Introduction

Intrinsically stretchable organic electronics hold great promise for applications in wearable electronics1,2,3,4,5 volumetric displays6,7,8 and energy power supplies9,10,11,12,13,14. Yet, achieving high performance in these devices demands a delicate reconciliation of mechanical deformability with optoelectronic efficiency15,16,17. In the case of intrinsically stretchable organic photovoltaics (IS-OPVs), mechanical strain perturbs the morphology of the donor-acceptor (D/A) active layer at multiple length scales18, including molecular conformation19, crystalline packing20, orientation21, phase separation22,23, and interfacial alignment24. These structural changes have a strong influence on exciton diffusion, exciton splitting, and charge carrier transport, thereby limiting the power conversion efficiency (PCE) of OPV devices under strain25.

To address this challenge, recent efforts have focused on the flexibility of D/A blend films through molecular design, such as introducing flexible segments26,27,28, or blending with elastomers and plasticizers29,30. These approaches enhance the stretchability by increasing intra-chain disorder31 and inter-chain interactions32 via non-covalent or elastomeric networks33. While these strategies have led to crack-onset strains (COS) exceeding 30%, they were often obtained at the cost of reduced electronic performance. The initial PCEs typically remained below 15%, and the PCE under 30% strain tended to drop below 80% of its original value34,35,36. Parallel advances in deformable interlayers, such as stretchable electrodes and charge transport layers, have further aided mechanical resilience by redistributing applied stress37,38. Despite these advances, enhancing the intrinsic toughness of the active layer itself, which governs crack resistance and structural integrity under strain, remains underexplored. Improving this parameter could offer a more direct route to sustaining high efficiency and practical applications for stretchable devices.

Here, we report a molecular crosslinking and filler toughening (MCFT) strategy that synergistically enhances both the stretchability and toughness of OPV blend films. A covalent network formed between crosslinkable polymer donors and non-fullerene acceptors (NFAs) efficiently transmits stress across molecular domains, while silica fillers serve as crack stoppers and promote ordered aggregation. Together, these dual reinforcements yield IS-OPVs with improved mechanical robustness and high performance retention under strain, setting a benchmark for intrinsically stretchable solar cells.

Results

MCFT Structure design

To construct a robust crosslinked network, we first engineered both the donor and acceptor materials to be chemically crosslinkable with vinyl groups (Fig. 1a, Supplementary Fig. 1). The donor polymer, PzA5, was synthesized by incorporating 1,2-di(thiophen-2-yl)ethene (TVT) and the flexible 1,8-di(thiophen-2-yl)octane (TOT) segments31 into the conjugated backbone of the imide-functionalized benzotriazole (TzBI) polymer PTzBI-dF polymer39. This backbone engineering enables elastic chain dynamics and facilitates crosslink formation. Systematic variation of TVT and TOT ratios identified 2%TVT:3%TOT as the optimal composition, based on second-derivative analysis of performance trends (Supplementary Fig. 3 a–b).

Fig. 1: Molecular crosslinking and filler toughening (MCFT) strategy proposed for IS-OPVs.
figure 1

a Chemical structures of the designed crosslinkable donor, acceptor, and crosslinker. b FTIR spectra for crosslinkable segment (TVT) and crosslinker (PETMP). UVC refers to the sample after UV light soaking. c Schematic illustration of the synergistic MCFT strategy. d Chemical structure and photographic images of the silica aerogel powder and its gel state in chloroform. e Dynamic light scattering (DLS) analysis of the particle size distribution of the filler in a dilute chloroform solution. f Calculated non-covalent IRI and binding energies between ASH and the organic semiconductors (left: donor polymer tetramer; right: acceptor) with different dimer aligning conformation.

Full size image

To complement the crosslinkable donor, we developed a small-molecule acceptor, Y5TV, by functionalizing the widely studied L8BO series acceptors with 2-(pent-4-en-1-yl)thiophene units at both indene carbonitrile termini. These alkene moieties enable efficient thiol-ene “click” crosslinking. We employed pentaerythritol tetra(3-mercaptopropionate) (PETMP) as the tetra-thiol crosslinker (Supplementary Fig. 2). Upon mild thermal (100 °C) and UV (365 nm) activation for 5 min, a stable and covalently bonded donor-acceptor network of PzA5:L8BO:Y5TV, termed UVC, was formed, as verified by the disappearance of S–H vibrational bands in Fourier transform infrared (FTIR) spectra (Fig. 1b).

To reinforce the mechanical integrity of the UVC network, we introduced an aerogel silica nanoparticle filler with an average size of ~100 nm, modified with 1,1,1,3,3,3-hexamethyldisilazane (ASH) (Fig. 1 c–e). The addition of ASH yielded a sharp trade-off in performance with a sudden increase in modulus and a corresponding drop in PCE at filler loadings above 3 wt%, as revealed by second-derivative analysis (Supplementary Fig. 3 c–d). This critical loading, 3 wt%, was selected as the optimal concentration for mechanical enhancement without compromising device performance. Surface energy analysis revealed that ASH incorporation markedly increased film hydrophobicity, as evidenced by contact angle measurements (Supplementary Fig. 4). This surface modification enhances compatibility with the organic semiconductor matrix, facilitating homogeneous dispersion in chloroform without sedimentation.

