A Mechanoluminescent ZnS:Cu/PDMS and Biocompatible Piezoelectric Silk Fibroin/PDMS Hybrid Sensor for Self‐Powered Sensing and Artificial Intelligence Control

Mechanoluminescence (ML) is luminescence induced due to mechanical stress, providing intuitive responses to strain‐related events. Piezoelectricity is the conversion of mechanical strain into electrical signals, offering a quantitative measurement of force/deformation. Combining ML and piezoelectricity within a single device provides a comprehensive understanding of mechanical events, providing qualitative and quantitative information about strain‐related phenomena. A ZnS:Cu/polydimethylsiloxane (PDMS) composite and a biocompatible silk fibroin/PDMS composite are prepared to generate ML and electrical signals, respectively. An innovative method for obtaining powder from silk fabrics is employed. The microstructure and composition of silk fibroin powder are also examined via X‐ray diffraction and Fourier Transform Infrared (FTIR) spectroscopy. Mechanical stimuli such as pressure, stretching, twisting, bending, vibration, and rubbing are applied to the device to demonstrate optical and electrical responses. Under pressure, a voltage of 3.82 V and an output current of 201.6 nA are generated at a force of 1 N. Furthermore, a handwritten test is conducted to qualitatively visualize letters based on ML effects and explore the feasibility of using artificial intelligence to classify voltage signals generated during writing into their corresponding letters. This biocompatible, dual‐modal self‐powered sensor demonstrates broad applicability in wearable technology, biomechanics, human–machine interaction, security, and energy harvesting.


Introduction
As a type of luminescence induced by mechanical stress, mechanoluminescence (ML) is a rapid visual representation of skin that can distinguish finger movements by harvesting triboelectric energy and ML. [11]Meanwhile, He et al. introduced a dual-mode strain sensor for monitoring human signals, utilizing triboelectric output and ML. [12]Additionally, Hajra et al. utilized PDMS/ZnS:Cu composites for triboelectric energy harvesting and ML intensity detection, demonstrating the versatility of ML across various applications. [13]It is important to note that ZnS:Cu has been considered for developing flexible and stretchable electroluminescent devices powered by triboelectric nanogenerator. [14,15]][18] Notably, compared to triboelectric devices, piezoelectric devices exhibit several advantages, such as higher resilience to environmental conditions, including humidity and temperature; a wider dynamic range; and low susceptibility to mechanical wear and degradation, etc. Owing to the numerous advantages associated with piezoelectric devices, the main aim of this study is the integration of the piezoelectric mode with ML to construct a hybrid sensor capable of photon emission as well as voltage in response to applied mechanical stimuli.
In this study, the main aim was to integrate a biocompatible silk-based piezoelectric material into a ML device.The integrated piezoelectric sensor within the ML device must effectively respond to different mechanical actions, aligning with the capabilities of the ZnS:Cu/PDMS composite across stretching, pressure, bending, rubbing, and vibration.Previous studies reported silkbased piezoelectric devices using a silk fiber or thin-film coatings prepared using a silk solution. [31,32]However, thin-film coatings on elastomers are brittle and lack adaptability to different types of mechanical deformation.Additionally, integrating silk fibers with an elastomer reduces the stretchability of the device.To address these limitations, in this study, a new method to synthesize silk fibroin powder (SFP) was reported, thereby simplifying the development of an elastomer composite by directly blending an elastomer and SFP.In addition, the microstructure and composition of SFP were characterized by X-ray diffraction and FTIR analyses.The piezoelectric/ML hybrid device structure was designed as an ML layer-piezoelectric layer-ML layer.The optimal SFP content (wt%) of PDMS was investigated for generating the maximum voltage output.Various mechanical stimuli, including pressing, bending, wind-driven vibration, and writing (rubbing), were employed to assess piezoelectric properties and ML output.Furthermore, the feasibility of employing artificial intelligence to categorize the voltage signals produced during writing into their respective letters was explored.This biocompatible dualmodal sensor demonstrates various potential applications, such as in wearable technology, [14] electronic skin, [15] biomechanics, human-machine interaction, security, and energy harvesting.

