An iontronic reservoir for highly robust neuromorphic prosthesis

Nature Materials (2026)Cite this article Neuromorphic prosthesis demands not only the assembly of neural architectures and functions but also robustness against unpredictable failures in dynamic physiological environments. While self-healing electronics have been demonstrated to restore synapse-like functions, their application to higher-order cognitive functions remains limited. Here we present a hydrogel-based iontronic reservoir that demonstrates exceptional physical and functional robustness for neuromorphic prosthesis. The nonlinear dynamics of the hydrogel–electrode interface can serve as a physical reservoir to preprocess time series, with minimized susceptibility to physical damage. Our system based on the hydrogel-based iontronic reservoir achieves 95% accuracy in speech recognition and can restore such capability within 0.02 s after reattaching the fractured points, outperforming biological systems in the neurorehabilitation process. Moreover, its pH-sensitive dynamics enable adaptive closed-loop neural stimulation control in a rat model, validating its potential for neural rehabilitation and sensorimotor function restoration. We expect such a hydrogel-based iontronic reservoir to improve both processing efficiency and robustness for next-generation neuroprosthetics and human–machine interfaces.This is a preview of subscription content, access via your institution Access Nature and 54 other Nature Portfolio journals Get Nature+, our best-value online-access subscription $32.99 / 30 days cancel any timeSubscribe to this journal Receive 12 print issues and online access $259.00 per yearonly $21.58 per issueBuy this articleUSD 39.95Prices may be subject to local taxes which are calculated during checkoutAll data supporting the findings of this study are available in the paper and its Supplementary Information. Other raw data are available from the corresponding authors upon request. Source data are available via figshare at https://doi.org/10.6084/m9.figshare.31153141 (ref. 56).MATLAB codes for data processing and algorithm implementation have been deposited at GitHub at https://github.com/ONEGroup-nju/Code-for-iontronic-reservoir.git.Harikesh, P. C. et al. Ion-tunable antiambipolarity in mixed ion–electron conducting polymers enables biorealistic organic electrochemical neurons. Nat. Mater. 22, 242–248 (2023).Article CAS PubMed PubMed Central Google Scholar Sarkar, T. et al. An organic artificial spiking neuron for in situ neuromorphic sensing and biointerfacing. Nat. Electron. 5, 774–783 (2022).Article Google Scholar Keene, S. T. et al. A biohybrid synapse with neurotransmitter-mediated plasticity. Nat. Mater. 19, 969–973 (2020).Article CAS PubMed Google Scholar Kim, Y. et al. A bioinspired flexible organic artificial afferent nerve. Science 360, 998–1003 (2018).Article CAS PubMed Google Scholar Lee, Y. et al. A low-power stretchable neuromorphic nerve with proprioceptive feedback. Nat. Biomed. Eng. 7, 511–519 (2023).Article PubMed Google Scholar Soman, S., Jayadeva & Suri, M. Recent trends in neuromorphic engineering.

Big Data Anal. 1, 15 (2016).Article Google Scholar Mead, C. How we created neuromorphic engineering. Nat. Electron. 3, 434–435 (2020).Article Google Scholar Zenke, F. et al. Visualizing a joint future of neuroscience and neuromorphic engineering. Neuron 109, 571–575 (2021).Article CAS PubMed Google Scholar Indiveri, G. & Horiuchi, T. K. Frontiers in neuromorphic engineering. Front. Neurosci. 5, 118 (2011).Article PubMed PubMed Central Google Scholar Harmon, L. D. Artificial neuron. Science 129, 962–963 (1959).Article CAS PubMed Google Scholar Tang, J. et al. Bridging biological and artificial neural networks with emerging neuromorphic devices: fundamentals, progress, and challenges. Adv. Mater. 31, 1902761 (2019).Article CAS Google Scholar Chen, S., Zhang, T., Tappertzhofen, S., Yang, Y. & Valov, I. Electrochemical-memristor-based artificial neurons and synapses—fundamentals, applications, and challenges. Adv. Mater. 35, 2301924 (2023).Article CAS Google Scholar Wang, T. et al. A chemically mediated artificial neuron. Nat. Electron. 5, 586–595 (2022).Article CAS Google Scholar Cao, Y. et al. Self-healing electronic skins for aquatic environments. Nat. Electron. 2, 75–82 (2019).Article Google Scholar Kang, J., Tok, J. B.-H. & Bao, Z. Self-healing soft electronics. Nat. Electron. 2, 144–150 (2019).Article Google Scholar Vo, N. T. P. et al. Autonomous self-healing supramolecular polymer transistors for skin electronics. Nat. Commun. 