To probe intermolecular interactions, density functional theory (DFT) calculations (B3LYP-D3/6-31 + G**) were performed. The ASH filler exhibited strong interaction region indicator (IRI) scattering plots and isosurfaces binding with both donor ( − 10.1 kcal mol⁻1 with PzA5) and acceptor ( − 28.0 kcal mol⁻1 with Y5TV) molecules (Fig. 1f and Supplementary Fig. 5). Consistently, X-ray photoelectron spectroscopy (XPS) results revealed downshifts in the Si 2p peak of ASH upon mixing with PzA5 (100.5 eV) and Y5TV (100.3 eV), relative to pristine ASH (100.8 eV) (Supplementary Fig. 6), confirming strong electrostatic interactions at the organic–inorganic interface. The Flory–Huggins interaction parameters (χ) derived from contact angle data were below 0.04 for both donor and acceptor with ASH, further supporting their favourable miscibility and stable dispersion (Supplementary Fig. 6). These findings underscore the feasibility of constructing a mechanically robust and molecularly cohesive network via the MCFT approach.

To elucidate the structural impact of the crosslinked D/A network and ASH filler, we conducted molecular dynamics (MD) simulations (Supplementary Fig. 7–8). The radius of gyration distributions revealed that replacing L8BO with Y5TV increased structural disorder (higher standard deviation, SD), which was further amplified by crosslinking and ASH addition. In blend systems, both donor and acceptor chains showed elevated SD values upon MCFT treatment, suggesting increased configurational heterogeneity and domain dispersion across length scales. This structural disorder likely contributes to enhanced entropic elasticity, which is critical for accommodating mechanical strain without catastrophic failure.

Enhanced stretchability while toughening

The mechanical enhancement conferred by the MCFT strategy was first assessed through uniaxial tensile testing of free-standing thin films (Fig. 2a), showing consistent results across three independent specimens for each sample, with the corresponding tensile parameters and standard deviations summarized in Supplementary Table 1. The pristine PzA5:L8BO blend exhibited brittle failure at low strain below 5%, while the UVC network significantly improved ductility, yielding a modulus of approaching 505 MPa and a crack-onset-strain (COS) of 44%. The further introduction of ASH resulted in a slightly reduced COS of 43%, likely due to local stiffness contrast from the rigid filler that promotes early strain localization. Nevertheless, interfacial debonding and viscoelastic dissipation around the filler enable efficient energy absorption, resulting in markedly enhanced toughness and modulus (1090 MPa for UVC + ASH). This combination of nearly unchanged COS with markedly reinforced modulus highlights the role of ASH in providing a toughening reinforcement effect.

Fig. 2: Mechanical performance of the blends.
figure 2

a Stress-strain curves of neat and blend films (three independent specimens per blend, with the maximum modulus marked). b Dynamic modulus of the blend films. c Stress distribution in finite element analysis on a matrix with filler particles of different modulus. d Snapshot of the stopped-extended crack in the UVC + ASH blend film (with molecular colours as the legend). e Stress-strain curves at the 100th cycle below yielding strain (3.1% for PzA5:L8BO, 8.8% for UVC, and 10.2% for UVC + ASH blend films, respectively). f Summary of various mechanical parameters for blend films.

Full size image

To probe the filler-induced network formation, dynamic mechanical analysis was conducted (Fig. 2b). The change in dynamic modulus between 0.2% and 2% strain (ΔE′0.2-2%), an indicator of the Payne effect40, was 9.5 kPa for the UVC + ASH film, significantly higher than that of the control samples (Supplementary Fig. 9a). Temperature-sweep DMA further revealed a pronounced rise in both storage and loss modulus near 100 °C, accompanied by a modest decrease in Tg (Supplementary Fig. 9b). The increase of E′ and E″ at elevated temperatures indicates the formation of a filler-induced reinforcing network that becomes active under strain, while the shift in Tg suggests altered chain dynamics associated with stronger filler–matrix interactions. These features consistently support the presence of a dynamic, strain-sensitive filler network that contributes to energy dissipation and mechanical reinforcement.

Finite element analysis (FEA) further clarified the role of ASH in mechanical stress distribution (Fig. 2c, Supplementary Fig. 10 a–c). Under tensile load, stress is concentrated at the filler–matrix interfaces, with ASH particles absorbing and redistributing stress while undergoing minimal deformation. This leads to a more uniform stress profile and the establishment of a rigid yet deformable mechanical skeleton within the composite. Additional FEA using a bilayer model revealed that the UVC + ASH blend also exhibits improved out-of-plane (OOP) toughness, with reduced bending-induced displacement compared to control films (Supplementary Fig. 10d–f).

To investigate fracture dynamics at the molecular level, we performed molecular dynamics (MD) simulations under periodic boundary conditions, which suppress edge effects and emphasize intrinsic tensile responses, with snapshots taken near the failure point (Fig. 2d, Supplementary Figs. 11 and 12). The PzA5:L8BO film showed abrupt displacement within L8BO-rich domains, indicative of a brittle fracture. By contrast, the UVC network delayed crack formation by enhancing covalent bonding and chain entanglement. When ASH was incorporated, cracks were arrested at filler interfaces while the surrounding network remained structurally intact, revealing a synergistic toughening mechanism where ASH particles act as crack stoppers and stress delocalizers. This is consistent with the optical imaging under strain (Supplementary Fig. 13), where ASH domains visibly interrupt fracture propagation.

