sc and similar elements; P stands for phosphorus in the ABP₂X₆ family; and X denotes chalcogens such as S and Se. e, The XRD pattern of LiInP₂S₆ crystals confirms phase purity with no secondary phases. f, The SEM combined with EDS verifies the presence of LiInP₂S₆ constituents, except Li, which is undetectable by EDS. g, XPS spectra of exfoliated LiInP₂S₆ display a Li 1s peak at ~55.2 eV, confirming its ionic state, along with dominant In²⁺ signals and minor In³⁺ contributions due to surface oxidation.

Figure 1d displays the crystal structure for the ABP₂X₆ class of compounds and Fig. 1e shows the X-ray diffraction (XRD) pattern of as-grown LiInP₂S₆ crystal, revealing sharp, well-defined peaks predominantly indexed to the (00l) planes, consistent with the strong out-of-plane crystallinity and preferential layer stacking. The presence of the (002) plane and its higher-order reflections confirm the trigonal symmetry corresponding to the P-3̅1c space group. Similar XRD patterns, obtained for LiInP₂Se₆, CuInP₂S₆, AgInP₂S₆ and AgScP₂S₆, are shown in Supplementary Fig. 3. SEM combined with energy-dispersive X-ray spectroscopy (EDS) on a representative LiInP₂S₆ flake (Fig. 1f) confirms the presence of In, P and S, with ratios consistent with LiInP₂S₆; lithium is not detectable by EDS but is inferred from the elemental stoichiometry and known structure/synthesis. Analogous SEM-EDS for other thiophosphates is reported in ref. ⁴³. The X-ray photoelectron spectroscopy (XPS) presented in Fig. 1g further verifies composition and bonding showing a clear Li 1s peak at ~55.2 eV indicative of ionic Li (versus metallic Li at lower binding energy)⁴⁴, along with predominant In²⁺ (sulfide-rich) and a minor In³⁺ component (limited surface oxidation). In addition, P 2p (~132 eV) and S 2p (~162 eV) match the expected [P₂S₆]⁴⁻ anion framework. The XPS results for LiInP₂Se₆, CuInP₂S₆, AgInP₂S₆ and AgScP₂S₆ are provided in Supplementary Fig. 4. These results confirm the ionic character of thiophosphates needed for ion–electron coupling in our vdW heterostructure-based temperature-sensing devices. Raman spectroscopy analysis of ABP₂X₆ crystals are provided in Supplementary Fig. 5.

Sensing mechanism of the 2D iontronic thermometer

To investigate the operational principle of the 2D iontronic thermometer, we first established electrostatic control of the MoS₂ channel using various ABP₂X₆ flakes as the top gate dielectrics. Figure 2a shows the dual-sweep transfer characteristics of a representative device using LiInP₂S₆ as the top-gate dielectric, with a channel length, L_CH = 1 µm and channel width, W_CH = 1 µm. The source-to-drain current, ${I}_{\mathrm{DS}}$, was obtained by sweeping the top-gate voltage (${V}_{\mathrm{TG}}$) from –6 V to +6 V at a fixed back-gate voltage, V_BG = 3 V, and a source to drain voltage, V_DS = 1 V. Extended Data Fig. 1 shows similar transfer characteristics obtained for MoS₂ FETs with LiInP₂Se₆, CuInP₂S₆, AgInP₂S₆ and AgScP₂S₆ as the top-gate dielectrics. All devices exhibit n-type conduction with a large current on–off ratio, confirming that the ABP₂X₆ family of compounds can electrostatically modulate ${I}_{\mathrm{DS}}$ and function as an active dielectric gate. The choice of a positive ${V}_{\mathrm{BG}}$ ensures sufficient carrier density in the channel, allowing the top gate, defined in an underlapped device geometry, to control the carrier density locally. The global back-gate transfer characteristics of MoS₂ FETs, provided in Supplementary Fig. 6, further confirm this design strategy. Interestingly, all devices exhibit counterclockwise hysteresis under dual-sweep top-gate operation, consistent with the presence of mobile ionic species in the ABP₂X₆ dielectrics. Although ferroelectric polarization switching can also give rise to counterclockwise hysteresis, the sweep-direction-dependent subthreshold swing (SS) observed here strongly indicates ionic migration as the dominant mechanism. Specifically, the steeper SS during the forward sweep and the broader SS during the reverse sweep reflect the finite response time of mobile ions to the applied gate field. This delayed ionic motion leads to a time-dependent redistribution of the internal electrostatic potential, giving rise to the observed hysteretic and asymmetric subthreshold behaviour. This behaviour was reproduced in all measured devices for each material, as summarized in Extended Data Fig. 2. In addition, the schematic shown in Extended Data Fig. 3 illustrates the electric-field and temperature-assisted ion migration within the top-gate ionic dielectric of 2D FETs. At room temperature, a slower ion migration causes incomplete redistribution and larger hysteresis, whereas at elevated temperatures, faster ionic equilibration reduces the hysteresis magnitude, highlighting the thermally assisted ionic relaxation dynamics.