SFP Preparation
Silk fabrics were prepared from Bombyx mori silk fibers (Sang-ju Myeongju, Korea).First, these fabrics were subjected to degumming by immersion in 0.01 n Na 2 CO 3 and boiling for 30 min.After degumming, the fabrics were rinsed with deionized water until the fabric appeared clear.Subsequently, the silk fabrics were dried at 80 °C for 30 min.Next, the dried silk fabrics were dissolved in a solution containing H 2 O:C 2 H 5 OH:CaCl 2 in a molar ratio of 8:2:1 at 75 °C for 24 h.The resulting silk fibroin solution was subjected to dialysis for 12 h using dialysis cassettes (Slide-A-Lyzer, Pierce, 3.5K molecular weight cut-off, USA), with the regular replacement of deionized water.Finally, the dialyzed silk fibroin solution was dried at 80 °C in an air oven for 4 days.The dried silk fibroin film was subsequently placed in a ball mill jar and subjected to ball milling using zirconia balls for 1 week.Figure S1a (Supporting Information) shows the entire fabrication process.

Fabrication of a Piezoelectric/ML Hybrid Device
The piezoelectric/ML hybrid device comprised five layers from top to bottom: ML, electrode, piezoelectric, electrode, and ML.To fabricate this device, SFP was first mixed with PDMS (PDMS; Sylgard, 184 Silicone Elastomer) at 450 rpm for 5 min using a planetary mixer.Subsequently, the SFP/PDMS composite was degassed using a vacuum for 30 min.The SFP/PDMS composite was subsequently poured onto a glass substrate and spread to a thickness of 1 mm by using the doctor blade method.The film was dried at 80 °C in an air oven for 1 h.The dried film was dipped into a solution of silver nanowires (Ag NW: length: 20 μm, diameter: 20 nm, NIC, Korea) and isopropyl alcohol using a dip coater.Subsequently, the film was dried at 80 °C in an air oven for 1 min.The dip-coating process was repeated for three cycles, which ensured a surface resistance of 80 ohm m −2 .As a result, two electrodes on two sides of the piezoelectric layer were deposited by the dip-coating method.Subsequently, the dried sample was cut into a size of 4 × 4 cm.Finally, to measure the voltage, two copper wires (250 μm, Nilaco Co., Japan) were attached on the bottom and top Ag NW electrodes using a silver paste.To fabricate two ML layers, a mixture of ZnS:Cu (LONCO, Hong Kong) and PDMS was initially prepared using a planetary mixer at 450 rpm for 5 min.Subsequently, the mixture was subjected to degassing under vacuum for 30 min.Next, the ZnS:Cu/PDMS composite was poured onto one of the surfaces of a Ag NW-coated piezoelectric film, and the doctor blade method was employed to achieve a film thickness of 250 μm.The resulting sample was subsequently transferred to a vacuum oven and dried at 80 °C for 1 h.Similarly, the second ML layer was fabricated on the remaining surface.Figure 1 shows a brief illustration of the overall fabrication process.Figure S1b (Supporting Information) shows the detailed fabrication process.

Material Characterization
The size, microstructure, and composition of the samples were characterized by field-emission scanning electron microscopy (FE-SEM; SU-8220, Hitachi, Japan).The crystalline structures of the SFP, silk fabric, and ZnS:Cu were analyzed by X-ray diffraction (XRD; X'Pert Pro MRD, PANalytical).The structures of SFP and the fabric were analyzed by Raman spectroscopy (inVia Reflex, Renishaw PLC, UK) and FTIR spectroscopy (PerkinElmer, Frontier, USA).
The electrical output of the device was systematically characterized by the application of pressure cycles using a pushing machine (PP machine system, SnM, Korea).The output voltage and current were measured using a digital multimeter (DMM7510, Keithley, USA).On the other hand, the optical signal of the device was systematically characterized by the application of pres-sure cycles using a universal testing machine (E3000, INSTRON, USA).The optical signal was measured using a photomultiplier tube (PMT).
Additional mechanical stimuli such as bending, stretching, twisting, vibration, and rubbing (handwritten test) were simulated by human hands.These experiments were conducted to reveal human-device interaction.

