Potassium ferricyanide [K₃Fe(CN)₆, ≥ 99.5%], potassium hexacyanoferrate [K₄Fe(CN)₆·3H₂O, > 99.0%], DMSO (≥ 99.9%), and AMPS (98%) were purchased from Shanghai Macklin Biochemical Technology Co., Ltd. (Shanghai, China). Acrylamide (AAm, ≥ 99.9%), ammonium persulphate (APS, ≥ 96%), and N,N'-methylenebisacrylamide (MBAA, ≥ 99%) were obtained from Shanghai Aladdin Biochemical Technology Co., Ltd. (Shanghai, China). Graphite plates were sourced from Qingdao Baofeng Graphite Produce Co., Ltd. All chemicals were used as received. Deionized water was employed throughout.

In a typical synthesis, AAm and AMPS (total mass of 2.0 g) were dissolved in 5.0 mL of deionized (DI) water at predetermined mass ratios to yield the desired copolymer compositions. Subsequently, 0.02 g of APS (initiator) and 0.02 g of MBAA (crosslinker) were added. After stirring at room temperature for 1 h at 300 rpm, the precursor was transferred into a polytetrafluoroethylene mold and polymerized at 60 °C for 1 h to produce a hydrogel matrix. By varying the AAm/AMPS ratio, pAAm, p(AAm-AMPS), and pAMPS hydrogels were obtained.

The as-prepared hydrogel matrices were then immersed in a 0.4 M Fe(CN)₆^4-/3- aqueous solution at 60 °C for 3 h to yield the thermogalvanic hydrogel electrolytes. Samples prepared without co-solvents were denoted as PAAm and P(AAm-AMPS). To establish a concentration gradient of the redox ions, DMSO was introduced into the immersion solution. The resulting DMSO-containing electrolytes were labeled as PAD-X and PSD-X for the PAAm and P(AAm-AMPS) systems, respectively, where X indicates the volume percentage (vol%) of DMSO.

Density functional theory (DFT) simulations were performed to evaluate electrostatic potentials (ESPs) and binding energy. For ESP analysis, geometry optimizations were carried out at the B3LYP/6-311G(d) level^[34,35], with an implicit solvent model using the Gaussian 16 package^[36]. Single-point energy calculations were then performed on the optimized structures to map ESP surfaces, followed by critical extrema analysis via the Multiwfn program. For binding energy calculations, simulations were conducted using the DMol3 module in Materials Studio 2023. The generalized gradient approximation with the Perdew–Burke–Ernzerhof functional and double numerical plus polarization basis sets were employed. The convergence tolerances for energy, force, and displacement were set to 1 × 10^-5, 0.002 Ha·Å^-1, and 0.005 Å, respectively. Individual ions and interacting pairs were fully relaxed before single-point energy calculations to determine the binding energy.

Fourier transform infrared (FTIR) spectra were collected using a Bruker Vertex 70V + Hyperion II spectrometer (Bruker, Germany) under vacuum in the range of 400-4,000 cm^-1. X-ray photoelectron spectroscopy (XPS) was performed on a PHI VersaProbe 4 system (ULVAC-PHI, Japan). Ultraviolet-visible (UV-vis) absorption spectra were recorded from 200 to 500 nm with a PerkinElmer LAMBDA 1050+ spectrophotometer (PerkinElmer, USA). Thermal conductivity was measured using a Hot Disk TPS 2500S analyzer (Hot Disk AB, Sweden). X-ray diffraction (XRD) patterns were collected on a Malvern PANalytical diffractometer (Malvern PANalytical, UK) with Cu Kα radiation (λ = 1.5406 Å) over the 2θ range 10°-60°. The morphology of the electrolyte was examined by scanning electron microscopy (SEM, Hitachi Regulus8100, Hitachi, Japan) at an acceleration voltage of 15 kV and an optical microscope (YM710TR, YUESCOPE, China). A universal testing machine (Instron 68TM, Instron, USA) equipped with a 50 N load cell was used to evaluate the mechanical properties at a deformation rate of 50 mm·min^-1. Viscosity measurements were conducted using a Brookfield DV3T LV rheometer (AMETEK Brookfield, USA).