Fig. 2: Ionic–electronic coupling in 2D iontronic transistor based on LiInP₂S₆.

a, The dual-sweep top-gate transfer characteristic of a LiInP₂S₆-gated MoS₂ FET obtained by sweeping the ${V}_{\mathrm{TG}}$ from −6 V to 6 V at a constant V_BG = 3 V and V_DS = 1 V. b, Minor hysteresis loops, observed during partial ${V}_{\mathrm{TG}}$ sweeps, reveal incomplete ion redistribution within the LiInP₂S₆, resulting in nested loops. The hysteresis window (${V}_{\mathrm{Hyst}}$) scales with the sweep range—narrower sweeps induce limited ion migration, leading to smaller ${V}_{\mathrm{Hyst}}$ values and vice versa. c, The dual-sweep transfer characteristics were measured at varying sweep rates (${t}_{\mathrm{sweep}}$: 1 ms to 1 s), showing enhanced ${V}_{\mathrm{Hyst}}$ and current on–off ratio with longer sweep times. d, The temperature-dependent top-gate transfer characteristics, measured across different temperatures, T_A = 30–100 °C. The current on–off ratio increases steadily with T_A, aligning with the thermally activated migration of Li⁺ ions in the LIPS dielectric. Elevated temperatures enhance ionic mobility, allowing more complete Li⁺ redistribution during each sweep cycle and thereby strengthening the electrostatic control over the MoS₂ channel. This leads to progressively higher on–off ratios, reinforcing temperature-dependent the ionic–electronic coupling in LiInP₂S₆-gated MoS₂ FETs.

To further probe the ionic dynamics, we performed sweep-range (${V}_{\mathrm{sweep}}$)-dependent measurements, revealing the emergence of minor loops in the transfer characteristics as shown in Fig. 2b and Supplementary Fig. 7 for LiInP₂S₆-, LiInP₂Se₆-, CuInP₂S₆-, AgInP₂S₆- and AgScP₂S₆-gated MoS₂ FETs respectively. These minor loops reflect the incomplete redistribution of mobile ions when the gate voltage is cycled over a reduced range, and their shape is a direct consequence of the viscous nature of ion transport in the layered ABP₂X₆ structure. The ions partially respond to the applied field but do not fully equilibrate, leading to history-dependent modulation of the channel conductance. Such behaviour is not expected in conventional dielectrics or purely ferroelectric systems, providing additional evidence of ionic motion as the dominant mechanism. To explore the kinetics of ion migration, we next varied the sweep rate (${t}_{\mathrm{sweep}}$) for a fixed ${V}_{\mathrm{TG}}$ range. As shown in Fig. 2c and Supplementary Fig. 8, slower sweeps allow more time for ions to respond to the applied field, resulting in larger current on–off ratio. This further supports a diffusion-limited ionic response, where the degree of electrostatic modulation depends on the competition between the applied field rate and the ionic mobility. Finally, we studied the temperature dependence of electronic transport in ABP₂X₆-gated MoS₂ FET by varying the ambient temperature (${T}_{{\rm{A}}}$) under fixed sweep conditions. As shown in Fig. 2d and Supplementary Fig. 9, the current on–off ratio increases monotonically with temperature, consistent with the thermally activated migration of Li⁺, Cu⁺ or Ag⁺ ions within the layered ABP₂X₆ dielectrics. At higher temperatures, enhanced ionic mobility enables more complete redistribution of ions during each sweep cycle, leading to stronger electrostatic modulation of the MoS₂ channel. This results in progressively larger on–off current ratios with increasing temperature, further confirming the role of temperature-sensitive ionic dynamics in governing the device behaviour and establishing the foundation for its application as a solid-state iontronic thermometer.