Cyclic tensile testing further confirmed the improved mechanical performance of UVC + ASH films (Fig. 2e), which dissipated energy up to 144 MJ m⁻2, far exceeding that of PzA5:L8BO and UVC blends. A summary of the key mechanical parameters underscores the effectiveness of the MCFT strategy in simultaneously enhancing ductility, stiffness, and fracture toughness (Fig. 2f).

MCFT-based efficient IS-OPVs

The photovoltaic performance of the MCFT-based blends was first evaluated using a conventional device architecture: ITO/PEDOT:PSS/active layer/PNDIT-F3N/Ag. As shown in Fig. 3a, the reference PzA5:L8BO device achieved a moderate performance, with an open-circuit voltage (VOC) of 0.884 V, a short-circuit current density (JSC) of 24.4 mA cm⁻2, and a fill factor (FF) of 76.3%, yielding a power conversion efficiency (PCE) of 16.4%. The crosslinked UVC device exhibited a similar PCE of 16.7% but with an increased JSC and a decreased VOC. The lower VOC is consistent with the presence of traps introduced by UV crosslinking, resulting in higher non-radiative recombination and lower SCLC mobilities (Supplementary Fig. 14a–b and Supplementary Table 2). This is reflected by the light-intensity dependences of VOC and JSC (Supplementary Fig. 14 c-d): n increases from 1.33 to 1.36, indicating higher trap-assisted recombination in UVC, whereas α increases from 0.886 to 0.956, indicating improved charge collection with reduced bimolecular recombination.

Fig. 3: Photovoltaic performance of the devices.
figure 3

a J–V curves of rigid devices using different blend films including PzA5:L8BO, UVC, and UVC + ASH. b EQE response of the corresponding rigid devices. c Transient absorption intensity at donor GSB wavelength. d J–V curves for stretchable devices fabricated on TPU substrate. (inset: the schematic architecture of stretchable devices). e J–V curves under stretching strains for UVC + ASH stretchable devices fabricated on TPU substrate. f Summary of photovoltaic performance and stretching strain of 80% PCE retention of IS-OPVs reported in the literature (dashed lines represent the upper limit and the average level of the performances). g Evolution of the 0-0/0-1 absorption peak ratio under increasing strains (D and A correspond to the peak of donor and acceptor, respectively) for PzA5:L8BO and UVC + ASH blend films. h Charge extraction by linearly increasing voltage (photo-CELIV) results of PzA5:L8BO and UVC + ASH IS-OPV devices under varying strains. i Dependence of the carrier mobility calculated from photo-CELIV on the stretching strain.

Full size image

In contrast, the UVC + ASH devices delivered a higher PCE of 18.7% (VOC = 0.888 V, JSC = 27.2 mA cm−2, and FF = 77.5%). Upon adding ASH, the transport is restored to over 4.5×10−4 cm2 V-1 s-1 with near-unity hole/electron mobility balance (μh/μe = 0.93). Moreover, α further approaches unity (0.989) and n decreases to 1.17, indicating more efficient carrier collection and reduced trap-assisted non-radiative recombination.

To assess the effect of ASH on interfacial charge transfer, we analyzed the donor ground-state bleach (GSB) growth using a biexponential model, referenced at 0.2 ps. The extracted rise lifetimes (τ1, τ2) are 1.18 ps, 8.88 ps (PzA5:L8BO), 2.82 ps, 27.77 ps (UVC), and 0.63 ps, 12.89 ps (UVC + ASH). Crosslinking slows both components, consistent with UV-induced traps, whereas adding ASH mitigates this degradation, restoring and accelerating the dynamics: τ1 shortens to 0.63 ps (faster hole-transfer build-up) and τ2 decreases from 27.77 ps (UVC) to 12.89 ps, indicating a quicker establishment of the charge-separated population. These kinetic improvements align with the enhanced carrier collection and device metrics of the UVC + ASH films (Supplementary Figs. 15–17). Together with the red-shifted, intensified EQE (Fig. 3b and Supplementary Fig. 18), these kinetic and transport improvements account for the enhanced charge generation and extraction in UVC + ASH.

The balance between mechanical properties and photovoltaic performance is translated well into intrinsically stretchable OPVs (IS-OPV), which were fabricated using the architecture of TPU/PH1000/PEDOT:PSS (4083)/active layer/EGaIn (Fig. 3d and Supplementary Fig. 19). The UVC + ASH-based stretchable device delivered a PCE of 16.5%, with a VOC of 0.88 V, a JSC of 26.2 mA cm−2, and an FF of 71.53%, outperforming both PzA5:L8BO and UVC analogues (Supplementary Table 3). Crucially, these devices maintained 80% of their initial efficiency under a stretching loading of 1 kg, resulting in tensile strain of approximately 40% (Fig. 3e, Supplementary Fig. 19–20), representing one of the highest reported levels of strain endurance among IS-OPVs. In addition, cyclic stretching tests at defined strains (5%, 10%, or 20%) over 100–200 cycles further underline the advantage of UVC + ASH, which exhibits significantly higher performance retention than PzA5:L8BO throughout repeated loading–unloading processes. This balance of stretchability, toughness, and performance underscores the potential of the MCFT strategy for enabling deformable solar energy harvesting platforms (Fig. 3f and Supplementary Table 4). Depending on the target scenario, either higher strain tolerance or maximized efficiency at moderate strain may be more relevant, suggesting the need for complementary design strategies across different strain regimes.

