Biomimetic microstructure design for ultrasensitive piezoionic mechanoreceptors in multimodal object recognition

Abstract

The challenge of achieving high recognition accuracy in artificial mechanoreceptors arises from the trade-off between sensitivity and stability in the sensing unit. Inspired by human skin, we developed a biomimetic approach that involves structural and engineering enhancements for ionic-conducting polyvinyl alcohol/Ti₃C₂T_x (PVA/MXene) composite hydrogel microneedles (HM) to enhance the sensitivity. By integrating the HM with a polyethylene terephthalate/indium tin oxide (PET/ITO) film, we create a non-faradaic junction that ensures stable electrical output without transmission loss under stimulation. Furthermore, the significant alteration in nanosheet spacing facilitates proton transport along the MXene microchannels, increasing the plasmonic gradient between the junction and the hydrogel's center, thereby boosting piezoionic efficiency. Consequently, the biomimetic sensing unit achieves a high power density of 165.6 mW m^-2 and exceptional sensing stability over 10,000 cycles. When combined with vertical memristor units, this system effectively captures and transforms characteristic signals from various objects, achieving a recognition accuracy of 90%.

Fig. 1. Object recognition system inspired by human skin.

a Schematic diagram of skin organs and ion channels on cell membranes. b The structure of HM-S (i): Polyvinyl alcohol/Ti₃C₂T_x (PVA/MXene) composite hydrogel microneedle (HM) was adhered on the polyethylene terephthalate/indium tin oxide (PET/ITO) substrate, followed by polydimethylsiloxane (PDMS) encapsulation. The conductance mechanism of HM-S (ii): Proton transport induced by interlayer spacing variation of MXene nanosheets. c Morphologies (i), elemental characterization, and FEA of HM (ii), with the application concept of human-machine interaction. d Multimodal object identification with an artificial neuromorphic network (ANN).

Fig. 2. Finite element analyses (FEAs) investigating the effect of geometric properties on mechanical response.

a A 3D plot comparing the displacement of HM-S and bulk hydrogel mechanoreceptor under a fixed load. b Schematic demonstration of the device’s modeling structure, surface morphology, and contact potential. (H and D represent the height and caliber of the cone, respectively. PDMS: Polydimethylsiloxane; PVA/MXene: Polyvinyl alcohol/Ti₃C₂T_x). c Schematic of the modeled circuit and comparison of capacitance changes (ΔC/C₀) between microneedle and bulk structures versus external stimuli. (C_object, C_bulk, C_air, C_cone, and C_base represent the capacitances of the contacting object, bulk structure device, air, cone part of the HM, and base part of the HM). d Comparisons of electric field distribution, and e energy density between the bulk mechanoreceptor and HM-S.

Fig. 3. Piezoionic effect generated by interlayer spacing variation-induced proton transport.

a Schematic illustration of the piezoionic effect in HM with proton-dominant transport. b Structure comparison between the bulk hydrogel and HM, with the cross-section morphology of HM. c FT-IR spectroscopy of HM with different MXene contents. d Rheological behavior of HM displaying the gel-point variation. e XRD, and f I-V curves of HM-2 with different water contents, indicating that the interlayer spacing variation induced the proton transport. The channel length (100 μm) was controlled. g Wideband dielectric testing of PDMS-encapsulated HM-2 under the temperature range of -20 to 60 °C (δ is the dielectric loss angle). h Conductivity (σ) and permittivity (ε) variation of the hydrogel versus temperature.

Fig. 4. Output/sensing performance and application demonstration of HM-S.

a V_oc, and b I_sc of HM-S with different sensing unit sizes. c V_oc of HM under slight-contact mode (0 % strain). d I_sc of HM-S under different strains and, e various frequencies. f The relationship between calculated power density (P) and resistance (R) of external loads (U and S represent the generated voltage loaded on resistance box and device area, respectively). The contact pairs were fixed as PE/HM-S for (a)-(e) and carbon fabric/HM-S for (f). g V_oc of HM-S and bulk hydrogel mechanoreceptor with distinct contact objects. The error bar illustrated in each column was obtained by calculating the mean and variance of the five peaks. h Texture recognition of cotton fabrics achieved by HM-S with shear/press modes (A is the fabric ampacity, reflecting the roughness of fabrics). i I_sc of HM-S while touching cotton fabrics, different am in shear (up) and press (down) modes. j Shape identification of polyvinyl chloride (PVC) balls by detecting the brightness and flicker frequency of the LED.

Fig. 5. The fabrication and algorithm-hardware combination of in-sensing and computing e-skin.

a Composition of sensing computing arrays, including HM-S and Ag-based memristor arrays. b Typical I-V characteristics of the ITO/Al₂O₃/Pt/Ag device in multiple tests under 10^-6 A compliance current (I_CC). c Current signals generated by a memristor array during regular contact with different sensing array materials. d Schematic of a three-layer artificial neural network. e The distribution of the high-dimensional feature vectors generated by the e-skin using the LDA method. The independent confusion matrices for (f) the training and (g) test sets. h The recognition accuracy of the in-sensing and computing e-skin.