Single Channel Based Interference‐Free and Self‐Powered Human–Machine Interactive Interface Using Eigenfrequency‐Dominant Mechanism

Abstract The recent development of wearable devices is revolutionizing the way of human–machine interaction (HMI). Nowadays, an interactive interface that carries more embedded information is desired to fulfill the increasing demand in era of Internet of Things. However, present approach normally relies on sensor arrays for memory expansion, which inevitably brings the concern of wiring complexity, signal differentiation, power consumption, and miniaturization. Herein, a one‐channel based self‐powered HMI interface, which uses the eigenfrequency of magnetized micropillar (MMP) as identification mechanism, is reported. When manually vibrated, the inherent recovery of the MMP causes a damped oscillation that generates current signals because of Faraday's Law of induction. The time‐to‐frequency conversion explores the MMP‐related eigenfrequency, which provides a specific solution to allocate diverse commands in an interference‐free behavior even with one electric channel. A cylindrical cantilever model is built to regulate the MMP eigenfrequencies via precisely designing the dimensional parameters and material properties. It is shown that using one device and two electrodes, high‐capacity HMI interface can be realized when the magnetic micropillars (MMPs) with different eigenfrequencies have been integrated. This study provides the reference value to design the future HMI system especially for situations that require a more intuitive and intelligent communication experience with high‐memory demand.

Specifically, the rapid development of IoT and artificial intelligence has now required a more effective HMI system to bridge the gaps between human and electric terminals. [34] To broaden the communication memory, one straightforward approach is to integrate device arrays. However, unlike rigid electronic materials, high-level integration is still a problem for flexible and wearable electronics. [35] Furthermore, the number of hardware configurations and signal wires increases along with the number of components in a flexible HMI interface, and the array inevitably brings the concern of portability and system complexity. From this perspective, it is particularly important to improve the information capability of individual device. [36,37] For example, Dai et al. proposed a flexible ternary sensor which can precisely perceive the bi-directional stimuli with non-overlapping response. [38] Upon inward bending, the optimized microstructures enabled more contact points for resistance decrease, while an outward bending generated increased resistance due to the applied strain. Using photolithography and thermal deposition, An et al. explored the response of metallic gratings to IR radiation from human hand for non-contact HMI, which can produce multiple commands based on the design of grating periods and duty cycles. [39] Recently, triboelectric hybrid coder was demonstrated, which combines the single-electrode and contact-separation mode to identify touch and press for distinguishable coding. [40] To date, however, to simply include a coding library into a single device for high-capacity flexible and wearable HMI is still challenging.
In nature, many existent elastic bodies, e.g. spring-block system, and cantilever beam, can freely vibrate under a specific eigenfrequency that is mainly determined by the inherent properties. [41,42] Inspired by this phenomenon, we consider that the design and establishment of an eigenfrequency library can be potentially applied to encode identifiable commands/signals without interference for HMI. Here, we report a flexible and wearable HMI interface, which is dominant by the eigenfrequency of a specific magnetized micropillar (MMP) using the cylindrical cantilever model. [43][44][45] When the MMP was deformed, the inherent recovery would cause the damped oscillation towards the equilibrium position. Based on Faraday's Law of Induction, the oscillation and corresponding variation of localized magnetic flux can be electrically perceived by the induced electromotive force (EMF)/electrical current in the conductive coil underneath. [46,47] More importantly, the conversion of electrical signals from time to frequency domain would figure out a specific eigenfrequency, which is mainly determined by the intrinsic property of the micropillar. On this basis, as depicted in Fig. 1a, the integration of MMPs with different eigenfrequencies on one device can enable a high-capacity HMI system for applications from information communication, robotic operation, to daily entertainment. Unlike conventional addressing-based multiple command system, one single channel is required here because the specific eigenfrequencies can provide the non-overlapping signals to precisely allocate subsequent commands without interference. Through the theoretical model, simulation, and experimental validation, we prove that the eigenfrequencies can be controllable via tuning the parameters of micropillar height, material modulus, or density, etc. Moreover, the MMPs are robust, and the stable eigenfrequency production ensures the reliability of eigenfrequency-dominant mechanism as an effective avenue to trigger intended command by one piece of device. Even though only one channel is used, the specific eigenfrequency (fij) can be precisely allocated for target commands (cij) without interference. b, Schematic diagram of the interface that contains MMP array with one coil substrate. Eigenfrequency (fij) of the MMP can be produced via vibrating the micropillars (pij), which is encoded with related commands for HMI. The colour of the micropillars indicates the elastic modulus (E) of the MMPA was tuned. c, Schematic diagram of a single magnetized micropillar. The height of the micropillar is H and the diameter of the cross-sectional circle is 2R. The right picture shows the schematic diagram of aligned NdFeB particles inside the pillar after magnetization. d, Optical images of the whole device with different MMPs on the human wrist. e, The simulation result of surrounding magnetic field around single magnetized micropillar. The slices indicate the magnetic scalar potential (A), and the arrow volume is related with the magnetic flux density. f, Process to generate the eigenfrequency based on the damped oscillation of an individual MMP via finger-induced deformation. The Prony analysis method was applied here to convert the electrical signals from time to frequency domain. g, Schematic diagram of the assembly with three specific MMPs of different eigenfrequencies. When the MMPs were vibrated consecutively, the voltage signals were captured via the single channel and converted to the distinguishable frequencies that can be identified by the terminal.