To understand the underlying morphological evolution under strain, in-situ absorption spectroscopy was performed. The PzA5:L8BO blend showed rapid spectral disordering beyond 0.5–1% strain, indicating a fragile morphology (Supplementary Fig. 21, Fig. 3g). In contrast, UVC + ASH exhibited a strain-induced alignment before yield, followed by partial structural relaxation and recovery during strain hardening, signatures of reversible network reorganization. These adaptive behaviours were further validated by in situ photo-CELIV measurements. UVC + ASH devices showed minimal mobility decay and even a slight increase in peak current at low strain levels, highlighting the role of MCFT-enhanced morphology in preserving carrier transport under mechanical deformation (Fig. 3 h–i, Supplementary Fig. 22).

MCFT morphology

The morphological characteristics of the blend films were investigated using a combination of nanoscale mapping and scattering techniques. To probe molecular ordering, grazing incidence wide-angle X-ray scattering (GIWAXS) was performed (Fig. 4a, Supplementary Figs. 26 and 27). The UVC blend, formed via UV-induced crosslinking of PzA5:L8BO:Y5TV, displayed a broadened and diminished (\(1\bar{1}1\)) reflection of L8BO in the in-plane (IP) direction, indicative of reduced crystallinity and a shorter coherence length. By contrast, the UVC + ASH blend showed significantly enhanced (\(1\bar{1}1\)) and (021) diffraction peaks, signifying improved L8BO crystallization. The strengthened π–π stacking would increase carrier mobility, benefiting both JSC and FF, while the improved crystalline order lowers trap-assisted recombination, providing a modest VOC gain. This enhancement likely arises from the gel-like rheological behaviour of ASH (Supplementary Fig. 28), which modulates phase separation kinetics during solvent evaporation and thus reshapes the thin film morphology.

Fig. 4: Morphology properties for the photoactive films.
figure 4

a Grazing incidence wide-angle X-ray scattering sector-averaged I–q curves in the IP (dashed lines) and OOP (solid lines) directions. b AFM-IR mapping of the blend films at the wavenumber of 1426 cm−1 (scale bar: 5 μm). The color contours represent the IR intensity. c NK-RSoXS averaged I–q curves for PzA5:L8BO, UVC, and UVC + ASH blends. d Schematic of the effect of crosslinker and filler in the blend. The purple translucent strips represent the synergetic interwoven crosslinked networks in the blend film.

Full size image

Atomic force microscopy coupled with infrared spectroscopy (AFM-IR) revealed increased-scale donor-acceptor (D/A) networks in the crosslinked UVC blend (Fig. 4b, Supplementary Fig. 23), consistent with enhanced elastic behaviour. Upon ASH incorporation, the UVC + ASH film exhibited stronger molecular aggregation, finer phase separation, and a denser percolating network. Power spectral density (PSD) analysis of the AFM topographies shows a pronounced peak emerging at k ≈ 0.0025 Å−1 (d = 251 nm) upon crosslinking and furtherly shifting to ∼0.0015 Å−1 for UVC + ASH, accompanied by a marked suppression of spectral power at low ( < 0.001 Å−1) and high ( > 0.01 Å−1) frequencies (Supplementary Figs. 24 and 25). These features indicate a growth of the characteristic domain size together with the damping of large-scale inhomogeneity and nanoscale roughness, aligning with the formation of purer yet more fibrillar phases. Such pronounced morphological optimization can be ascribed to the strong interfacial interactions between ASH and the donor/acceptor components, which facilitate more favorable packing and phase organization. This fibrillar morphology facilitates continuous charge percolation pathways, while the finer yet purer domains promote efficient exciton diffusion and dissociation, jointly underpinning the observed JSC enhancement in devices.

Nitrogen K-edge resonant soft X-ray scattering (NK-RSoXS) was used to assess phase separation and domain formation in the blends. Energy contrast analysis identified 399.2 eV (N 1 s → π*) as optimal for visualizing Y5TV-rich domains (Supplementary Fig. 29 a–b). The scattering profiles revealed a distinct hump at q ≈ 0.0107 Å⁻1 in the UVC blend, corresponding to a characteristic domain size of ~58.5 nm (Fig. 4c, Supplementary Fig. 29c). Upon ASH incorporation, this feature became more intense and shifted to lower q (~0.0095 Å⁻1), reflecting the formation of larger and purer domains.

The combined results of GIWAXS, AFM-IR, and NK-RSoXS reveal that structural evolution under ASH incorporation is in line with the enhanced EQE response, where increased molecular ordering and refined, purer phase-separated domains collectively broaden absorption shoulders, extend the continuity for exciton diffusion and carrier transport across multiple length scales, and reinforce JSC and FF gains, consistent with results of TA spectroscopy and device EQEs. A schematic summary of the morphology evolution and the synergistic roles of crosslinker and filler is illustrated in Fig. 4d, highlighting the improved fibrillar domains and networks that underpin the enhanced mechanical and optoelectronic performance of the UVC + ASH system.