The QTEC was assembled in a sandwich configuration of platinum (Pt)|hydrogel electrolyte (3 × 1 × 0.3 cm)|Pt. Two thermocouples were attached to both electrodes to monitor the temperature gradient (ΔT). The temperature of these plates was precisely regulated by a circulating water-cooling system (HRC2 Lite, IKA, Germany). The S_e was determined from the slope of the steady-state voltage vs. ΔT plots, with the thermoelectrochemical voltage and ΔT monitored by a Keithley DAQ6510 sourcemeter (Keithley Instruments, USA). Current–voltage and power–voltage characteristics were captured using a Keithley 2450 sourcemeter (Keithley Instruments, USA) by sweeping the voltage from the open-circuit value down to 0 V, with the load resistance determined from the recorded current at each voltage step. Discharge-charge cycles were performed using a protocol consisting of thermal charging (open circuit), galvanic discharge, and thermal recovery. Ionic conductivity (σ) was determined by electrochemical impedance spectroscopy (EIS, PARSTAT 3000A, AMETEK, USA) and calculated as σ = L/(RA), where L, R, and A respectively denote the hydrogel electrolyte thickness, bulk resistance, and electrode-electrolyte contact area.

Figure 1A schematically compares the network structures of the pristine PAAm and P(AAm-AMPS) hydrogel electrolytes. The PAAm backbone is non-electroactive and provides minimal electrostatic interaction with redox ions. In contrast, copolymerization with AMPS monomers introduces abundant sulfonate groups (−SO₃^-) that act as fixed anionic centers. ESP maps confirm that the AMPS monomer carries a strongly negative charge distribution localized on the sulfonate moiety, whereas the AAm unit remains essentially neutral [Figure 1B]. This concentrated negative charge creates a strong local electric field, which offers a driving force for valence-selective interactions with the redox ions.

FTIR spectra verify the integration of both polymer components [Figure 1C]. The peak at ~1,450 cm^-1 corresponds to the −CH₂ bending vibration of the PAAm segments^[37,38], while the bands at 1,042 and 1,195 cm^-1 are assigned to the stretching vibrations of the S–O and S=O groups in AMPS chains^[39,40]. The coexistence of these characteristic bands confirms the successful formation of P(AAm-AMPS) copolymer networks. XPS spectra further corroborate this result through the emergence of a distinct S 2p signal in the P(AAm-AMPS) electrolyte [Figure 1D]. Moreover, the high-resolution N 1s spectra [Figure 1E] display a clear change in line shape for the C–NH₂ and O–NH–C components^[41,42], consistent with the altered chemical environment expected for the copolymer. The P(AAm-AMPS) hydrogel also exhibits good mechanical robustness and stretchability, as confirmed by its resistance to twisting, folding, and stretching without structural failure [Figure 1F and Supplementary Figure 1]. This mechanical integrity ensures that the as-prepared QTECs can accommodate complex deformation during operation without electrolyte leakage.

The heat-to-electricity conversion in the QTEC relies on temperature-dependent redox reactions of the Fe(CN)₆^4-/3- couple [Figure 2A]. At the hot electrode, Fe(CN)₆^4- oxidizes to Fe(CN)₆^3- and releases an electron; this electron travels through the external circuit to the cold electrode, where it reduces Fe(CN)₆^3- back to Fe(CN)₆^4-. To evaluate the effect of network composition, Figure 2B and Supplementary Figure 2 examine the S_e of the P(AAm-AMPS) electrolytes as a function of the AAm/AMPS mass ratio. The pristine PAAm electrolyte (1:0 ratio) exhibits an S_e of 1.31 mV·K^-1, consistent with previous reports^[26,30]. Incorporation of AMPS markedly influences S_e, an effect attributable to emerging ion-polymer interactions. S_e reaches a maximum of 1.61 mV·K^-1 at a 1:1 mass ratio, approximately a 23% improvement over PAAm. Further increasing AMPS content to a 1:3 ratio, however, reduces S_e to 1.42 mV·K^-1. Excess AMPS chains saturate the local coordination environment and consequently provide little additional solvation contrast between the redox ions. The overcrowded sulfonate groups also introduce steric hindrance and induce electrostatic chain aggregation, which further reduces ΔS_solv. This non-monotonic dependence highlights the need for a balanced copolymer composition.

To elucidate the origin of thermopower enhancement in P(AAm-AMPS) electrolytes, we performed UV-vis spectroscopy to probe the specific interactions between AMPS-containing polymer and two redox ions. As shown in Figure 2C, the characteristic absorption peak of Fe(CN)₆^3- remains unchanged upon the addition of AMPS. In contrast, the Fe(CN)₆^4- signal exhibits a noticeable shift from 218 to 214 nm [Figure 2D], indicating a stronger interaction with the AMPS chains^[43-45]. DFT calculations further support this interpretation. The AMPS−Fe(CN)₆^4- complex exhibits a substantially more negative ESP minimum than AMPS−Fe(CN)₆^3- [Supplementary Figure 3], along with a more favorable binding energy [Supplementary Figure 4]. The pronounced UV-vis divergence, together with DFT calculations, demonstrates that the copolymer network preferentially interacts with Fe(CN)₆^4- over Fe(CN)₆^3- [Figure 2E]. Such valence-selective ion-polymer interactions effectively alter ΔS_solv to boost S_e.