15, 3433 (2024).Article CAS PubMed PubMed Central Google Scholar John, R. A. et al. Self healable neuromorphic memtransistor elements for decentralized sensory signal processing in robotics. Nat. Commun. 11, 4030 (2020).Article CAS PubMed PubMed Central Google Scholar Stojek, Z. in Electroanalytical Methods: Guide to Experiments and Applications (eds Scholz, F. et al.) 3–9 (Springer, Berlin, Heidelberg, 2010).Bohinc, K., Kralj-Iglič, V. & Iglič, A. Thickness of electrical double layer. Effect of ion size. Electrochim. Acta 46, 3033–3040 (2001).Article CAS Google Scholar Saboorian-Jooybari, H. & Chen, Z. Calculation of re-defined electrical double layer thickness in symmetrical electrolyte solutions. Results Phys. 15, 102501 (2019).Article Google Scholar Tsunegi, S. et al. Physical reservoir computing based on spin torque oscillator with forced synchronization. Appl. Phys. Lett. 114, 164101 (2019).Article Google Scholar Everschor-Sitte, K., Majumdar, A., Wolk, K. & Meier, D. Topological magnetic and ferroelectric systems for reservoir computing. Nat. Rev. Phys. 6, 455–462 (2024).Article CAS Google Scholar Virgen-Ortíz, J. J. et al. Polyethylenimine: a very useful ionic polymer in the design of immobilized enzyme biocatalysts. J. Mater. Chem. B 5, 7461–7490 (2017).Article PubMed Google Scholar Naficy, S., Spinks, G. M. & Wallace, G. G. Thin, tough, pH-sensitive hydrogel films with rapid load recovery. ACS Appl. Mater. Interfaces 6, 4109–4114 (2014).Article CAS PubMed Google Scholar Appeltant, L. et al. Information processing using a single dynamical node as complex system. Nat. Commun. 2, 468 (2011).Article CAS PubMed PubMed Central Google Scholar Pei, M. et al. Power-efficient multisensory reservoir computing based on Zr-doped HfO2 memcapacitive synapse arrays. Adv. Mater. 35, 2305609 (2023).Article CAS Google Scholar Yang, Y. et al. Reservoir computing based on electric-double-layer coupled InGaZnO artificial synapse. Appl. Phys. Lett. 122, 043508 (2023).Article CAS Google Scholar Williams, G. & Watts, D. C. Non-symmetrical dielectric relaxation behaviour arising from a simple empirical decay function. Trans. Faraday Soc. 66, 80–85 (1970).Article CAS Google Scholar Doyne Farmer, J. Chaotic attractors of an infinite-dimensional dynamical system. Phys. D 4, 366–393 (1982).Article Google Scholar Kaczmarek, P., Mańkowski, T. & Tomczyński, J. putEMG—a surface electromyography hand gesture recognition dataset. Sensors 19, 3548 (2019).Article PubMed PubMed Central Google Scholar Lecun, Y., Bottou, L., Bengio, Y. & Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE 86, 2278–2324 (1998).Article Google Scholar Conrads, T. P. et al. High-resolution serum proteomic features for ovarian cancer detection. Endocri. Relat. Cancer 11, 163–178 (2004).Article CAS Google Scholar Fisher, R. A. The use of multiple measurements in taxonomic problems. Ann. Eugen. 7, 179–188 (1936).Article Google Scholar Rahman, T. COVID-19 Radiography Database. Kaggle https://www.kaggle.com/datasets/tawsifurrahman/covid19-radiography-database (2026).Kermany, D. S. et al. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell 172, 1122–1131.e9 (2018).Article CAS PubMed Google Scholar Chowdhury, M. E. H. et al. Can AI help in screening viral and COVID-19 pneumonia? IEEE Access 8, 132665–132676 (2020).Article Google Scholar Tawsifur, R. A. et al. Exploring the effect of image enhancement techniques on COVID-19 detection using chest X-ray images. Comput. Biol. Med. https://doi.org/10.1016/j.compbiomed.2021.104319 (2021).Zhong, Y. et al. A memristor-based analogue reservoir computing system for real-time and power-efficient signal processing. Nat. Electron. 5, 672–681 (2022).Article Google Scholar Goldberger, A. L. et al. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation 101, E215–E220 (2000).Article CAS PubMed Google Scholar Chen, C., Jafari, R. & Kehtarnavaz, N. UTD-MHAD: a multimodal dataset for human action recognition utilizing a depth camera and a wearable inertial sensor. In 2015 IEEE International Conference on Image Processing (ICIP) 168–172 (IEEE, 2015).Liu, D. et al. A wearable in-sensor computing platform based on stretchable organic electrochemical transistors. Nat. Electron. 