Design principle
The design principle to apply MMP eigenfrequencies for effective HMI is provided in Figs. 1a-b, which show that the micropillar arrays can be assembled together and only one electrical connection is required for signal collection. As discussed subsequently, the dimension and material property of the micropillars can both affect the eigenfrequency values. It is thus possible that the micropillars can be assembled with different heights, mechanical strength, or density, to realize the control of eigenfrequency. Via precisely designing the MMPs, the specific eigenfrequencies (fij) can be accompanied with the oscillation of the micropillars (pij), which can carry the embedded command (cij) to complete the intended HMI process. Consequently, with the integrated MMP array containing m rows and n columns, a coil device can totally produce command capacity of mn. Note that even the electrical signals are captured using the same channel, the specific eigenfrequency can be accurately identified to allocate related commands for communication. For example, when the micropillars P12 and P11 were vibrated, they will generate the eigenfrequency signals of f12 and f11, respectively. As the values of eigenfrequency are different, the mechanical inputs to vibrate specific MMPs can finally be transmitted and converted to the corresponding commands of c12 and c11 for HMI. The micropillars, composed of NdFeB particles, Polydimethylsiloxane (PDMS), and Ecoflex composite, were prepared based on a specific mass ratio. Detailed fabrication methodology of the MMPs and the conductive coil underneath are demonstrated in Supplementary Fig. 1a and In this work, the height and the radius of the micropillar were defined as H and R, respectively (Fig.   1c). After the magnetization along the in-plane direction, the embedded NdFeB particles aligned within the polymer base and the micropillar could serve as a flexible magnet with defined south (S) and north (N) poles. Due to high energy product of NdFeB [48] , the MMP shows a large remanent magnetization with excellent flexibility as indicated by the magnetic hysteresis curvature of NdFeB/silicone polymer composite (Supplementary Fig. 3). Thanks to the flexibility of the conductive copper coil and the MMP assembly, the device could be attached to the human skin for the wearable human-machine interactions. Fig. 1d shows the customized device on the human wrist.
As shown in the inset, both the MMPs and the coil can be properly bent according to the human wrist. The flexibility of the micropillars not only results in excellent wearable performance, but more importantly, the MMP can serve as a flexible magnet to deform and vibrate for signal generation.
The inset also shows that the micropillars on one device can be prepared with different heights, which was realized by defining the depth of the microholes in the mold (Supplementary Fig. 1b). position, the magnetic field distribution above the coil changes simultaneously. [49] Due to the electromagnetic induction, a voltage profile (EMF) will then be produced in the conductive coil according to the micropillar vibration. Through the time-to-frequency transformation, the frequency characteristics of the micropillar's oscillation could be determined. In principle, this eigenfrequency is mainly defined by the micropillar, which allows us to customize the MMPs for specific eigenfrequency production and command allocation. Details about the signal process of time to frequency domain conversion (Prony method) were discussed in Supplementary Note 1. As depicted in Fig. 1g, even though three micropillars have been integrated onto the same coil substrate with one communication channel, the vibration can produce distinguishable electrical signals. The oscillating signals can be converted to the non-overlapping signals in frequency domain, and thus the pillar-based eigenfrequency can be applied as an effective and accurate solution for the specific command allocation. In principle, with more integrated MMPs, the generation of specific eigenfrequencies can be allocated for more commands to build a multifunctional interface with one communication channel.