Proof-of-concept for applications of IS-OPV in powering

Compared to conventional solar cells on rigid substrates, IS-OPVs offer advantages for powering devices on curved and deformable surfaces. These include applications such as portable energy sources for camping tents, inflatable rescue gear, and wearable electronics for real-time bio-signal or motion tracking in dynamic environments. Such scenarios demand not only high flexibility and stretchability but also robust mechanical resilience to avoid device failure under repetitive or intense mechanical loads. To showcase real-world utility, we integrated the MCFT-based IS-OPV into a wearable, wireless electromyography (EMG) monitoring system (Fig. 5b), where it served as a self-powered energy module. The system successfully recorded muscle electrical signals, distinguishing between contraction and relaxation states (inset, Fig. 5b). Such functionality is directly applicable to biomechanical monitoring, including myoelectric prosthesis control, soft exoskeletons, human-machine interfaces, and fatigue detection via frequency-domain EMG analytics.

Fig. 5: Proof-of-concept for IS-OPV.
figure 5

a A wearable wireless electromyography (EMG) monitoring prototype charged by IS-OPVs fitted and packed as part of the elastic wristband. b The collected signals and the maximum operation time of the wireless EMG monitoring prototype in outdoor sunlight sports scenes (with and without IS-OPV power supply devices). c The power output of IS-OPVs with different areas under different light intensities estimated based on a PCE of 16.5%.

Full size image

In the proof-of-concept demonstration, the IS-OPV wristband integrated three serially connected large area devices with a total effective area of 4.5 cm2 (1.5 cm2 × 3, PCE = 14.25%, Supplementary Fig. 30). Under standard solar illumination (1 sun, 100 mW cm⁻²), this corresponds to a theoretical maximum output power of 65 mW (P = Ilight × area × PCE). The standalone wireless EMG monitoring system exhibited an average power load of 360 mW, of which the signal acquisition and processing module accounted for ~100 mW, while the wireless data transmission consumed the majority ( ~ 250 mW). With a battery capacity of 0.7 Wh, the system operated for 1.96 hours without external charging. When coupled with the IS-OPV charger, the operational time was extended to 2.30 hours, a 17% increase (Fig. 5c). Extrapolating this concept, enlarging the active area beyond 30 cm² would yield >435 mW under full sunlight (Fig. 5d), a level approaching the continuous power requirement of the device, and thereby potentially enabling near-continuous operation in outdoor conditions. These results underscore the feasibility of MCFT-enhanced IS-OPVs as mechanically robust, energy-autonomous power sources for next-generation wearable and off-grid electronics.

Discussion

In this work, we have demonstrated an MCFT strategy that enables IS-OPVs with improved mechanical durability and high photovoltaic efficiency. By integrating a UV-triggered covalent network between conjugated polymer donors and crosslinkable non-fullerene acceptors with a hydrophobic silica filler, the MCFT approach simultaneously enhances film stretchability, toughness, and charge transport. Mechanistically, the covalent D/A network provides long-range stress transfer and molecular stabilization, while the ASH filler acts as a crack stopper and morphological modulator. The synergistic interactions between these two components not only suppress crack propagation under mechanical load but also promote favourable domain size, crystallinity, and phase separation. These structural features lead to improved charge-transport and recombination kinetics, shorter exciton diffusion and dissociation times, more balanced carrier mobilities, and reduced degradation under strain, culminating in high PCE retention (80% at 40% strain) among reported IS-OPVs. Importantly, the MCFT strategy circumvents the typical trade-off between flexibility and performance seen in conventional softening methods, where elastomer or plasticizer blending often compromises device efficiency. Our approach achieves a modulus as high as 1090 MPa while maintaining a PCE of 16.5% in stretchable devices, benchmarks that position this system at the forefront of deformable solar energy technologies.

Beyond OPVs, the modular nature of MCFT, combining chemical crosslinking and physical filler reinforcement, potentially offers a broadly applicable framework for enhancing mechanical resilience in other organic optoelectronic systems, including light-emitting diodes and thin-film transistors. Its scalability and compatibility with diverse molecular systems further open avenues for robust, efficient, and form-adaptive soft electronics. As a demonstration, we integrated our IS-OPVs into a self-powered, wireless EMG monitoring system, highlighting their promise for energy-autonomous wearable electronics. Therefore, our study provides a foundational blueprint for advancing mechanically adaptive organic optoelectronics at the intersection of molecular engineering, materials mechanics, and device performance.

Methods

Materials and synthesis

Monomers and catalysts are purchased from Sigma-Aldrich, Strem, or Volt-Amp Optoelectronic Tech Co.Ltd, Dongguan, China, and used without further purification. All the solvents are purchased from Sigma-Aldrich and used without further purification.