Introducing a ΔC_r of the redox ions provides an effective route to boost S_e [as captured by Equation (1)]. Since this strategy hinges on the solubility contrast between the two redox species, we incorporated DMSO as an organic co-solvent to tune their dissolution thresholds [Supplementary Figures 5 and 6]. The S_e of PAAm and P(AAm-AMPS) electrolytes were then systematically examined with varying DMSO contents. Figure 3A and Supplementary Figure 7 depict S_e as a function of the DMSO volume fraction for the two hydrogel matrices. For the PAAm-based electrolyte containing DMSO (denoted PAD-X), S_e rises almost linearly with increasing DMSO concentration. In contrast, the P(AAm-AMPS)-based counterparts (denoted PSD-X) shows a non-monotonic, volcano-shaped dependence. The S_e reaches a maximum of 2.30 mV·K^-1 at 10 vol% DMSO, clearly surpassing that of PAD-10 (2.02 mV·K^-1), while maintaining excellent mechanical properties [Supplementary Figure 8]. In addition to the in-plane performance, the cross-plane S_e was also measured [Supplementary Figure 9].

This enhancement arises from the interplay between the selective ion-polymer interaction in the P(AAm-AMPS) network and the DMSO-induced ΔC_r. Minor DMSO addition mainly enhances ΔC_r through differential solubility between redox ions, with minimal impact on ΔS_solv. At higher content, DMSO increasingly enters the solvation shell of redox ions and competes with sulfonate groups for ion coordination, potentially diminishing their ΔS_solv contribution. Excess DMSO (> 10 vol%) also induces shrinkage of the hydrogel matrix [Supplementary Figure 10], which further restricts the local coordination space. The resulting loss in ΔS_solv outweighs the gain in ΔC_r, thereby dictating the non-monotonic trend in S_e. The PAD-X electrolytes, by contrast, lack active functional groups, so S_e follows the monotonic increase in ΔC_r.

Beyond S_e, the ionic conductivity (σ) of the hydrogel electrolytes also depends strongly on composition. At each DMSO concentration, the PSD electrolytes consistently outperform their PAD counterparts in σ [Figure 3B]. This improvement originates from anionic sulfonate groups in the P(AAm-AMPS) network, which promotes dissociation of redox ions and elevates effective charge carrier density^[46]. With increasing DMSO content, σ declines for all electrolytes, a trend attributable to combined effects of reduced ion concentration and increased viscosity [Supplementary Figure 11]. Despite this modest reduction in conductivity, the PSD-10 electrolyte delivers superior power output under various ΔT. The voltage-current and voltage-output power density curves of the PSD-10 based-QTEC exhibit characteristic linear V-I and parabolic V-P relationships [Figure 3C and D]. At a ΔT of 10 K, the device yields a short-circuit current density of ~5.4 A·m^-2, an open-circuit voltage of ~23.3 mV, and a maximum power density (P_max) of ~31.9 mW·m^-2. When ΔT is raised to 30 K, P_max increases to ~278.8 mW·m^-2. The normalized power density (P_max/ΔT²) of PSD-10 based-QTEC remains stable at approximately 0.32 mW·m^-2·K^-2 over the tested temperature range [Figure 3E]. This composition is confirmed as the optimal DMSO content [Supplementary Figure 12], at which the overall power density is maximized driven by the S_e and σ trends shown in Figure 3A and B. A comparison with recently reported QTECs [Supplementary Table 1] reveals that our design strategy yields competitive thermoelectrochemical performance.

Beyond these gains in thermoelectrochemical performance, the design also effectively suppresses thermal conductivity (κ). A low κ is essential to sustain a stable ΔT across the QTEC. Specifically, the value drops from 0.584 W·m^-1·K^-1 for pristine PAAm electrolyte to 0.561 W·m^-1·K^-1 for P(AAm-AMPS) electrolyte, and reaches a minimum of 0.512 W·m^-1·K^-1 for PSD-10 [Figure 3F]. Swelling measurements [Supplementary Figure 10] reveal the expansion of the polymer matrix after moderate DMSO (≤ 10 vol%) incorporation. This looser network architecture reduces continuous heat conduction pathways and adds polymer-solvent interfaces that enhance phonon scattering. This effect, together with the elevated viscosity induced by DMSO, contributes to the more pronounced decrease in thermal conductivity after DMSO addition compared with copolymerization^[47,48].