Characterization and authentication
To characterize the real-time vibration of the micropillar, the Laser Doppler Vibrometer (LDV) was employed to track the displacement of MMP throughout the process. We applied a tweezer to deform the free end of a typical MMP (H4.0P0.5E0.5), and the laser spot of LDV was focused at the free end to track the instantaneous displacement. Here, H4.0P0.5E0.5 indicates that the height of the MMP is 4 mm, and the polymer matrix consists of 50% PDMS and 50% Ecoflex in mass ratio. Note that the radius (R) of all MMPs in this work is fixed at 0.5 mm. The entire process is presented in following the waveform of damped sinusoid. In addition, we used LDV to record the dynamic velocity of MMP in a typical oscillation process, which also exhibits a periodical oscillation in damped sinusoidal behavior (Supplementary Fig. 4b).
As the flexible micropillar has been magnetized, the localized magnetic field would vary periodically to induce electrical currents in the coil underneath. Fig. 2c shows the schematic diagram to vibrate the MMP for measurement of the electrical current, and the optical image of the entire setup is displayed in Supplementary Fig. 5.
As discussed below, the impact location on the micropillar can be flexibly adjusted to investigate the deformation influence, while the rotation speed of the blade was applied to consider the effect from the external impact. Via vibrating the micropillar (H5.0P0.5E0.5) for five consecutive times, the typical induced electrical currents within the coil was shown in Fig. 2d. The red curve in the inset shows the enlarged real-time profile of the current in a damping mode, which is mainly caused by the recovery and periodical oscillation of the micropillar. The characteristics of the induced current basically follow the damped behavior of the free-end motion (displacement) as shown in Fig.   2a and Fig. 2b, which ensures that the electromagnetic induction is also an effective approach to reflect the vibration process. To explore the oscillation frequency that is related with a specific MMP, the conversion from time to frequency domain was performed on the induced current. A nonlinear fitting (OriginLab) was firstly introduced to process the original signal, and three main components in damped sinusoidal form, y(t) = cos (2 + ), were shown in Fig. 2e. Here, Ai is the related amplitude, fi is the sampling frequency, and θi is the phase constant. Note that all the signals were shifted down to cancel the noise from the electrical instrument before the data processing. Table S1 provides the details of fitting parameters, and the Prony method for conversion from time to frequency domain is explained in details in Supplementary Note 1. In principle, the oscillation eigenfrequency is determined by the inherent property, e.g. the dimension, and the mechanical strength, of the MMP. The Prony energy spectrum (Fig. 2f) shows a sharp peak at ~159.55 Hz across the whole frequency domain, which can be considered as the inherent eigenfrequency of the micropillar (H5.0P0.5E0.5) under the specific conditions. From this perspective, the incorporation of electromagnetic induction into vibrating micropillars can effectively generate the signals to determine the eigenfrequency. As discussed subsequently, the experimentally-obtained eigenfrequency is consistent with the theoretical model, thus enabling the precise adjustment of the eigenfrequency to realize a frequency-dominant interactive interface.
Supplementary Fig. 6 further compares the eigenfrequencies from the oscillating electrical signals in Fig. 2d. The stable generation of frequency signals across different tests ensures the reliability of the proposed mechanism and interface. We notice that for the real applications, the human finger may not be able to touch the MMP at a fixed location due to the small micropillar size Prony energy spectrum was given in Fig. 2g. It can also be observed that the shapes of the frequency spectra are similar, and the peaks of the eigenfrequencies locate almost at the same position ( Supplementary Fig. 8b). The inset further exhibits a slight "right shift" of the eigenfrequency when the impact position was moved up from Location 3 to Location 1. This behavior might be attributed by the damping effect from the air, which imposes a tiny revision to the eigenfrequency.