Results and Discussion
In Figure S2 (Supporting Information), the SEM images exhibits the effect of ball milling time on the size of SFP.It is evident that the SFP subjected to 1 day of ball milling exhibits significantly larger particles, with an average size of 18.77 μm (Figure S2a, Supporting Information).Conversely, in the case of the SFP ball milled for 3 days, the particles appeared relatively finer, with an average particle size of 12.65 μm (Figure S2b, Supporting Information).Figure 2a shows the morphology of the SFP particles after the 7-day milling process.The extended ball-milling duration led to an average SFP particle size of 3.8 μm (Figure 2b).Notably, irregular SFP particles were observed, which were attributed to the prolonged ball-milling duration and the significant shear stress involved.In contrast, commercial ZnS:Cu particles exhibited a regular morphology (Figure S3a, Supporting  Figure 3a shows the XRD patterns of the SFP, silk fabric, and ZnS:Cu (inset image).The XRD patterns of SFP and the silk fabric revealed distinct crystalline structures observed at 9.5°and 21°. [33,34]Additionally, a peak corresponding to silk I, a known metastable structure, was clearly observed at 24.4°.Notably, the 9.5°peak identified as silk II, was observed in both materials.However, in the case of silk fabric, this peak exhibited a negligible intensity.This difference in intensity suggests that SFP likely possesses a more pronounced silk II crystalline structure compared to silk fabric.Since the silk II structure is associated with enhanced piezoelectric properties, the higher intensity of this peak in SFP indicates that it may indeed exhibit better piezoelectric characteristics than silk fabric. [31,32]This inference is based on the understanding that a more prominent silk II structure typically correlates with improved piezoelectric performance.Therefore, the XRD analysis supports the hypothesis that SFP is expected to demonstrate superior piezoelectric properties compared to silk fabric.As shown in Figure S4a (Supporting Information), the XRD spectra of ZnS:Cu particles were consistent with those of ZnS-sphalerite (PDF #5-0566) and ZnS-Wurtzite-2H(PDF #36-1450), indicative of a combined cubic and hexagonal phase structure. [35]Notably, Ma et al. previously reported that ZnS exhibited a biphasic structure, contributing to an improved ML performance. [36,37]gure 3b shows the Raman spectra, highlighting key observations.The amide I bands, attributed to the C═O stretching vibration, were observed at 1668 cm −1 for silk fabric and at 1677 cm −1 for SFP.Similarly, the amide III bands, attributed to N-H bending ((N-H) vibration), were observed at 1236 cm −1 for silk fabric and at 1260 cm −1 for SFP.These results were consistent with those reported by Monti et al., where peaks observed at 1236 and 1260 cm −1 for the amide III bands with a silk I arrangement were characterized by random coils and -sheets, respectively. [38]Notably, SFP exhibited a noticeable peak shift, with the peaks corresponding to the amide bands and -sheet slightly shifting to the right.To investigate the origin of the peak shift, we analyzed the Raman spectra of the dissolved fabric solution and the resulting SFP after 1, 3, and 7 days of ball milling.As depicted in Figure S4b (Supporting Information), the dissolution of the silk fabric led to a rightward shift in the Raman peaks corresponding to the Amide III and Amide I bands.Subsequently, SFP obtained after 1 day of milling exhibited a slight left shift compared to the dissolved fabric.However, SFP from 3 and 7 days of milling showed a continuous right shift compared to the dissolved fabric.This suggests that both the dissolution and ball milling processes contributed to the shift in the Raman peaks.The Raman peak shift can occur due to various reasons, such as changes in the bonding length, doping, and crystal size. [39]igure 3c shows the FTIR transmission spectra of silk fabrics and SFP, highlighting key bands such as amide I, amide II, and amide III.These bands were observed at 1619, 1515, and 1229 cm −1 , respectively, corresponding to C═O stretching, N-H bending/C-N stretching, and C-H stretching/N-H in-plane bending.The amide A band at 3300 cm −1 corresponded to the O-H stretching vibrations.Interestingly, the transmittance of the amide V bands for SFP was less than that of the silk fabric, attributed to the increased crystallinity as reported by Kim et al. [31] Figure 3d shows a magnified view of the FTIR spectrum in the wavenumber range from 2000 to 800 cm −1 .The FTIR spectrum revealed peaks at 1645 and 1619 cm −1 corresponding to the amide I bands, representing an amorphous region (composed of a random coil and -helix) and a crystallization region, respectively.Similarly, the peaks at 1265 and 1229 cm −1 corresponding to the amide III bands revealed crystalline and amorphous regions.The crystallinity index (CI) was calculated using the absorbance values of the amide I and III peaks at each wavenumber, as shown in Equation 1. [31,40] CI for amide (%) A 1619 cm −1 ∕ ( The CI values of the silk fabric and SFP were determined to be 55.23% for amide I and 60.03% for amide I, respectively.These results suggested that the dissolution and ball-milling process increased crystallinity, transforming amorphous structures into crystalline ones and increasing the relative abundance of -sheets.
Owing to their versatile responses to various mechanical stimuli, flexible and stretchable piezoelectric devices are revolutionizing the collection and utilization of data, offering a wide range of applications in healthcare, engineering, robotics, and beyond.Traditional piezoelectric devices primarily rely on generating voltage signals in response to external mechanical stimuli, while the introduction of mechanoluminescent properties facilitates an entirely new dimension to their functionality.Therefore, demonstrating the dual-sensing capability of the proposed hybrid sensors under various mechanical stimuli, including pressure, stretching, bending, twisting, vibration, and rubbing, is intriguing.This approach enhances the versatility of the sensor and its potential applications across different domains.