XPS analysis confirms the successful integration of the Fe(CN)₆^4-/3- in PSD-10. The high-resolution Fe 2p spectrum shows coexisting Fe²⁺ and Fe³⁺ states, a prerequisite for the thermogalvanic effect [Figure 4A]. During the operation of PSD-10-based-QTEC, the dynamic migration of redox ions under a ΔT was monitored by in situ Raman spectroscopy [Figure 4B]. The characteristic peaks at 2,058 and 2,093 cm^-1 correspond to Fe(CN)₆^4-, whereas the 2,132 cm^-1 signal belongs to Fe(CN)₆^3-[49]. At the cold side, the Fe(CN)₆^3- stretching vibration gradually loses intensity, consistent with its reduction to Fe(CN)₆^4-. The spectral features of two redox ions evolve in distinctly anti-correlated trends, a direct reflection of ΔT-driven ion migration and the accompanying redox reactions at the electrodes.

To establish the link between S_e enhancement and ΔC_r, we examined the precipitate formed in the DMSO-containing electrolytes. All characterizations consistently identify that the precipitate contains Fe(CN)₆^4- as the anionic component. It exhibits a yellow appearance identical to that of K₄Fe(CN)₆·3H₂O rather than the red hue characteristic of K₃Fe(CN)₆ [Figure 4C]; the XRD pattern matches the standard reference of K₄Fe(CN)₆ [Figure 4D]; Raman peaks are identical to those of pristine K₄Fe(CN)₆ [Figure 4E]. Together, these observations demonstrate that only Fe(CN)₆^4- selectively crystallizes.

UV-vis spectroscopy of the supernatant quantitatively corroborates the above selective crystallization. The Fe(CN)₆^4- concentration drops substantially after DMSO incorporation, whereas Fe(CN)₆^3- content remains nearly unchanged [Figure 4F]. This solubility disparity arises from differential ion-solvent interactions between the two redox ions^[15,50], and the resulting thermodynamic equilibrium is highly temperature-sensitive. Temperature-dependent UV-vis profiles [Figure 4G] capture a prominent recovery of Fe(CN)₆^4- absorbance upon heating, indicating that the precipitated K₄Fe(CN)₆ crystals redissolve as temperature rises. Direct visual confirmation is provided by optical microscopy images [Supplementary Figure 13]. This reversible crystallization–dissolution transition establishes ΔC_r under a ΔT, which directly contributes to the observed S_e enhancement.

The DMSO-induced crystallization of Fe(CN)₆^4- ions is physically confined within the PSD-10 hydrogel matrix. SEM images reveal micrometer-sized crystals embedded throughout the polymer network [Figure 4H]. The corresponding Raman spectra also display markedly stronger Fe(CN)₆^4- signals compared to the P(AAm-AMPS) electrolyte [Figure 4I]. This signal enhancement is attributable to physical scattering from the DMSO-induced solid Fe(CN)₆^4- precipitates. The localization of such precipitates within the hydrogel framework, consequently, facilitates a stable ΔC_r across the electrolyte.

The long-term stability of the PSD-10-based QTEC was evaluated under a quasi-continuous discharge mode [Figure 5A and B]. Upon applying a ΔT = 10 K, the open-circuit voltage rises to approximately 23.1 mV within a thermal accumulation period of about 10 min. The subsequent repeated discharge-recovery cycles show only a gradual, minimal voltage decay, attributed primarily to polarization effect^[51]. After 200 consecutive cycles, the voltage stabilizes at ~21.2 mV, corresponding to a retention rate of 91.8%. The power output remains highly stable over 200 cycles as well [Supplementary Figure 14]. Although the device remains stable over this timescale, gradual ion redistribution is inevitable over extended use, and intermittent operation with resting periods could potentially further improve the cycling performance. Figure 5C demonstrates that the output voltage remains stable during a continuous 10-hour operation under a constant ΔT. After encapsulation with polyimide tape, the assembled QTEC also maintained good performance retention when tested daily over 7 days [Supplementary Figure 15]. These results highlight the excellent operational durability of the PSD-10-based QTECs for sustained low-grade heat harvesting.

To meet practical voltage requirements under small temperature gradients, multiple QTEC units were connected in series using segmented PSD-10 hydrogel electrolytes, adjacent graphite electrodes, and copper ribbons interconnects [Figure 5D]. Increasing the number of series-connected units from 2 to 12 raises the thermovoltage from 39.1 to 294.7 mV at ΔT = 10 K, which confirms the scalability of the device. Elevated temperature gradients enhance the power generation capability, as demonstrated by the ~0.73 V output of the 12-unit module at ΔT = 40 K within a short time [Figure 5E]. Combined with a voltage amplifier, this module can power a small electronic device [Figure 5F], which demonstrates the practical applicability of the PSD-10 electrolyte. The hydrogel electrolyte is easy to prepare and can be fabricated at scale, which makes it compatible with advanced packaging technologies. These features highlight its potential for large-scale ionic thermoelectric energy harvesting.