7, 1176–1185 (2024).Article Google Scholar Imperadore, P., Shah, S. B., Makarenkova, H. P. & Fiorito, G. Nerve degeneration and regeneration in the cephalopod mollusc Octopus vulgaris: the case of the pallial nerve. Sci. Rep. 7, 46564 (2017).Article CAS PubMed PubMed Central Google Scholar Barbary, S., Dap, F. & Dautel, G. Finger replantation: surgical technique and indications. Chir. Main 32, 363–372 (2013).Article CAS PubMed Google Scholar Kadak, U. Modeling brain information flow dynamics with multidimensional fuzzy inference systems. Inf. Sci. 677, 120807 (2024).Article Google Scholar Arenas Gómez, C. M., Sabin, K. Z. & Echeverri, K. Wound healing across the animal kingdom: crosstalk between the immune system and the extracellular matrix. Dev. Dyn. 249, 834–846 (2020).Article PubMed PubMed Central Google Scholar Wang, F. et al. Self-healing memristors based on SA/PVA/STB hydrogel. J. Mater. Sci. Mater. Electron. https://doi.org/10.1007/s10854-023-10942-5 (2023).Wang, K. et al. A biodegradable, stretchable, healable, and self-powered optoelectronic synapse based on ionic gelatins for neuromorphic vision system. Small 20, 2404566 (2024).Article CAS Google Scholar Gao, Y. et al. Artificial synapses based on organic electrochemical transistors with self-healing dielectric layers. Chin. Chem. Lett. 35, 108582 (2024).Article CAS Google Scholar Qiu, R. et al. Bilingual bidirectional stretchable self-healing neuristors with proprioception. ACS Nano 17, 12652–12662 (2023).Article CAS PubMed Google Scholar Han, S. et al. An antifatigue and self-healable ionic polyurethane/ionic liquid composite as the channel layer for a low energy cost synaptic transistor. Eur. Polym. J. 174, 111292 (2022).Article CAS Google Scholar Van Der Werf, J., Buchholz, V. N., Jensen, O. & Medendorp, W. P. Neuronal synchronization in human parietal cortex during saccade planning. Behav. Brain Res. 205, 329–335 (2009).Article PubMed Google Scholar Shoaran, M. Next-generation closed-loop neural interfaces: circuit and AI-driven innovations. IEEE Solid State Circuits Mag. 15, 41–49 (2023).Article Google Scholar Zlokapa, A. et al. Fault-tolerant neural networks from biological error correction codes. Phys. Rev. E 110, 054303 (2024).Article CAS PubMed Google Scholar Ichinose, N., Kawashima, T., Yada, T. & Wada, H. Dynamical robustness and its structural dependence in biological networks. J. Theor. Biol. 526, 110808 (2021).Article CAS PubMed Google Scholar Deco, G., Rolls, E. T. & Romo, R. Stochastic dynamics as a principle of brain function. Prog. Neurobiol. 88, 1–16 (2009).Article PubMed Google Scholar Mengjiao, P. Source data. figshare https://doi.org/10.6084/m9.figshare.31153141 (2026).Download referencesThis work was supported by the National Key Research and Development Program of China (grant nos 2023YFE0208600 to C.W., 2021YFA1202600 to C.W., 2022YFA1203802 to Y. Li and 2021YFA0715600 to Y. Li), the National Natural Science Foundation of China (grant nos 62174082 to C.W., 92364106 to C.W., 92364204 to C.W., 82302321 to L.L., 62374081 to Y. Li and T2322010 to B.X.) and the Nanjing Municipal Science and Technology Bureau (grant nos 202305001 and 202205020 to C.W.). We thank the Leading Innovation and Entrepreneurship Team of Zhejiang Province, Ministry of Education Engineering Research Center for Optoelectronic Materials and Chip Technology and Nanjing University International Collaboration Initiative for support.These authors contributed equally: Mengjiao Pei, Tian Gao, Li Liu.School of Electronic Science and Engineering, National Laboratory of Solid-State Microstructures, Nanjing University, Nanjing, P. R. ChinaMengjiao Pei, Haotian Long, Hangyuan Cui, Xiang Li, Qinyong Dai, Kailu Shi, Lesheng Qiao, Baocheng Peng, Qianye Xing, Manhua Wen, Mengtao Han, Zhenhua Wan, Yun Li, Yi Shi & Changjin WanCollaborative Innovation Center of Advanced Microstructures, School of Physics, Nanjing University, Nanjing, ChinaTian Gao, Bin Xue & Yi CaoDepartment of Radiology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, ChinaLi LiuInstitute of Materials Research and Engineering (IMRE), Agency for Science, Technology and Research (A*STAR), Singapore, SingaporeWenlong Li, Yifei Luo & Xiaodong ChenInnovative Center for Flexible Devices (iFLEX), Max Planck-NTU Joint Laboratory for Artificial Senses, School of Materials Science and Engineering, Nanyang Technological University, Singapore, SingaporeWenlong Li, Yifei Luo & Xiaodong ChenKey Laboratory for Organic Electronics and Information Displays and Jiangsu Key Laboratory for Biosensors, Institute of Advanced Materials, Jiangsu National Synergetic Innovation Centre for Advanced Materials, Nanjing University of Posts and Telecommunications, Nanjing, P. R. ChinaZhaogang