To provide comprehensive insights on the physical mechanism of ion migration, we performed sweep-time- and temperature-dependent transfer measurements, time-resolved ${V}_{\mathrm{th}}$ evolution and molecular dynamics simulations (Supplementary Figs. 10–18). The emergence of a characteristic hysteresis peak time that shifts to shorter values at higher temperatures, together with Arrhenius-extracted activation energies and drift–diffusion-derived ion mobilities, unambiguously confirms thermally activated Li⁺/Cu⁺/Ag⁺ transport. Molecular dynamics simulations further corroborate the experimental migration barriers, establishing a quantitative, multiscale picture of ionic dynamics in ABP₂X₆ dielectrics. Note that such strong temperature-dependent ionic–electronic coupling is rarely observed in conventional dielectric-gated FETs, where electrostatic modulation is typically limited to purely electronic effects. This pronounced thermal sensitivity makes the vdW thiophosphates a promising platform for realizing miniaturized, FET-based solid-state thermometers.

Time-domain sensing and energy-speed trade-off

To realize a practical thermometer, we use a time-domain sensing scheme that translates temperature-dependent ionic dynamics into a measurable voltage response—a mechanism that inherently enables temperature-to-time conversion. The device is operated in current-mode, where a constant ${I}_{\mathrm{DS}}$ is injected at the drain terminal, while maintaining constant V_BG = 3 V and V_TG = –4 V. Under these fixed gate biases, the drain voltage (${V}_{{\rm{D}}}$) is monitored over time. Initially, the device is in a high-conductance state, but as ions gradually redistribute under the applied field, the channel threshold voltage of the MoS₂ FET becomes more positive leading to decrease in channel conductance and accordingly ${V}_{{\rm{D}}}$ rises. This temporal evolution continues until ${V}_{{\rm{D}}}$ reaches the compliance voltage (1 V) set by the measurement instrument, marking the transition to a low-conductance regime.

Figure 3a illustrates the temporal evolution of ${V}_{{\rm{D}}}$ for various temperatures (${T}_{{\rm{A}}}$) ranging from 80 °C to 110 °C at fixed I_DS = 100 pA, 500 pA, 1 nA, 5 nA for LiInP₂S₆-gated MoS₂ FET. This temperature range is particularly critical for integrated circuits, as most processors begin to reduce their speed at so-called throttle temperatures, typically around 85–95 °C, to avoid thermal stress¹⁸ and risk catastrophic failure above 105 °C (ref. ⁴⁵). Accurate on-chip thermometry, particularly near these critical thresholds, is essential for preventing device degradation and ensuring long-term stability. Extended Data Fig. 4 presents similar results for LiInP₂Se₆-, CuInP₂S₆-, AgInP₂S₆- and AgScP₂S₆-gated MoS₂ FETs. Figure 3b presents bar plots of ${\tau }_{{\rm{D}}}$, that is, the time required to reach the compliance voltage, as a function of ${T}_{{\rm{A}}}$ for different ${I}_{\mathrm{DS}}$ values for the LiInP₂S₆-gated MoS₂ FET. As expected, at any given ${I}_{\mathrm{DS}}$, ${\tau }_{{\rm{D}}}$ decreases with increasing ${T}_{{A}}$ reflecting enhanced Li⁺ ion mobility at higher temperatures. Moreover, for any fixed ${T}_{{\rm{A}}}$, a higher ${I}_{\mathrm{DS}}$ enables faster sensing, that is, lower ${\tau }_{{\rm{D}}}$. This is because a smaller change in channel conductance is sufficient to drive ${V}_{{\rm{D}}}$ to the compliance threshold. Extended Data Fig. 5a presents similar bar plots for ${\tau }_{{\rm{D}}}$ corresponding to LiInP₂Se₆-, CuInP₂S₆-, AgInP₂S₆- and AgScP₂S₆-gated MoS₂ FETs, measured at a given ${I}_{\mathrm{DS}}$. A similar trend, that is, ${\tau }_{{\rm{D}}}$ decreasing with increasing ${T}_{{\rm{A}}}$ is observed for all materials. Figure 3c and Extended Data Fig. 5b summarize the corresponding energy expenditure per sensing event (${E}_{{\rm{T}}}$), calculated as the product of ${I}_{\mathrm{DS}}$ and the area under the ${V}_{{\rm{D}}}(t)$ curve during the sensing interval for the LiInP₂S₆-, LiInP₂Se₆-, CuInP₂S₆-, AgInP₂S₆- and AgScP₂S₆-gated MoS₂ FETs, respectively. Other dissipating components such as the top- and back-gate leakage currents (Supplementary Fig. 19) are at least two orders of magnitude smaller than ${I}_{\mathrm{DS}}$ and do not add meaningfully to overall ${E}_{{\rm{T}}}$.