As the parasitic drag of the surrounding medium (air) is proportional to the square of the instantaneous speed, the lower impact position would induce a higher oscillating speed to cause the decay of the eigenfrequency. [50] The results confirm that the impact position on the MMP has negligible influence on the eigenfrequency of a given micropillar. We further discussed the influence on the eigenfrequency when the impact velocity of the blade was changed. The vibrating velocity could be flexibly tuned by changing the motor speed. As shown in Supplementary Fig. 5, the motor speed was adjusted from ω0 to 4ω0 (ω0 is 10% of the speed upper limit) to impact the micropillar at  Fig. 10a). However, the eigenfrequency, which was mainly determined by the micropillar, remains almost identical without obvious interference from the position on the coil ( Supplementary Fig. 10b). Fig. 2i further shows the long-term stability of the signals when the MMP (H5.0P0.5E0.5) was exposed to cyclic deformation for ~1400 s. No obvious variation of the eigenfrequency peaks was observed after the fatigue test, indicating the robustness of the microscaled MMP as a reliable interface for practical wearable interactions (Fig. 2j). It can be expected that the flexible and elastic MMP can withstand the mechanical deformation that is induced by the human finger under daily conditions.

Customization of eigenfrequencies
For a reliable, multifunctional and effective HMI, it's crucial to generate identifiable eigenfrequencies so that the customized commands can be allocated to the specific micropillars.
Considering this, a cylindrical cantilever model was adopted to investigate the parameters that can possibly tune the eigenfrequency of the MMP. The schematic diagram exhibits that via tuning the dimension (height, H) and material property (modulus/density, E/ρ) of the MMP, the related eigenfrequency (f1 and f2) can be possibly regulated (Fig. 3a). Based on the specific eigenfrequency design, the micropillar array with controllable eigenfrequency can thus enable a multi-functional interface for interactions, which is established on one coil device. For example, when the electrical terminal receives the frequency of f1, a command can be performed, while another frequency of f2 can trigger another pre-encoded command without interference.  Fig. 11 and Table S2). According to the cylindrical cantilever model, the relationship between eigenfrequency (f) and related parameters is given by: where H is the height of the MMP, E is elastic modulus, R is radius of cross-sectional circle, and ρ  Fig. 11d). To further confirm the frequency difference, a high-speed camera was used to record the oscillating behavior of the MMPs within one whole cycle (Fig. 3e and Video S1). As shown in the snapshots, one oscillation period is roughly 3.6 ms for H4P0.5E0.5 and about 8.0 ms for H6P0.5E0.5, which well matches the theoretical and simulated eigenfrequencies. To experimentally exam the eigenfrequency, we tuned the height (H) and the modulus/density (E/ρ) of the micropillar and the induced currents in the coil were recorded as shown in Supplementary Fig. 13 and Supplementary Fig. 14 Fig. 3g shows the experimental, theoretical, and simulated eigenfrequency of MMP for different heights from 4 mm to 6 mm, while keeping the E/ρ at 777.6 Pa·m 3 /Kg (related with case of P0.5E0.5 as shown in Supplementary Fig. 11d). The    Fig. 4a provides the comparison between the conventional and eigenfrequency-based high-capacity HMI system. Normally, the electrical channels of mn are required for conventional interface to establish a command capacity of mn. [51,52] When the mechanical input was applied to one specific sensor, the corresponding channel received the electrical pulse and the pre-set operation or command would be delivered to the terminal. Consequently, this addressing-based approach normally requires the electrode amount of mn2 to avoid interference, which will bring complexity to the interface with increased values of m and n. As discussed above, the eigenfrequency of MMP can be customized to build-up the interaction system that is mainly based on the differential frequency and P4-H6.0P0.5E0.5) on one coil substrate to establish the HMI system. The design of these MMPs will generate four different eigenfrequencies, f1, f2, f3, and f4, that presents the progressive decrease owing to the increased height from 4.0 mm to 6.0 mm. The optical image (Fig. 4b) shows that the four MMPs can be integrated to build up a multifunctional HMI interface with two electrode connections. When a typical MMP was vibrated by the human finger, the coil underneath would perceive the magnetic field variation that is related with the micropillar oscillation. As shown in Fig.   4c, the one-output terminal can collect the signals from the vibration of the MMP, and convert to individual frequency for the interaction with multi-commands. In principle, not only the height but also the mechanical property of the MMPs can be applied to tune the eigenfrequency if more commands are required.