In this study, to use SFP as a piezoelectric material for sensing, determining the optimal composition of the SFP-PDMS composite is imperative.Therefore, four hybrid sensors were constructed by varying the weight percentages of SFP (15, 25, 35, and 45 wt%, respectively), where SFP was taken after 7 days of milling.The SFP obtained from 7 days of milling was chosen for the hybrid device due to its superior piezoelectric responses compared to the SFP obtained from 1 and 3 days of milling (Figure S5, Supporting Information).This increase in piezoelectric response with longer milling time can be attributed to the smaller particle sizes, which enhance the material's surface area.This enhancement facilitates stronger interactions with mechanical stimuli, resulting in a more robust piezoelectric response.Additionally, the concentration of ZnS:Cu wt% in PDMS remained consistent at 40 wt% for the ML layer of the hybrid sensor.This concentration exhibited the highest ML intensity under compressive load cycles of 50 N, as illustrated in Figure S6a (Supporting Information).The ML emission initially increased with the rising wt% of ZnS:Cu up to 40 wt%, but began to decrease after reaching 50 wt% of ZnS:Cu.This indicates that achieving maximum ML emission involves a trade-off between the ZnS:Cu content and the stiffness of the composite. [42,43]he correlation between the SFP content and output performance was systematically examined by the application of pressure cycles in the range from 1 to 50 N.Notably, each load was applied for three consecutive cycles.Figure 4a shows the schematic The results revealed a notable relationship between the weight percentage (wt%) of SFP and the magnitudes of voltage and output current.Specifically, with an increase in the SFP content, the voltage and output current values increased significantly, reaching their peak at an SFP content of 35 wt%.Beyond this threshold, however, the magnitudes of voltage and output current decreased.This phenomenon can be attributed to the results reported by Lee et al., who suggested that a piezoelectric composite exhibited a high dielectric constant when it contained an excess amount of the piezoelectric material within a confined volume. [41]This excess material can adversely affect the electromechanical coupling effect, consequently reducing the output performance.Subsequent investigations were conducted using the device with 35 wt% SFP.It's noteworthy to highlight that in our previous study, silk fabric yielded an output of 6.62 V and 0.65 μA under a 60 N load. [31]Interestingly, a comparable output response was achieved with 35 wt% SFP at a 50 N load.This suggests that the enhanced piezoelectric performance observed in SFP may be attributed to its more prominent silk II crystalline structure compared to silk fabric, as evidenced by the XRD patterns shown in Figure 3a.The durability of the device was assessed by subjecting it to 10 000 pressure cycles of 120 N. Figure 4d plots the voltage response of the device throughout this durability test, revealing the capacity of the device to endure prolonged mechanical stress and maintain its functionality.Notably, the average output voltage during this test was determined to be ≈15 V.
Figure 4a shows the schematic of the ML responses of a device measured from the bottom ML layer using a PMT.device emitted photons from both sides of the device in response to mechanical actions.The ML layer on both sides of the piezoelectric layer not only enabled the emissions of photons from both sides but also ensured the presence of flexible and stretchable electrodes between the ML and piezoelectric layer.After the doctor blade technique, the liquid PDMS within the ML layer was encapsulated around the Ag NW networks on the piezoelectric layer surface, ensuring a secure hold after drying.The effective function of the device was dependent on the presence of flexible and stretchable electrodes for demanding mechanical environments.From the results shown in Figure 4e, at 1 N, the PMT did not detect ML emissions, possibly attributed to the minimal emission below the sensitivity threshold of the PMT.However, starting from 10 N, ML emissions were consistently detected, revealing two luminescent peaks for each loading cycle.Recent studies suggested that this unique ML phenomenon was attributed to the triboelectric effect between the ZnS:Cu phosphor and the PDMS matrix within the device. [43]When the force was applied and then released, these materials rubbed against each other due to the contact and separation caused by loading and unloading. [43]his friction resulted in the ZnS:Cu becoming positively charged and the PDMS becoming negatively charged, leading to the emission of light via triboelectricity during loading and releasing cycles.It is noteworthy that beyond an applied load of 30 N, the intensity of the second peak (during unloading) exceeded that of the first peak (during loading), although it remained similar below 20 N.This asymmetrical emission could be attributed to the viscoelastic properties of PDMS.At low loads, elastic deformation prevails over viscous deformation, resulting in similar emissions during loading and unloading.However, as the load increases, viscous deformation becomes dominant over elastic deformation, leading to uneven peak intensities.Similar uneven intensities were also observed in the previous reports. [9,43]The magnitude of the applied force determined the surface area of contact between ZnS:Cu and the PDMS matrix.Consequently, when a higher force was applied, it resulted in more extensive contact, leading to a more pronounced triboelectric effect.This relationship was clearly demonstrated in Figure 4e, where ML consistently increased in direct proportion to the magnitude of the applied force.It is worth highlighting that conventional ML materials typically require preirradiation with UV light to initiate the ML phenomenon. [18]However, ML from ZnS:Cu is entirely mechanically driven, offering a distinct advantage in practical applications since there's no need for harmful UV light.
From the results of the systematic study of the pressure test, the proposed hybrid device exhibited a dual-sensing capability.The device maintained its functionality under prolonged mechanical stress, suggesting that SFP can exhibit long-life performance.Moreover, the electrode layers also endured prolonged mechanical stress.The durability test of ML already has been investigated previously, which has been reported to be greater than 10 5 cycles [44] and it has not been conducted herein.
To demonstrate the response of the device under more different mechanical stimuli such as bending, twisting, stretching, vibration, and rubbing, the experiment was conducted in a way to demonstrate human-machine interaction.Therefore, different mechanical actions on the device were simulated by human hands (Figure 5). Figure 5a-d shows the voltage and ML responses under bending, twisting, stretching, and wind flow, respectively.From the uniaxial tension test shown in Figure S6b (Supporting Information), it was found that the device can be stretched up to 82% strain with a tension strength of 0.66 MPa.The vibration on the device in Figure 5d was stimulated using an air gun (150 L min −1 ).Both electric signal and ML were detected (Figure S8, Supporting Information).The monitoring of wind flow demonstrates potential applications in medical devices such as ventilators, respiratory therapy equipment, cleanroom environments, and laboratory hoods. [13]The precise monitoring of wind flow is crucial for upholding the safety and optimal functioning of these systems.Furthermore, an additional experiment was conducted by blowing air from the mouth to simulate lower flow rates.While electrical signals during this process were observed, ML emission was not detected, as illustrated in Figure S9 (Supporting Information).This suggests that the deformation induced by the lower air flow was not sufficient to trigger ML emission.