Fig. 3: Time-domain sensing in LiInP₂S₆-based 2D iontronic thermometer.

a, The temporal evolution of drain voltage (${V}_{{\rm{D}}}$), monitored for LiInP₂S₆-gated MoS₂ FETs, across the temperature (${T}_{{\rm{A}}}$) range of 80–110 °C, under fixed ${I}_{\mathrm{DS}}$ values of 100 pA, 500 pA, 1 nA and 5 nA. b, A bar plot of ${\tau }_{{\rm{D}}}$, that is, the time required to reach the compliance voltage, extracted from a as a function of ${T}_{{\rm{A}}}$ for different ${I}_{\mathrm{DS}}$ values. c, The energy expenditure per sensing event (${E}_{{\rm{T}}}$) is obtained by multiplying ${I}_{\mathrm{DS}}$ with the area under the ${V}_{{\rm{D}}}$(t) curve over the sensing interval for LiInP₂S₆-gated MoS₂ FETs.

Nevertheless, as expected, there exists a fundamental trade-off between sensing speed and energy consumption: whereas a larger ${I}_{\mathrm{DS}}$ enables faster temperature readout by accelerating the ${V}_{{\rm{D}}}$ rise, it also increases ${E}_{{\rm{T}}}$. Conversely, operating at lower ${I}_{\mathrm{DS}}$ extends ${\tau }_{{\rm{D}}}$ but reduces the energy cost per measurement. Notably, the 2D iontronic thermometer based on LiInP₂S₆-gated MoS₂ FET achieves sensing speeds as fast as τ_D = 6 ms at T_A = 110 °C with I_DS = 5 nA and maintains a relatively fast response of τ_D ~264 ms at T_A = 80 °C. The corresponding ${E}_{{\rm{T}}}$ values were found to be ~19 pJ and ~470 pJ, respectively. Further reducing the operating current to ${I}_{\mathrm{DS}}$ = 100 pA increases the response time at 110 °C to τ_D ≈ 40 ms (∼7× slower) but substantially lowers ${E}_{{\rm{T}}}$ by ~30× to ~640 fJ, highlighting the tunability of the device performance through bias optimization. This tunability allows the sensing protocol to be adapted for performance- or power-constrained environments. In addition, note that the active footprint of the 2D thermometer is ~1 µm². Moreover, we benchmarked our 2D iontronic temperature sensors against commercial on-chip sensors based on bipolar junction transistors (AD590), diodes (BPW21R) and resistors (LM35) and found that our devices operate at ultra-low current and power levels (Supplementary Fig. 20). Furthermore, Supplementary Table 1 compares the performance of the 2D iontronic thermometer against state-of-the-art thermal sensors. We also note that, beyond temperature-to-time encoding, the same iontronic thermometer can be read out more simply and much faster using the instantaneous device current as the temperature signal with energy per measurement in the subpicojoule regime (Extended Data Fig. 6).

Accurate temperature readout requires periodic, short sensing windows (shorter than thermal transients) and a reset between measurements to restore the dielectric’s initial ionic configuration; otherwise, ${\tau }_{{\rm{D}}}$ records a convoluted thermal history rather than a single temperature point. This is achieved by applying a positive V_TG > 5 V for 100 ms, which drives Li⁺ ions back towards the gate electrode. Supplementary Fig. 21 shows the response of a representative device measured at a fixed T_A = 85 °C following five sequential resets. The measured ${\tau }_{{\rm{D}}}$ remains consistent across all cycles, confirming the reliability and repeatability of the sensing process. Moreover, we measured the dual-sweep top-gate transfer characteristics of a representative LiInP₂S₆-gated MoS₂ FET at T = 30 °C and 80 °C, over multiple thermal cycling measurements as shown in Extended Data Fig. 7. The transfer curves were recorded by sweeping ${V}_{\mathrm{TG}}$ at a constant V_BG = 4 V and V_DS = 1 V. The results show minimal variation across successive cycles, with the device consistently retracing the same transfer characteristics under thermal stress. Similarly, we assessed the temporal evolution of drain voltage (${V}_{{\rm{D}}}$), monitored for LiInP₂S₆-gated MoS₂ FETs, across at T = 30 °C and 80 °C and over multiple cycles as depicted in Extended Data Fig. 7. The recorded ${V}_{{\rm{D}}}$ traces exhibit negligible drift or degradation across repeated heating and cooling cycles, indicating that the sensing characteristics remain highly stable and reproducible. This consistent behaviour confirms that Li⁺ migration within the dielectric is fully reversible under thermal excitation and relaxation and that no cumulative ionic displacement or defect generation occurs during repeated operation. Long-term stability was verified by measuring two iontronic FET sensors based on LiInP₂S₆ and CuInP₂S₆ across T_A = 80–100 °C for three consecutive days, where ${\tau }_{{\rm{D}}}$ decreased monotonically with temperature yet remained consistent day-to-day with negligible drift (Extended Data Fig. 8). We also evaluated the performance of five distinct LiInP₂S₆-gated MoS₂ FETs at a fixed temperature, T_A = 85 °C as shown in Supplementary Fig. 22. As expected, ${\tau }_{{\rm{D}}}$ values exhibited some device-to-device variation, primarily owing to flake thickness, which could be reduced via calibration or, more scalable, by large-area growth of uniform vdW thiophosphate films.

