Remarkable piezoelectric properties in thin �lms of cellulose nano�bers after electrochemical poling

Internet-of-everything (IoE) is de�ned as networked connections of things, people, data and processes. IoE nodes, preferably shaped as printed �exible systems, serves as the frontier outpost of the Internet and comprises devices to record and regulate states and functions. To power distributed IoE nodes in an ecofriendly manner, technology to scavenge energy from ambience and self-powered devices are developed. For this, piezoelectricity is regarded as a key-property, however current technology typically based on polyvinylidene di�uoride (PVDF) co-polymers, are expensive and produced via toxic protocols. We report piezoelectric characteristics of electrochemically poled cellulose nano�ber (CNFs) thin �lms processed from water dispersions. Poling these �lms at humid conditions cause breaking and reorientation of CNF segments, which results in enhanced crystal alignment rendering the resulting material piezoelectric. Generators based on poled CNF show similar piezoelectric voltage and coe�cient, here measured to d 33 = 46 pm/V, as for devices including PVDF copolymer layers of similar thickness. Our �ndings promise for low cost and printable ecofriendly piezoelectric-powered IoE nodes.


Introduction
Internet-of-Everything (IoE) [1] is currently one of the strongest tech-trends with many great applications already in operation and a vast array of opportunities forecasted.To settle this technology as an ecofriendly technology, we need to make sure that all included materials are derived from sustainable origins and that they are processed and recycled under protocols with a minimal impact on the environment [2].To power and to integrate self-powered devices into IoE nodes, various properties and functions are being explored to enable heavily distributed and remote autarchic nodes within future Fog Computing systems [3].In this respect, piezoelectric materials have been widely applied in energy harvesting and signal transducing applications.
Piezoelectric materials couple mechanical stress, such as displacement, pressure, vibration and ows to electrical polarization [4,5], which have been used for powering portable electronic devices and wireless sensors [6,7] as well as in actuators and acoustic devices [8].The strength of the mechano-to-electric coupling described as the piezoelectric coe cient is de ned as the ratio between the generated charge and the applied mechanical force [9].In recent decades, piezoelectric polymers have become the focus because of the possibility to use atmospheric synthetic routes as well as the ability to make exible thin lms by solution processing, such as printing technologies, despite their typically lower piezoelectric coe cient compared to inorganics [10].The best-performing piezoelectric polymers are uoropolymers based on polyvinylidene di uoride (PVDF) and its copolymer P(VDF-TrFE) (polyvinylidene uoride tri uoroethylene).Recently, a high piezoelectric coe cient of 52 p CN −1 was reported for P(VDF-TrFE) composited with Mxene nanosheets [11].With the awareness of the human footprint on the environment, IoE applications imply massive number of small devices released in society and nature.Here, the uorine chemistry and the release of HF toxic fume upon thermal degradation of the PVDF material are issues for chemical synthesis and recyclability [12].Hence, a new generation of green piezoelectric materials, comprising features such as low cost, water solution processibility, and an overall environmental friendliness is in urgent need.So far, the highest piezoelectric coe cient found in biomaterials reaches 38 pC N −1 at nanoscale characterized by piezoresponse force microscopy (PFM) [13].Although this high value was not correlated to piezoelectric devices of practically large area, it showed the potential of biomaterials for piezoelectricity.
As the most abundant and green material on earth, cellulose-based materials have triggered a tremendous research attention over the last decades.The piezoelectricity in wood was discovered already in 1950 [14], followed by explanation that the origin of shear piezoelectricity is correlated to the reorientation of dipole moment from hydroxyl groups in the crystalline cellulose [15].The intra-and inter molecular hydrogen bonds along the side of the cellulose chain introduce a non-centrosymmetric crystal structure in cellulose [16,17].The piezoelectric coe cients of those bulk wood-based materials are very low, which hinders their practical applications [18].More recent, the topic went through a phase of renaissance thanks to the exploration of nanocelluloses that opened up new possibilities to make cellulose material systems with unique properties.Cellulose nanocrystals (CNCs) of diameter between 5-20 nm and a length of about 100-500 nm are the smallest crystalline features extracted from wood via mechanical and chemical treatments.Cellulose nano bers (CNFs) can be several micrometers long with amorphous and crystalline domains (mentioned as CNC) arranged alternatively.
For a single cellulose crystal, apart from the shear piezoelectric constants, longitudinal and transverse piezoelectric constants are also present [19][20][21] because of the permanent dipole moment along their axis [22].The longitudinal piezoelectric coe cient d 33 enables the most promising applications because maximum deformation occurs when the output voltage and the applied mechanical strain are in the same direction.Orienting randomly arranged CNC to aligned structure in a particular direction could increase the piezoelectric constant.However, most of the reported methods only succeeded to align CNFs and CNCs parallel to the substrate [20,23,24]; thus not leading to any virtual enhancement of d 33 .The di culty to orient the nanocellulose homogeneously across large area and the extreme fragility of the CNCs lms prevent cellulose to be implemented in e cient and high-performing piezoelectric devices [21].The practical piezoelectric devices composed of nanocellulose only exhibit a low piezoelectric constant (5 pC N −1 ), that is 4-5 times smaller than ones based on PVDFs [19].Till now there is no research showing that pure cellulose can exhibit competitive piezoelectric performance compared to PVDF derivatives in practical devices.
In this work, we report that electrochemical poling greatly enhances the piezoelectric property of CNF thin lms, which can be implemented in macroscopic devices (area of about 1 cm 2 ) to exhibit competitive performance with P(VDF-TrFE)-based piezoelectric devices of similar thickness.Different from the electric poling process used to polarize typical ferroelectric materials to improve piezoelectricity, the proposed electrochemical poling mechanism involves absorbed water from air, breaking and reorientation of the CNF under electrochemical reactions.Atomic force microscopy (AFM) was here used to reveal the role of humidity and visualizes the changes of CNF thin lm morphology after electrochemical poling.The mechanism is supported by impedance spectroscopy and grazing-incidence small-angle X-ray scattering (GISAXS) characterization of the CNF lms before and after poling.We also con rmed the improved d 33 piezoelectric coe cient of CNF thin lms and measures values reaching up to 46 pm V −1 using PFM, which is comparable to P(VDF-TrFE) (45 pm V −1 ).Our work demonstrates the possibility of nanocellulose to compete with PVDF copolymers processed from solution.Our ndings pave the way for ecofriendly novel all-printed energy technology and sensors for IoE application in general and may also nd its way into some niche-application for large area energy harvesting systems.