The response of the device to rubbing was investigated by writing letters on the device's surface.In recent years, several researchers reported the potential applications of MLgenerated handwriting as an electronic signature system and optical anticounterfeiting. [10]Furthermore, Zhou et al. introduced machine learning to digitalize ML-generated handwriting. [16]Machine learning was also introduced to translate complex bite ML patterns into specific data inputs for high-accuracy remote control and operation of various electronic devices. [45,46]In this study, ML-generated handwriting as well as the voltage response from the piezoelectric layer were demonstrated, and machine learning was applied to classify the written letters solely from their generated voltage signals.
Each of the seven letters ("p," "i," "e," "z," "o," "m," and "l," respectively) was written on the device surface using a pen. Figure 6a shows the ML images where handwritten letters were clearly visualized due to the emission of ML originating from the friction between the device surface and pen.Notably, the entire action of writing a letter was recorded by increasing the exposure time of the camera sensor, thereby enabling the direct visualization of the ML pattern corresponding to the handwritten letter.Meanwhile, a voltage signal was generated while writing due to the pressure applied by the pen. Figure 6b shows the corresponding voltage outputs of each type of written letter, and compares the three types of output.The voltage signals demonstrated that each letter exhibited a unique output response.Moreover, outputs while writing the same letter revealed some degree of variations.
In this study, the architecture of the machine learning algorithm was built using a convolutional neural network (CNN), which is schematically illustrated in Figure 6c.CNNs typically comprise an input layer, a feature extraction layer, a fully connected layer, and an output layer.Herein, the input shape was considered to have a size of 200 × 1; therefore, all of the original signals were resized equivalent to the input shape.Each convolutional layer of the feature extraction layer comprised a kernel of size seven, a stride of unit one, and an activation function of ReLu.A total of four fully connected layers were used with the ReLu activation Moreover, the dropout layers were included to minimize overfitting with a 20% dropout rate.The output layer contained seven neurons for seven classes and was assigned to be the softmax activation function.The models were trained using the Adams optimizer to determine the global minima of the categorical cross-entropy loss function with a given learning rate of 0.0001.The model was run for 200 epoch with a batch size of four.
The dataset required for training and testing was generated through repetitive writing.In particular, each letter was repetitively written 100 times, where 80 signals were allocated for training and 20 signals were allocated for testing.In fact, the model was trained and tested from the voltage signals generated while handwriting of letters written by one person.While this may limit the generalizability of the model to handwriting from other individuals, our aim was to demonstrate the feasibility of our approach using a controlled dataset.Figure 6d shows the test results displayed in a confusion matrix plot.The test accuracy of the model was ≈98.6%, indicating the model's high performance in accurately classifying handwritten letters based on the voltage signals (Figure 6d).Achieving a high accuracy score indicated that the model was robust and could handle various handwritten styles, variations in pressure, and other factors that might affect the voltage signals generated during writing.Such a high accuracy is crucial for practical applications that rely on the recognition of handwritten letters, such as electronic signature systems, optical anticounterfeiting, document verification, and humanmachine interaction.It ensures that the technology can be effectively deployed in real-world scenarios.The framework described above can be expanded to create a more generalized machine learning model by incorporating handwritten letters from a diverse set of individuals.Since CNN operates on supervised learning principles, the test dataset should align closely with the range of the training dataset.Therefore, to improve the model's generalization, it would require a large number of handwritten voltage signals from various individuals during the training phase.Future research could investigate the model's performance using data from multiple individuals to evaluate its broader applicability and robustness across different handwriting styles.
Previous studies have explored integrating the triboelectric nanogenerator mode with ML devices to generate self-powered for their triboelectric nanogenerator device, whereas Zhang et al. reported 2 N V −1 . [11,16]For the former device, a rough surface was created to enhance sensitivity.Despite using piezoelectric material to generate electric signals in our device, the sensitivity was measured at 3 V N −1 , within an acceptable range for tactile sensing.Additionally, Zhang et al. reported a dynamic range of 1-12 N for their device, while Zhou et al. reported 1-25 N. [11,16] In comparison, our device exhibited a larger dynamic range of 1-125 N. Hajra et al. reported a durability of 3000 cycles for their device, similar to the durability reported by Zhou et al. and Zhang et al., which was over 2000 cycles. [11,13,16]In our present work, we demonstrated durability exceeding 10 000 cycles.It is noteworthy that the ML layer and piezoelectric layer in our device are chemically bonded, as PDMS was used in all layers, ensuring longer durability.Neither the previous reports nor our current work have verified performance under various factors like humidity and surface contamination to determine environmental stability.However, the triboelectric effect relies on surface charge transfer, which can be hindered by these environmental factors.In contrast, piezoelectric sensors rely more on internal material properties and less on surface interactions, making them less susceptible to environmental changes.Therefore, it is justifiable to conclude that our device could be more stable against environmental factors.