Range and precision

Operational range and temperature resolution are key for thermal sensors in advanced systems. In our iontronic FET thermometer, both are set by ion-migration dynamics in the vdW dielectric; therefore, bias conditions must be chosen to balance range and precision. Specifically, ${I}_{\mathrm{DS}}$ should be high enough that ${V}_{{\rm{D}}}$ reaches compliance within a practical window at low temperature where ionic mobility is reduced but not so high that compliance is reached too quickly, reducing resolution and wasting energy. Figure 4a presents the extracted ${\tau }_{{\rm{D}}}$, measured for a constant I_DS = 5 nA across a wide temperature range (0–110 °C) for a LiInP₂S₆-gated MoS₂ FET. The corresponding temporal ${V}_{{\rm{D}}}$ profiles are provided in Extended Data Fig. 9. As expected, ${\tau }_{{\rm{D}}}$ increases substantially at lower temperatures from 6 ms at 110 °C to 10 s at 0 °C. To mitigate this slowdown at lower temperatures, the response speed can be enhanced by increasing the magnitude of ${V}_{\mathrm{TG}}$ and/or ${I}_{\mathrm{DS}}$. Applying a more negative ${V}_{\mathrm{TG}}$ increases the electric field across the dielectric, which accelerates ion migration and reduces the time required for the device to transition from a high- to low-conductance state. This effect is demonstrated in Fig. 4b, where applying a stronger V_TG = −6 V at I_DS = 100 pA and T_A = 80 °C yields an approximately fivefold faster response reducing ${\tau }_{{\rm{D}}}$ from 422 ms at V_TG = −4 V to 82 ms. Similarly, Fig. 4c shows that increasing ${I}_{\mathrm{DS}}$ from 5 nA to 20 nA at T_A = 30 °C yields an approximately fivefold reduction in response time, with ${\tau }_{{\rm{D}}}$ decreasing from 1,800 ms to 400 ms. In short, the use of 2D iontronic thermometer may require the precalibration of biasing conditions to match the target temperature range and desired sensing speed.

Fig. 4: Range and precision in 2D iontronic thermometer.

a, The extracted ${\tau }_{{\rm{D}}}$ as a function of ${T}_{{\rm{A}}}$ for a constant ${I}_{\mathrm{DS}}$ of 5 nA and a ${V}_{\mathrm{TG}}$ of −4 V, for a representative LiInP₂S₆-gated MoS₂ FET over the entire temperature range of 0–100 °C. b, The extracted ${\tau }_{{\rm{D}}}$ as a function of ${V}_{\mathrm{TG}}$, at a ${T}_{{\rm{A}}}$ of 80 °C, illustrating that a higher negative ${V}_{\mathrm{TG}}$ enhances the electric field across LiInP₂S₆, accelerating Li⁺ migration and shortening the transition time to the low-conductance state. c, The ${\tau }_{{\rm{D}}}$ as a function of ${I}_{\mathrm{DS}}$ at ${T}_{{\rm{A}}}$ of 30 °C and ${V}_{\mathrm{TG}}$ of −4 V, exhibiting a monotonic reduction in ${\tau }_{{\rm{D}}}$ with increasing ${I}_{\mathrm{DS}}$. d, A violin plot of ${\tau }_{{\rm{D}}}$ as a function of ${T}_{{\rm{A}}}$, ranging from 85 °C to 90 °C, for five sensing cycles. Each degree increment yields a clearly resolvable change in ${\tau }_{{\rm{D}}}$, with no overlap between adjacent ${T}_{{\rm{A}}}$ points. The standard deviation of ${\tau }_{{\rm{D}}}$ is between 1 ms and 5 ms.