Demonstration of eigenfrequency-dominant high-capacity HMI
To illustrate the reliability of the interface, each MMP was continuously vibrated for 50 times by manual operation and the frequency of MMPs' oscillations was recorded by a LabVIEW-based script (Supplementary Fig. 15). As presented in Fig. 4d Fig. 4e, the vibration of specific MMPs can be perceived and lighten the lamps of 1-4 (Video S2). When an MMP with a particular frequency is vibrated by a finger, the electrical signal was induced and the corresponding eigenfrequency could be determined by the software in the computer terminal, which finally was reflected by the highlighted number as displayed. To further investigate the interface reliability for identification of the input source, five volunteers participated in the test to successively vibrate P1 and P2 for ten times and the accuracy was summarized in Fig. 4f. As shown in Video S3, the two MMPs were identified with accuracy of over 90%, showing that the eigenfrequency-based mechanism is of reliable potential in the daily human-machine interactions towards a broad spectrum of users.
Based on four MMPs design, we demonstrated that the assembly could be applied to produce different allocated commands for effective robotic arm control and the authentication system. The schematic diagram in Fig. 5a shows that the integration of multiple MMPs is capable to produce different commands of "Lift", "Rotate", and "Grasp", etc. With two electrodes and one channel, the distinguishable eigenfrequencies can be used to allocate the specific command with accuracy. For robotic control application, we show that the four MMPs system is effective to move the target from one to another location (Fig. 5b). As shown in Fig. 5c, the experimental setup consists of the wearable interface (4 MMPs and the coil), the computer system and the robotic arm. The oscillation of the MMPs were captured by the coil beneath and transmitted to the software script for frequency determination, then a command was sent to the robotic arm to perform the corresponding operation that has been encoded. When P1 was vibrated, the interface received the corresponding eigenfrequency signal, and the pre-defined command of "Move down" was transmitted to the robotic arm. A "move down" action was performed by the robotic arm to approach the tomato target (Fig.   5d). When P2, P3 and P4 were vibrated successively, corresponding commands, e.g. to grasp the tomato, to rotate right and release the tomato, were conducted on demand. As shown in Video S4, this entire interaction process requires only one MMPs-based interface to achieve the multicommands, instead of using the device array with complex wiring connections. We further demonstrated that the MMPs-based interface could be applied for effective password coding and decoding in wearable HMI (Fig. 5e). To improve the security level, it is inevitable to introduce more digits, which imposes the concern of wiring complexity for the entire system. Alternatively, the eigenfrequency-based mechanism allows the establishment of authentication in a more effective manner. For example, a total amount of 5040 password combinations can be generated if ten MMPs were integrated and four different digits were selected for permutation (Fig. 5f). Note that the interface uses the eigenfrequency for number identification (0-9) and only one channel is required to connect the mechanical input and the electric terminal. Fig. 5g shows the process of password inputting and unlocking based on our proposed system by means of eigenfrequency using four as a numeric keypad. When an MMP is vibrated, the corresponding number would be typed into the input window based on the received eigenfrequency. If the vibration sequence matches the predefined password, the software could detect the inputs and the system was unlocked successfully.
The vibration sequence of P1, P4, P3, and P2 results in the input of 1-4-3-2 to unlock the software interface (Video S5). Supplementary Fig. 16 also shows that via re-defining the sequence of micropillar vibrations, another password ("4123") can be encoded through the interface. The vibration order of P4, P1, P2, and P3 was thus applied to realize the unlocking function. From this perspective, a password setting with higher confidentiality, e.g. more digits, can be accessible based on more MMPs with different eigenfrequencies. The design would not bring the wiring or space burden to the entire interface, because the designed micropillars can be integrated onto one coil and only one electrical output was required to enable the recognition process. As mentioned above, the increased demand of command capacity would bring the complexity of wiring system or electrode connections, especially for the cases when values of m and n are becoming larger. However, the eigenfrequency-based approach can effectively avoid the limitation and two electrodes are normally required even with more MMPs in the same