TengYongjiang Laboratory (Y-LAB), Ningbo, ChinaQing Wan & Changjin WanSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarSearch author on:PubMed Google ScholarW.L. and M.P. conceived the initial idea and organized the research frame. M.P., T.G. and H.L. contributed to sample preparation and device fabrication. M.P. performed the simulations and hardware measurements and analysed the results. L.L., Z.T. and M.P. performed the implantation experiments and analysed the results. H.C. collected speech signals. X.L. implemented the GIST algorithm. K.S. and Q.D. performed the electrochemical impedance spectroscopy measurements. L.Q., B.P. and Q.X. assisted in the discussion on device physics. M.W., M.H. and Z.W. contributed to the mathematical fitting of the device models. M.P., C.W., T.G., W.L., Y. Luo and L.L. wrote and refined the paper. C.W., X.C., Q.W., B.X., Y.C., Y. Li and Y.S. supervised the project. All authors discussed the results and implications and commented on the manuscript at all stages.Correspondence to Yun Li, Bin Xue, Yi Shi, Qing Wan, Xiaodong Chen or Changjin Wan.The authors declare no competing interests.Nature Materials thanks Kyung Min Kim, Tae-Woo Lee and Yoeri van de Burgt for their contribution to the peer review of this work.Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.a-c, VOUT-VIN hysteresis curves (red squares) and input current (black squares) characteristics for different electrode sequences. Output responses to electrical pulses (0.1 s duration) with varying input amplitudes for different electrode configurations: d, VOUT positioned near VIN electrode. e, VOUT positioned near the GND electrode. Impedance analysis and model fitting of the HIRE device. f, Schematic diagram of the device equivalent model. g, Impedance fitting with the fitting model shown in the inset. h, Statistical analysis of parameters from the fitting results.a, Output results triggered by electrical pulses (0.1 s) with different amplitudes of inputs. The extracted measurements (symbols) for the b, increasing processes and c, decreasing processes. The corresponding dashed lines are the fitting results. d, Fitting parameters as a function of current input voltage VIN(t) in the increasing processes. e, Fitting parameters as a function of the last state output voltage VOUT(t-1) in the decreasing processes.a, Preprocessing of the raw signal data by feature extraction and masking, then converting it to the input voltage of the device. b, One of the input time frames to be applied to devices and c, the corresponding outputs of the HIRE. d, Output weight matrix. e, Recognition results (confusion matrix) for ten voices. The recognition accuracy is 95.40%. f, Preprocessing of the raw signal data by feature extraction and masking, then converting it to the input voltage of the device. g, One of the input time frames to be applied to devices and h, the corresponding outputs of the HIRE. i, Output weight matrix. j, Recognition results (confusion matrix) for seven gestures. The recognition accuracy is 92.14%.a, Preprocessing of the raw signal data by feature extraction and masking, then converting it to the input voltage of the device. b, One of the input time frames to be applied to devices and c, the corresponding outputs of the HIRE. d, Output weight matrix. e, Recognition results (confusion matrix) for five types of grips. The recognition accuracy is 98.00%. f, Preprocessing of the raw signal data by feature extraction (Gabor convolution results) and masking, then converting it to the input voltage of the device. g, One of the input time frames to be applied to devices and h, the corresponding outputs of the HIRE. i, Output weight matrix. j, Recognition results (confusion matrix) for images of ten numbers. The recognition accuracy is 90.63%.a, Preprocessing of the raw signal data by feature extraction (Gabor convolution results) and masking, then converting it to the input voltage of the device. b, One of the input time frames to be applied to devices and c, the corresponding outputs of the HIRE. d Output weight matrix. e, Recognition results (confusion matrix) for three cases. The recognition accuracy is 90.30%. f, Preprocessing of the raw signal data by masking, then converting it to the input voltage of the device. g, One of the input time frames to be applied to devices and h, the corresponding outputs of the HIRE. i, Output weight matrix. j, Recognition results (confusion matrix) for diseased and non-diseased patients. The recognition accuracy