Finally, the precision of the 2D iontronic thermometer is evaluated by examining the reproducibility and sensitivity of the response time within narrow temperature windows. As shown in Fig. 4d, the device demonstrates the ability to distinguish temperature differences as small as 1 °C within the critical range of 85–90 °C, using I_DS = 100 pA. Each degree increment yields a clearly resolvable change in ${\tau }_{{\rm{D}}}$, with no overlap between adjacent temperature points. To validate the robustness of this precision, measurements were repeated across five independent sensing cycles, confirming high reproducibility and minimal statistical variation in the extracted ${\tau }_{{\rm{D}}}$ values as presented in Extended Data Fig. 10. Achieving higher precision, however, requires slowing down the measurement to ensure that small temperature differences produce sufficiently distinct ${\tau }_{{\rm{D}}}$ shifts. This necessitates operating at a lower ${I}_{\mathrm{DS}}$, which stretches the overall ${\tau }_{{\rm{D}}}$ window and enhances the resolution between adjacent temperatures. Although reducing ${I}_{\mathrm{DS}}$ increases the sensing time, it simultaneously lowers the energy expenditure per measurement, offering an advantageous trade-off for precision-critical applications where measurement speed is less important than sensitivity.

To further evaluate the sensing dynamics of the 2D iontronic platform, we have expanded our analysis to assess the temperature-dependent top-gate characteristics of 2D iontronic devices. As shown in Supplementary Fig. 23, each device exhibits a distinct and reproducible modulation of the drain current (${I}_{{\rm{D}}}$) with temperature when measured at a fixed top-gate voltage (${V}_{\mathrm{TG}}$). Notably, in the subthreshold regime, the ${I}_{{\rm{D}}}$ is highly sensitive to even small temperature variations, allowing precise detection of thermal changes through changes in ${I}_{{\rm{D}}}$. In the critical temperature range of 85–90 °C, the traces are separable down to 1 °C, with robust discrimination at 2–3 °C, with LiInP₂Se₆ and AgInP₂S₆ showing the strongest contrast. We believe that the resolution can be further improved by sharpening the SS and optimizing the top-gate dielectric thickness. In addition, note that the measurement can be performed with high temporal resolution, and the sensing response can be detected within ~100 ns, corresponding to the maximum temporal resolution of our instrument, suggesting that the intrinsic thermal response of the 2D iontronic devices may be even faster.

Conclusion

This work introduces a new class of 2D iontronic sensors that harness thermally activated ionic motion in layered bimetallic thiophosphates to modulate charge transport in monolayer semiconductors. By integrating LiInP₂S₆, LiInP₂Se₆, CuInP₂S₆, AgInP₂S₆ and AgScP₂S₆ as ionically active dielectrics with MoS₂ FETs, we demonstrate absolute temperature sensing with high resolution (down to 1 °C), fast response (as low as 100 ns), low energy consumption (as low as ~2 fJ) and an ultracompact footprint (~1 µm²). These metrics surpass or rival those existing complementary metal–oxide-semiconductor-compatible thermal sensors. Beyond thermometry, this platform establishes bimetallic thiophosphates as a versatile new family of vdW ionic dielectrics and provides a general framework for solid-state iontronic devices. The coupling of mobile ions with 2D semiconductors opens a rich design space for multifunctional electronics.

Methods

Large-area monolayer MoS₂ film growth

Monolayer MoS₂ films were grown on a 2-inch prescored, double-side polished c-plane sapphire substrate using metalorganic chemical vapour deposition. A cold-wall horizontal reactor with an inductively heated graphite susceptor and wafer rotation was used to ensure uniform deposition, as detailed in previous work⁴⁶. Molybdenum hexacarbonyl (Mo(CO)₆) and hydrogen sulfide (H₂S) served as the precursors. The metal precursor, Mo(CO)₆, was maintained at 10 °C and 950 torr in a stainless-steel bubbler, delivering a flow of 0.036 sccm, while 400 sccm of H₂S was supplied. The deposition took place at 1000 °C and 50 torr under a hydrogen atmosphere, achieving monolayer growth in 18 min. Before growth, the substrate was preheated to 1,000 °C in H₂ for 10 min. Following deposition, the substrate was cooled in H₂S to 300 °C to prevent decomposition of the MoS₂ film. Additional details can be found in previous studies^39,47,48.