Polymerization of PzA5

TzBI-dF-2Br (396.3 mg, 0.5050 mmol), BDT-Th-Sn2 (500.0 mg, 0.5316 mmol), TOT-2Br (7.0 mg, 0.0160 mmol), TVT-2Br (3.7 mg, 0.0106 mmol), and 12.3 mg Pd(PPh3)4 were dissolved in 15 mL anhydrous chlorobenzene in a 50 mL pressure tube under N2 protection. The mixture was sealed and stirred at 140 °C for 72 h. The crude product was collected by precipitation in methanol after cooling to room temperature. Then, Soxhlet extraction using a 0.22 μm membrane was carried out with methanol, acetone, n-hexane, dichloromethane, tetrahydrofuran (THF), and chloroform stepwise.

Synthesis of trimethyl(5-(pent-4-en-1-yl)thiophen-2-yl)stannane (TV-Sn)

Under −78°C and nitrogen flow protection, 200.0 mg of 2-(pent-4-en-1-yl)thiophene (1.31 mmol) was dissolved in 20 mL anhydrous tetrahydrofuran in a 3-neck flask. 0.78 mL n-BuLi solution (2.0 M in cyclohexane) was added dropwise and stirred for 1.5 h at 20°C afterwards. The flask was cooled down to −78°C again, and the chlorotrimethylstannane solution (1.0 M in THF) was added dropwise. It’s then stirred under 30°C for another 8 h. The crude product was washed and extracted with saturated potassium fluoride water solution and dichloromethane, respectively. It was then concentrated using vacuum rotary vapour and preliminary purified using recrystallization with DCM:MeOH=1:1. An oily yellow crude product was obtained and used for the next step.

Synthesis of Y5TV

Under the protection of nitrogen, Y5-2Br (200 mg, 0.1067 mmol), TV-Sn (84.0 mg, 0.2668 mmol), and Pd(PPh3)4 (12.3 mg, 0.0107 mmol) are dissolved in 100 mL anhydrous chlorobenzene, heated to 100°C, and stirred for 10 h and collected by precipitation into methanol. The crude product is purified with a polypropylene gel chromatography column with tetrahydrofuran as the eluent, obtaining a dark green solid with a yield of 32%. 1H NMR (400 MHz, Chloroform-d) δ 9.16 (s, 2H), 8.94 (dd, J = 8.4 Hz, 1.5H), 8.68 (dd, 1.5H), 8.09 (d, J = 5.2 Hz, 1.5H), 7.94 – 7.88 (m, 4H), 7.80 (d, J = 8.0 Hz, 1.5H), 7.40 (d, J = 10.6 Hz, 3H), 7.12 – 6.98 (m, 3H), 5.86 (m, 3H), 5.05 (m, 6H), 4.76 (m, 4H), 4.65 (d, 2H), 4.20 (s, 2H), 3.23 (d, 4H), 2.69 (m, 6H), 2.15 (m, 8H), 2.01 – 1.67 (m, 11H), 1.56 (m, 6H), 1.55 – 0.69 (m, 88H).

Simulation methods

The DFT calculation is carried out with the Gaussian 16 and Multiwfn software. The structure optimization for the ASH filler cluster and the molecular is conducted based on B3LYP-D3/6-31 g**. The dimer conformations are obtained with the forcite molecular dynamics. The exact situation of the dimer interaction is simulated based on B3LYP-D3/6-31+g** and illustrated with the Multiwfn and VMD software, with blue, green, and red colour scaling the electron density corresponding to the bonds, the vdW interaction, and the steric hindrance.

The radius of gyration is obtained from the forcite with the most stable structures going through the following procedures:s

For NFA molecular: (1) the crystal structure, including 108 molecules, is built. (2) The cells are subjected to successive dynamics of 100 ps molecular dynamics (NVT, T = 300 K), 50 ps molecular dynamics (NPT, T = 400 K), and 100 ps molecular dynamics (NVT, T = 300 K) to simulate the change in the crystal.

For the blends: (1) An amorphous cell with decamer chains of the polymer donor and NFAs (initial density = 0.5 g cm−3) is built using the structures from the DFT calculation. (2) 5 annealing cycles from 300 to 1000 K (NPT) are conducted to obtain random distributions. (3) The cells are subjected to successive dynamics of 100 ps molecular dynamics (NVT, T = 300 K), 200 ps molecular dynamics (NPT, T = 400 K), and 100 ps molecular dynamics (NVT, T = 300 K) to simulate the change in the cell. The final structures are used for further analysis.

The MD stress–strain simulations were performed using a Perl script. Each system was first equilibrated for 200 ps under zero external stress (300 K, NPT), followed by the application of successive uniaxial stresses of 0.001, 0.05, 0.10, 0.15, 0.25, 0.30, 0.35, and 0.40 GPa. At each stress level, the cell was equilibrated for 100 ps (300 K, NPT) to extract the actual stress and strain values of the cell. Fracture onset was identified from the stress response. In the visualization, PzA5 donor, L8BO acceptor, crosslinked PzA5:Y5TV network, and ASH filler are shown in red, blue, purple, and cyan, respectively. Representative snapshots at selected steps were taken to illustrate the locations where cracks initiate.