Conclusion
A self-powered piezoelectric/ML hybrid device with dual-output capabilities is introduced, exhibiting electrical and optical responses to various mechanical stimuli.The integration of ML and piezoelectricity in the device provides a comprehensive understanding of mechanical events, encompassing qualitative and quantitative aspects of strain-related phenomena.A novel approach to synthesizing powder from silk fabrics is presented herein, simplifying the development of a biocompatible piezoelectric elastomer composite by simply mixing the elastomer and SFP.Furthermore, the device is fabricated using the doctor blade method, a simple and cost-effective technique.Moreover, the device exhibits exceptional durability, withstanding prolonged mechanical stress; hence, it is extremely suitable for longterm applications.Furthermore, the device is adaptable to various mechanical stimuli, including bending, twisting, stretching, and even vibration, demonstrating its potential for diverse real-world scenarios.The high accuracy (98.6%) achieved in letter recognition using a CNN, which is based on the electrical signals generated during writing, underscores its practicality and robust performance.In essence, the dual-sensing piezoelectric/ML hybrid device, with its SFP component, demonstrates promise for a wide range of applications in wearable technology, biomechanics, human-machine interaction, security, and energy harvesting.Its versatility, durability, and biocompatibility make it a valuable asset in the realm of multifunctional sensing devices.