is 96.67%.a, Different mask matrices applied to 5 devices based on decay-based positive integer sequences. b, Input sequences after mask matrix processing. c, One of the input time frames applied to devices and d, the corresponding outputs of the HIRE. e, Output weight matrix. f, Recognition results (confusion matrix) for healthy and arrhythmic heartbeats with recognition accuracy of 98.31%. g, Different mask matrices applied to 24 devices based on ±1 binary sequences. h, Input sequences after mask matrix processing. i, One of the input time frames applied to devices and j, the corresponding outputs of the HIRE. k, Output weight matrix. l, Recognition results (confusion matrix) for healthy and arrhythmic heartbeats with recognition accuracy of 96.13%. The proposed method employs a 3D decay-based positive integer sequence input mask, where 3D refers to the dimensionality of the mask matrix. This approach generates more complex and separable reservoir states, thereby enabling high-accuracy linear regression while maintaining low computational complexity.a, Preprocessing of the raw signal data by masking, then converting it to the input voltage of the device. b, One of the input time frames to be applied to devices and c, the corresponding outputs of the HIRE. d, Output weight matrix. e, Recognition results (confusion matrix) for three species of iris flowers. The recognition accuracy is 95.44%. f, Preprocessing of the raw signal data by masking, then converting it to the input voltage of the device. g, One of the input time frames to be applied to devices and h, the corresponding outputs of the HIRE. The first 100 points were selected for the waveform to be visible. i, Output weight matrix. j, Recognition results (confusion matrix) for 27 human actions. The recognition accuracy is 94.20%.a, Impedance-frequency relationship of original samples under different pH environments. b, Nyquist plots. c, VOUT-VIN hysteresis curves. d, IIN–VIN characteristics. e, Output responses triggered by electrical pulses (0.2 s duration) with input amplitudes ranging from 0.1 V to 1 V. Illustration in a created in BioRender. Pei, M. https://BioRender.com/s5u94ya (2026).a, Process flow for controlling mouse sciatic nerve stimulation using voice commands through pH-adaptive SHH-based bio-hybrid neuromorphic systems. b, Responses of the hydrogel reservoir to identical time series inputs under different pH conditions during voice command (‘Run’) recognition tasks. c, Recognition results under identical input command sequences for open-loop and closed-loop systems. d, The responses of the hydrogel reservoir to the same time series inputs before (green) and after (yellow) implantation. e, Confusion matrix for the spoken recognition task. f, H&E-stained sections (scale bar: 50 μm) of major organs from the sacrificed rat 9 days after implantation. Illustration in a created in BioRender. Pei, M. https://BioRender.com/io6ouei (2026).a, Schematic illustration of PDMS-based soft device. b, VOUT-VIN hysteresis loops and corresponding input current. c, Response to electrical pulse stimulation with different amplitudes ranging from 0.1 to 1 V and pulse width of 0.2 s. d, Forecasting results of the Mackey–Glass time series, where the pink line represents the actual data to be estimated and the black dotted line represents the predicted output based on the device model. e, 2-D display of the predicted results. f, Classification accuracy for nine different intelligent tasks using standard datasets. Data are presented as mean values ± SD with n = 10 (10-fold cross-validation).Supplementary Figs. 1–14, Notes 1–6, Tables 1–3 and Captions for Videos 1–3.The test environment set-up, and the cutting and healing processes of the HIRE device.Real-time tests including speech recognition, partial damage and functional recovery of the HIRE-based system.Activities of rat after implantation.Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.Reprints and permissionsPei, M., Gao, T., Liu, L. et al. An iontronic reservoir for highly robust neuromorphic prosthesis. Nat. Mater. (2026). https://doi.org/10.1038/s41563-026-02532-7Download citationReceived: 19 February 2025Accepted: 05 February 2026Published: 09 March 2026Version of record: 09 March 2026DOI: https://doi.org/10.1038/s41563-026-02532-7Anyone you share the following link with will be able to read this content:Sorry, a shareable link is not currently available for this article. Provided by the Springer Nature SharedIt content-sharing initiative