Growth of LiInP₂S₆ and LiInP₂Se₆ crystals

Lithium (99.9%), indium ingot (99.999%), red phosphorus (99.999%), sulfur (99.999%) and selenium shot (99.999%) were used as received without further purification. Lithium chalcogenides (Li₂S and Li₂Se) were synthesized following a modified literature procedure, and owing to their air-sensitive nature, all handling was carried out in a nitrogen-filled glovebox. Phosphorus pentachalcogenides (P₂S₅ and P₂Se₅) were prepared by loading phosphorus with sulfur or selenium, respectively, in stoichiometric ratios into fused-silica tubes, which were flame-sealed under vacuum (~3 × 10⁻³mbar), heated to 500 °C over 12 h, held for 72 h and furnace-cooled to room temperature. For bulk synthesis, Li₂S (or Li₂Se), In, P₂S₅ (or P₂Se₅) and excess S (or Se) were combined in a carbon-coated fused-silica tube using a slightly Li- and chalcogen-rich molar ratio (Li:In:P:X ≈ 1.05:1:2.04:6.12, X = S or Se). Single crystals of LiInP₂S₆ and LiInP₂Se₆ were grown using the chemical vapour transport (CVT) method in a two-zone furnace. Presynthesized^49,50 polycrystalline material (~2.5 g) was mixed with iodine (50 mg) as the transport agent, loaded into a fused-silica tube (16–18 mm inner–outer diameter, 28 cm length), evacuated and flame-sealed. The source zone was ramped to 560 °C over 12 h, held for 4 h, further heated to 660 °C over 3 h and maintained for 144 h, followed by cooling to room temperature over 12 h, whereas the deposition zone was maintained at 560 °C to promote crystal growth. The resulting LiInP₂S₆ and LiInP₂Se₆ crystals were collected inside a nitrogen-filled glovebox to avoid moisture exposure and used directly without further processing. The crystals exhibit large-area, flat surfaces suitable for device fabrication.

Growth of AgInP₂S₆, CuInP₂S₆ and AgScP₂S₆ crystals

Single crystals of AgInP₂S₆, CuInP₂S₆ and AgScP₂S₆ were grown using the CVT method in evacuated quartz ampoules. High-purity elemental Ag (or Cu), In (or Sc), P and S were mixed in stoichiometric ratios corresponding to ~12 g of the target compound and loaded into quartz ampoules (40 mm × 220 mm) together with ~0.5 g of iodine as the transport agent. The ampoules were evacuated to pressures below 1 × 10⁻³Pa using an oil diffusion pump with a liquid-nitrogen cold trap and flame-sealed under vacuum. The sealed ampoules were prereacted horizontally in a muffle furnace by heating to 450 °C at 1 °C min⁻¹ for 25 h, followed by annealing at 500 °C for 50 h and 600 °C for 50 h with a reduced heating rate of 0.1 °C min⁻¹ and a cooling rate of 1 °C min⁻¹. The prereacted material was then mechanically homogenized inside the ampoule and transferred to a horizontal two-zone furnace for CVT growth, where the source and growth zones were initially held at 600 °C and 750 °C, respectively, for 2 days. The temperature gradient was subsequently reversed, with the source zone at 700 °C and the growth zone at 600 °C for 10 days to promote crystal growth. After cooling to room temperature, the growth zone was heated to 200 °C for 2 h while keeping the source zone near room temperature to remove residual transport agent. The ampoules were finally opened in an argon-filled glovebox, and well-formed AgInP₂S₆, CuInP₂S₆ and AgScP₂S₆ single crystals were collected.

Raman spectroscopy

Raman spectroscopy of the MoS₂ films were performed using a Horiba LabRAM HR Evolution confocal Raman microscope with a 532-nm laser. The laser power was initially set to 34 mW and reduced to 1.7 mW using a 1% filter. A 100× objective with a 0.9 numerical aperture was used, with a grating spacing of 1,800 grooves mm⁻¹ for Raman spectroscopy and 300 grooves mm⁻¹ for PL spectroscopy. For the 2D dielectrics, the Raman spectra was collected using confocal SPID Nano Raman and TERS System, equipped with an excitation wavelength of 532 nm.