Figure 1 .
Figure 1.Schematic of the fabrication process of a piezoelectric/ML hybrid device.

Figure 2 .
Figure 2. Morphology and dispersion of SFP and ZnS:Cu in PDMS.a) FE-SEM images of SFP, (b) Particle size distribution of SFP and ZnS:Cu, (c) Dispersion of 35 wt% SFP in PDMS, (d) Dispersion of 40 wt% ZnS:Cu in PDMS.
Figure 2c,d shows the dispersion of 35 wt% SFP and 40 wt% ZnS:Cu in PDMS, respectively.Figure S3b-d (Supporting Information) shows the surface images of the 15, 25, and 45 wt% SFP/PDMS composite, respectively.All of these figures confirmed the homogeneous dispersion of ZnS:Cu and SFP within the device.

Figure 3 .
Figure 3. Microstructure and composition of SFP and ZnS:Cu: a) XRD patterns of the silk fabric, SFP, and ZnS:Cu, b) Raman spectra of the silk fabric and SFP, c) FTIR spectra of the silk fabric and SFP, and d) Magnified FTIR spectra of silk fabric and SPF in the wavenumber range of 2000-800 cm −1 .

Figure 4 .
Figure 4. Output performance of the piezoelectric/ML hybrid device with 35 wt% SFP.a) Schematic of the dual-sensing capability of the proposed hybrid sensor, b) output voltage, c) output current, d) durability test at 120 N during 10 000 cycles, and (e) ML intensity.

Figure 5 .
Figure 5. Dual response of the piezoelectric/ML hybrid device under different mechanical stimuli.a) bending, b) twisting, c) stretching, and d) wind flow.

Figure 6 .
Figure 6.ML and voltage responses of the device while writing.a) Visualization of handwritten letters via ML effects.b) Electrical responses of the device to handwritten letters.c) The architecture of the convolutional neural network designed to train electrical responses of the device to handwritten letters shown in panel (b).d) Model performance displayed as a confusion matrix.