Finite element analysis of the mechanical properties

To provide an intuitive visualization of the tensile process, finite element models were constructed in Abaqus. The matrix dimensions were set to 30 × 40 (units), and spherical fillers with diameters of 8 and 4 (units) were embedded. For computational efficiency, a linear elastic constitutive model was adopted. The Young’s modulus of the matrix was set to 370 MPa, while that of the fillers was either 100 MPa (representing soft elastomeric additives) or 2000 MPa (representing rigid fillers). The unit system was kept consistent across the model. This simplification does not affect the validity of the observed trends, as the purpose of the simulation was to reproduce the qualitative behavior of uniaxial tensile experiments. A static analysis step was employed: one side of the model was fully fixed, while a constant tensile force of 10 N was applied along the X-axis at the opposite side. The resulting stress distributions are presented in Supplementary Fig. 10.

The comparison between the elastomer and filler blending is using the model in Supplementary Fig. 10, with the balls representing the elastomer or filler particles. A stretching force is applied to the right side with the left side of the model fixed. With moduli much less than or exceeding the matrix, the stress distribution in the matrix differs a lot. The sample using low-modulus blends has a significantly larger displacement. Notably, the stress concentrates at the interface between the low-modulus additive and the matrix, demonstrating heterogeneous intra-stress distribution due to the mismatch of modulus and elongation rate. In the circumstance of high-modulus filler, the filler particles take large shares of the stress due to their high modulus, which could alleviate the stress distribution in the photoactive materials. Besides, the negligible deformation of the filler particle prevents the stress concentration at the interface. These explain the improvement in stress resistance for photoactive layers using the high-modulus filler.

Supplementary Fig. 10d is the bilayer model with two different moduli; the two sides of the bar are fixed, and an upward force is applied to bend it. The model using the modulus of that measure for UVC + ASH film shows the least displacement under the bending force, demonstrating the improvement in stress durability for devices using photoactive layer with higher moduli.

Device fabrication and photovoltaic performance

The solid organic solar cells were prepared with a device structure of ITO/PEDOT:PSS/active layer/PNDIT-F3N-Br/Ag. Patterned ITO substrates were cleaned with tetrahydrofuran, deionized water, and isopropanol in an ultrasonic cleaner for 15 minutes each and 3 cycles. After drying at 75 °C, the ITO substrates were plasma-cleaned for 150 seconds. PEDOT:PSS was then spin-coated on the ITO substrates at 3000 rpm for 30 s and annealed at 150 °C for 15 min. The active layer (donor and acceptors are 5.4 and 6.5 mg mL−1, respectively) was dissolved in chloroform altogether and stirred for 12 h. The thickness of this layer was controlled between 90 and 110 nm, measured by a Bruker Dektak XT profilometer. It’s followed by thermal annealing at 100 °C for 5 min. The PNDIT-F3N-Br layer was then spin-coated onto the active layers in a methanol solution. Finally, Ag electrodes ( ~ 100 nm) were deposited on the PNDIT-F3N-Br.

For devices with active layers more than binary compositions, Y5TV takes 20% parts of the acceptors (1.3 mg mL−1) and L8BO takes the rest (5.2 mg mL−1). PETMP is dissolved in the blend solutions with a concentration of 0.065 mg mL−1. ASH is added to the blend solutions with a concentration of 0.3 mg mL−1.

The stretchable organic solar cells were prepared with a device structure of TPU/PH1000/PEDOT:PSS(4083)/active layer/EGaIn. TPU substrates were cleaned with deionized water in an ultrasonic cleaner for 15 minutes and 3 cycles. After drying at 30 °C, the TPU substrates were plasma-cleaned for 180 seconds. PH1000 (with 5 vol% dimethyl sulfoxide, 2 vol% polyethylene glycol, 0.5 vol% Zonyl fluorosurfactant FS-30 blended to improve the performance) was then spin-coated on the TPU substrates at 2000 rpm for 60 s and annealed at 100 °C for 10 min. PEDOT:PSS(CLEVIOS P VP AI 4083) was then spin-coated at 3000 rpm for 60 s and annealed at 100 °C for 10 min. The active layer materials (donors and acceptors are 5.4 and 6.5 mg mL−1, respectively) are dissolved in chloroform altogether and stirred for 12 h, then spin-coated as the parameters used in solid devices. Methanol was then spin-coated onto the active layers. Finally, EGaIn electrodes were deposited on the active layers by spraying as shown in Supplementary Fig. 19 with the masks of 0.04 cm2 area and 1.5 cm2 area, respectively. The area of the IS-OPV devices is estimated by multiplying the strain by (1+ε).

The hole-only devices were configured by ITO/PEDOT:PSS/active layer/MoO3 /Ag, while the electron-only device structure was ITO/ZnO/active layer/PFN-Br/Ag. The fabrication process was similar to the device above. Mobilities are calculated with the Mott-Gurney equation: \(J=\frac{9{\varepsilon }_{0}{\varepsilon }_{R}{\mu }_{h}{V}^{2}}{8{d}^{3}}\), which means \({\mu }_{h}=\frac{{k}^{2}{d}^{3}}{2.987\times {10}^{13}}\) (cm2V−1s−1). The J–V curves were measured with a Keithley 2400 source meter under 100 mW cm−2, AM 1.5 G solar simulator (Enlitech SS–F5). The J–V curves were measured with forward scan mode from −0.1 V to 1.0 V, and the scan step was 0.01 V with a dwell time of 1 ms. The light intensity was calibrated with a standard silicon solar cell certified by NREL. Light-intensity dependences were measured by varying the incident flux via the illumination aperture, followed by calibrating the intensity with the standard silicon solar cell. The EQE spectra were tested on a commercial QE measurement system (Enlitech, QER3011). Hole-only devices were recorded with a Keithley 236 sourcemeter under a dark chamber.