XRD

Powder XRD patterns of synthesized crystals were obtained using a Rigaku Miniflex600 XRD system equipped with a Dtex silicon 1-D detector. The Cu Kα radiation (wavelength λ = 1.5406 Å) was generated by applying a voltage of 40 kV and a current of 15 mA. The radiation was filtered with a graphite monochromator and a Kβ foil filter. A zero-background silicon holder was used for the measurements.

XPS

XPS measurements were performed using a Thermo Scientific NEXSA G2 XPS equipped with an electron flood gun and a scanning ion gun at room temperature. The crystals were placed on copper tape, which served as the adhesive substrate. The data were analysed using Thermo Scientific Avantage Data System software.

AFM

AFM was utilized to investigate the thickness profile of exfoliated multilayer flakes before and after plasma treatments. RTESPA-150 probe tips were used with a Bruker dimension icon AFM. All images were collected in peak force tapping mode with a peak force of 12 nN and a scan rate of 0.5 Hz. Images were processed and exported using Nanoscope.

SEM

The SEM of the 2D MoS₂ transistors used in this study was conducted using a Zeiss Gemini 500 field emission SEM system at an accelerating voltage of 5 kV.

EDS

The EDS analysis was performed on micrometre-sized bulk crystal flakes using an ESEM Q250 SEM instrument. A tungsten source with an accelerating voltage of 30 kV was utilized to determine the elemental composition of the crystals.

Fabrication of dual-gated 2D FETs

To begin with, monolayer MoS₂ was transferred on 25-nm-thick Al₂O₃ substrates. The substrate was spin-coated with PMMA A6 (4,000 rpm for 45 s) and baked at 180 °C for 90 s. The resist was then exposed using a Raith EBPG5200 e-beam lithography tool and developed with a 1:1 mixture of 4-methyl-2-pentanone (MIBK) and isopropyl alcohol (IPA) for 60 s, followed by a 45 s IPA rinse. The exposed monolayer MoS₂ film was etched using an SF₆ reactive ion etching process at 5 °C for 10 s. Afterwards, the sample was cleaned in acetone and IPA to remove the e-beam resist. Next, the 2D dielectrics were mechanically exfoliated on top of the prepatterned MoS₂ channels. A second lithography step was performed to form the source-drain and top-gate electrodes. The substrate was spin-coated at 4,000 rpm for 45 s with methyl methacrylate (MMA) EL6 and PMMA A3; these resists were baked at 150 °C and 180 °C for 90 s each. An e-beam lithography was used again to pattern the source and drain, followed by development with MIBK/IPA and an IPA rinse, as before. For MoS₂ FETs, 20 nm of Ni and 60 nm of Au were deposited via e-beam evaporation to form the electrodes. Finally, a lift-off process was conducted to remove excess Ni/Au by soaking the sample in acetone for 1 h, followed by a 10-min IPA rinse to clean the substrate.

Reactive force field for molecular dynamics simulations

A reactive force field (ReaxFF) was developed for the Li–In–P–S chemical system, targeting both LiInPS₄ and InPS₄. The training database includes atomization energies and elastic properties of β-Li₃PS₄ (Pnma) and orthorhombic InPS₄. Reference structural parameters and elastic constants were obtained from density functional theory (DFT) calculations available in the Materials Project database. The elastic dataset comprises the bulk modulus ($B$) and nine independent elastic constants (${C}_{11}$, ${C}_{22}$, ${C}_{33}$, ${C}_{12}$, ${C}_{13}$, ${C}_{23}$, ${C}_{44}$, ${C}_{55}$ and ${C}_{66}$). To generate training data for elastic properties, small lattice distortions were applied to β-Li₃PS₄, and the resulting changes in total energy ($\Delta E$) were computed as a function of strain using the Taylor-expanded relationship between elastic constants and strain-induced energy variations. Very small strain amplitudes were used to minimize errors arising from higher-order terms in the expansion. The bulk modulus was incorporated into the training set by fitting energy–volume data to the Birch–Murnaghan equation of state. The elastic constants ${C}_{{i}{j}}$ were extracted from second-order polynomial fits to the energy–strain relationships following the methodology described in ref. ⁵¹. Overall, the developed ReaxFF force field accurately reproduces the elastic response of the material under small deformations and captures the strain-dependent energetics in close agreement with DFT calculations.

Electrical characterization

The electrical characterization of the fabricated devices was performed using a semi-automated Formfactor 11000 probe station under atmospheric conditions with a Keysight B1500A parameter analyser.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.