system. Figs. 6a-b present the demonstration of a wearable electronic piano that can produce seven musical tones based on different MMPs. Within an identical coil, seven MMPs (P1-P7) with heights of 4.0 mm, 3.8 mm, 3.6 mm, 3.4 mm, 3.2 mm, 3.0 mm, and 2.8 mm were integrated and each micropillar represents a specific tone from "Do" to "Si". The MMPs were integrated in a row (71), and arranged in the form of continuously increased eigenfrequency from P1 to P7. As shown in Fig. 6c, the recorded eigenfrequencies of P1, P4, and P7 are 221.78 Hz, 298.65 Hz, and 447.03 Hz, respectively. The nonoverlapping behavior of eigenfrequencies allows the users to allocate specific musical tones to the micropillars instead of using a wiring system for location addressing. [53,54] For example, the related keys of "Do", "Fa", and "Si" were recognized when the micropillars P1, P4, and P7 were manually deformed ( Fig. 6d and Supplementary Fig. 17). Video S6 also records the complete process to produce musical tones from "Do" to "Si", which can be easily performed based on the wearable interface that has been integrated with seven MMPs of different eigenfrequencies. Furthermore, the vibration of MMPs can be potentially applied to reflect the motion trajectory as each micropillar can serve as a specific location that can be identified via the non-overlapping eigenfrequencies. [46,55] When a specific eigenfrequency is received, the related location can be determined and the continuous deformation of the micropillars can be combined to reflect the trajectory from different pixels, such as "MACAU" (Fig. 6e). As a proof of concept, a 33 MMPs array was fabricated to exhibit the potential for trajectory mapping and interactive writing. Fig.6f shows the assembly of nine MMPs with a total area of ~2.2 cm2.2 cm, which was then assembled with the coil for HMI demonstration (Supplementary Figs. 18a-b). The heights of the MMPs in Fig. 6g were designed from 4.2 mm (P1) to 2.8 mm (P9), and the related eigenfrequencies were recorded as shown in the normalized spectrum in Fig. 6h. With increased height, the eigenfrequency continuously decreases from 471.25±0.80 Hz to 229.55±0.84 Hz. The frequency trend follows the governing formula of ∝ −2 , where a larger MMP height (H) would lead to the decreased eigenfrequency in exponential behavior (Fig. 6i). With the non-overlapping eigenfrequencies, it is thus possible to identify the positions of mechanical inputs and convert to the real-time trajectory. Fig. 6j displays the vibration path of P3, P5, and P7 can be applied to represent the diagonal trajectory of Path 1, and a letter "U" was successfully realized with a more complex triggering path of P3, P6, P9, P8, P7, P4, and P1 ( Fig. 6k). Another path (Path 2) and letter "C" were also provided in Supplementary Fig. 18c through the combinational oscillation of different MMPs, and the complete interaction process was recorded in Video S7. With design of more MMPs, we believe that the eigenfrequency-based mechanism can provide the possibility to realize more functions for further HMI, while without imposing the burden on the overall system consumption and the number of connecting electrodes.

Conclusion
In this work, we reported a wearable HMI interface which uses the regulation of eigenfrequency as the dominant perception mechanism. When the MMPs were deformed, the intrinsic oscillation caused the variation of the magnetic field distribution and the instant current was received in the conductive coil. By converting the signals from time to frequency domain, the eigenfrequency that is mainly determined by the micropillars could be identified as the marker for subsequent coding and decoding. Along with the theoretical model and simulation results, we experimentally proved that the eigenfrequency can be flexibly customized via the material property and dimension of the micropillars. Based on one coil device, we built up a conceptual interactive platform which consists of four MMPs with different heights and thus the eigenfrequencies. The platform could conveniently produce multiple instructions through the intrinsic oscillation of specific MMPs, exhibiting the potential for uses in robotic arm control and password recognition system. With more MMPs, the non-overlapping eigenfrequencies allow us to realize the electronic piano or the trajectory mapping/hand writing on the platform using one electrical channel. The demonstrations confirm the improved capacity and functionalities would not bring the concern of electrode numbers or connection system that might affect the wearability. Thanks to the robustness of the MMP, the interference-free eigenfrequency generation also verifies the developed interface can be a reliable and accurate medium for daily applications. We believe that the design of eigenfrequency-based HMI device could be more effective to address the continuously increasing demand of multiple commands for the next-generation HMI and IoT era.