The output power of the IS-OPV module was calculated using the standard relation:

$${P}_{{OPV}}={I}_{{lig}ht}\times {A}_{{eff}}\times {{{\rm{PCE}}}}$$
(1)

where \({I}_{{lig}ht}\) = 100 mW cm−2 (AM 1.5 G, 1 sun), \({A}_{{eff}}\) = 4.5 cm2 (three devices in series, each 1.5 cm2), and PCE = 14.4%. This corresponds to a theoretical maximum output power of ~65 mW. For the wearable demonstration, a wireless Canum Venaticorum Arduino UNO-based electromyography (EMG) monitoring prototype was powered by a lithium-polymer battery (nominal voltage 3.7 V, effective capacity ~0.70 Wh), ensuring stable device operation. The average load power of the system was estimated from the measured discharge time (1.96 h) to be ~360 mW, with ~100 mW attributed to biosignal acquisition and processing and ~260 mW consumed by wireless data transmission. Three large-area IS-OPV devices were connected through a 330-type boost converter and interfaced with an SPV1050 solar power management module, supplying the EMG charging port to partially offset power consumption or recharge the battery. The effective contribution of photovoltaic charging was quantified by comparing discharge durations with and without solar input, with the improvement attributed to the continuous ~65 mW output from the IS-OPV module under 1 sun illumination. For scaling estimation, the same formula was applied to larger module areas, demonstrating that >30 cm² of active area could generate >430 mW under full sunlight in the future, approaching the continuous operating power of the EMG system.

Characterization methods

The NMR measurements were conducted on a Bruker AVANCE 400 spectrometer operating at 400 MHz, with tetramethyl silane (TMS) used as the internal standard. UV-vis absorption spectra of the solutions or the films spin-cast on quartz glass were measured using a Shimadzu UV-3600 UV-vis-NIR spectrometer. The FTIR is measured on a Nicolet IS50 - Nicolet Continuum, Thermo Fisher Scientific. The XPS is measured on an Axis Supra + , Kratos.

The IR-AFM is measured on an Anasys nanoIR3 system, Bruker. Intensity maps were mean-subtracted and Hann-windowed, followed by a 2D Fourier transform and azimuthal averaging to obtain the PSD.

The fully free-standing tensile test was performed using a HY-0230 horizontal tensile system (Shanghai Hengyi Precision Instrument Co., Ltd), following procedures described in our previous work31. The process includes four steps:

  1. (1)

    Thin film preparation: A PEDOT:PSS (Clevios P VP AI 4083) layer was spin-coated onto oxygen plasma-treated ITO glass at 2000 rpm, followed by thermal annealing at 150 °C for 10 minutes. The active layer blend solution was then spin-coated on top and annealed at 100 °C for 5 minutes inside a nitrogen glove box.

  2. (2)

    Film transfer: The coated substrates were floated on deionized water, allowing the PEDOT:PSS layer to dissolve and the ITO glass to sink. The hydrophobic active layer remained afloat, forming a freestanding film on the water surface.

  3. (3)

    Film mounting: The floating film was gently attached to the tensile stage clamps using van der Waals forces from two frosted aluminium blocks. Water was then slowly removed, and the film was air-dried on the stage.

  4. (4)

    Tensile testing: The film was stretched at a constant rate of 0.03 mm min⁻1 to record the stress-strain response. For cyclic testing, the film was extended to 5% strain at 3 mm min⁻1, held for 5 seconds, and then returned to its original position at the same rate.

GIWAXS experiments were carried out at beamline 7.3.3 of the Advanced Light Source (ALS), Lawrence Berkeley National Laboratory (LBNL). Thin films were deposited on PEDOT:PSS/silicon wafer substrates under the same processing conditions as those employed for device fabrication. The X-ray energy was fixed at 10 keV. Measurements were performed with the samples mounted on a stage inside a helium chamber, using an incident angle of 0.16°. The sample-to-detector distance was ~280 mm, calibrated with silver behenate. Scattering patterns were recorded with a Pilatus 2 M detector, featuring a pixel size of 0.172 mm × 0.172 mm.

RSoXS was performed at beamline 11.0.1.2 of ALS. The neat and blend thin films were prepared on PEDOT:PSS-coated wafer substrates under the same conditions as device fabrication, and then floated on deionized water. After dissolution of the PEDOT:PSS layer, the thin films were transferred onto PDMS membranes with a pin-hole (diameter = 2.5 mm) at the center. All samples were loaded onto a plate in the sample chamber and measured under high vacuum. The X-ray beam energy was screened within the N K-edge (390.0–400.4 eV). Scattering patterns were recorded using a Princeton Instrument PI-MTE CCD camera, with a pixel size of 0.027 mm × 0.027 mm and an exposure time of 5 seconds.

Reporting summary

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