Fabrication of MMP and conductive coil
Laser patterning of the copper coil was conducted with laser engraving machine (LPKF ProtoLaser U4, LPKF Laser& Electronics AG, Germany). The width of each loop is ~70 µm and the distance between two adjacent conductive lines is ~80 µm. The coil consists of two layers, which are separated by an insulating polyimide film, and each layer contains 50 turns. After engraving, the coil was bathed in citric acid to remove the oxidized layer, and a drop of conductive silver glue was deposited on the hole to connect both layers. Finally, a plastic capsulation was applied to the coil to avoid oxidation during the use. For the MMP, a plastic Polymethylmethacrylate (PMMA) mold with pre-designed micro-hole array was used as the template for the micropillar (NdFeB/PDMS/Ecoflex) preparation. The mold was fabricated using the engraving machine (CNC-3020, JingYan Instruments& Technology Co.). First, we uniformly mixed the Polydimethylsiloxane (PDMS) gel, Ecoflex, and NdFeB particles with a specific mass ratio, and the composite was poured into the plastic mold and cured on the hot plate at 80 ℃ for 30 min to ensure complete solidification. After solidification, the cured PDMS/Ecoflex/NdFeB composite was peeled off from the mold, and placed in the magnetic field with strength of ~3 T and in-plane orientations for magnetization. Once magnetized, the MMP sample was stuck onto the copper coil after surface-treated by plasma cleaner (Harrick Plasma, USA), which helps to improve the interfacial adhesion.
For the MMPs used in the demonstration of robotic control and password setting, the plastic mold was firstly fabricated with four micro-holes (radius of 0.5 mm, distance of 20 mm). The micro-holes with different depths were drilled on the PMMA board (5 cm×5 cm×1 cm) using a milling cutter (radius of 0.5 mm). The depths of the four holes are 4 mm, 4.5 mm, 5.5 mm, and 6 mm, respectively. Then, Ecoflex, PDMS, and NdFeB particles were mixed uniformly with the mass ratio of 1:1:4, and poured onto the prepared mold. The total assembly was then placed in the vacuum chamber to remove the gas that has been trapped in the micro-

LabVIEW interface design for HMI demonstration
Firstly, we configured the data acquisition parameters (including physical channels, sampling rate, sampling numbers, and trigger conditions, etc.) via the built-in DAQmx series subVIs of LabVIEW. Since the preamplifier noise at this setting is in range of 0-1 µA, we set the triggering condition at 1 µA. In addition, we set the sampling time as 100 ms because the micropillar oscillation usually lasts for tens of milliseconds.
When the acquisition was triggered, data in 100 ms were transmitted to the Spectral Measurement Express VI for spectrum processing and the maximum frequency is extracted using the Tone Measurement Express VI. The outputs of the Tone Measurement Express VI are used as inputs and matched to the pre-set frequency bands to realize multiple functions for different applications.

Simulation of magnetic field
The software, COMSOL Multiphysics 5.6, was employed to simulate the magnetic field around the MMP.
To be consisted with the experiment, we adopted a three-dimensional model. Geometry of MMP was imported with 3D printing files with different dimensions. To simplify the model, the N54 (sintered NdFeB) was given to the micropillars. Air atmosphere was set with the dimension of 10 mm × 10 mm. The mesh was controlled by the physics interfaces with regular size. Magnetic field (no current) module was employed to trace the magnetic scalar potential and magnetic flux density, with the governing equations = −∇ and ∇ • = 0, where is the magnetic field vector, = 0 is the magnetic flux density vector and the magnetic scalar potential. The boundary conditions were set as • = 0. The initial value of magnetic scalar potential was set as 0. The constitutive relation between the magnetic field and magnetization was governed by = 0( + ), where is the magnetization vector.

Simulation of eigenfrequency
COMSOL Multiphysics 5.6 was further employed to simulate the eigenfrequency of MMP with different dimensional and physical parameters. Solid mechanics and eigenfrequency module were chosen to compute eigenmodes and eigenfrequency of the MMP. Cylinder models with certain dimensions were constructed in built-in geometry module. PDMS was given to the MMPs, and the material property (density, Young's modulus, etc.) was kept consistent with the experimental results. One end of the cylinder was applied a fixed constrain, and the other end was set at free. To determine the vibration state of the MMP at different frequencies of forces, a boundary load of 0.1 N/m 2 was applied to different MMPs and a probe was attached at a location of 3 mm from the fixed end. The displacement at different frequencies of boundary load was recorded. When a maximum displacement was obtained, the applied frequency was considered as the resonant frequency, and thus the eigenfrequency of the investigated MMP in the model was confirmed.

Statistical analysis
The data were expressed as the "mean±standard deviation". Error bars in all figures are the standard deviations obtained from at least five independent measurements unless otherwise stated. All the data were analysed and performed by Origin Software. Supplementary Tables. Table S1. Fitting parameters of the oscillating signal which are related with the data shown in Fig.  2e.  Table S2. Material properties based on the different mass ratios of PDMS and Ecoflex.