Piezoelectric voltage of CNF thin lms
The CNF thin lms were prepared on doped silicon wafers by drop-casting a xed amount of water dispersion of CNF with 0.5 wt% solid content (Fig. 1a).The conductive silicon wafer and the top Au electrode were connected to form a close circuit for both electric poling and piezoelectric characterizations (Fig. 1b).The photograph and dimension of a result device is shown in the inset gure in Fig. 1c.The CNF lm was poled with gradually increasing voltage from 5 V to 110 V.After each voltage step, the piezoelectric response of the same lm was recorded while applying the same force.Fig. 1c presents the piezoelectric performance for a Si/CNF/Au device at ambient humidity (RH=40%).Those traces correspond to the open circuit piezoelectric output voltage upon a periodic applied force (24 N) at frequency of 3 Hz.There is no obvious piezoelectric response above the non-periodical noise level (<10 mV) for the device before poling.For poling voltages below 30 V, a periodic output voltage following the 3Hz input force is observed with an amplitude of around 30 mV.Above 30 V, the piezoelectric voltage increases and reaches a maximum of 100 mV at a poling voltage of 70 V.For poling voltages higher than 70 V, there is a degradation of the performance with a slight decrease in output voltage.We also tested the piezoelectric response of the same device 30 minutes after poling (with the same force F=24 N), see Fig. 1c, there is no obvious degradation of the output voltage.The asymmetry between positive and negative voltage is most likely due to the measurement frequency limit as found in other piezoelectric materials [5,10].The same device was characterized again after two weeks and the piezoelectric response maintained similar (Supplementary Fig. 1).These results demonstrate that the piezoelectric performance of the CNF thin lms can be greatly improved by electric poling at room humidity and the induced poling effect remains.
The piezoelectric response of CNF lms are found to depend on the humidity of the atmosphere during poling process.In Fig. 1d, CNF thin lms poled at four different relative humidity levels (empty columns) are compared with that of non-poled ones (dashed columns) using the same force.The poling of CNF lms was performed in a climate chamber with xed humidity after keeping the samples inside the chamber for 2 hours, and the piezoelectric output voltage was measured in ambient humidity afterwards.
Here, we select the samples with the average thickness close to 1μm (±0.1 μm), and then poled them with an increasing voltage between 5 to 110V, and the maximun output votlage is presented in Fig. 1d (corrisponding poling voltage is presented in Supplementary Table 1).The piezoelectric output voltage is found to be the highest for samples poled in room humidity (40% RH) and smallest values were found from samples poled at high humidity.This might be due to more disociated ions in the samples poled at high humidity that screens the piezoelectric polarization [25].To prove this, we dried the CNF lms poled at 70% RH and then tracked the piezoelectric reponse.As shown in Fig. 1e, the reponse of the same sample increased from 52 mV to 80 mV after drying in low humidity (10% RH) for 30 min.
Electrochemical poling: poling with electrolysis Electric poling has been successfully applied to enhance the alignment of the electric dipoles in ferroelectric materials, thus improving their piezoelectric performance [26].However, we did not observe typical ferroelectric poling current as typically found for P(VDF-TrFE) (Supplementary Fig. 2) when sweeping the CNF lms with increasing voltage between -110V to 110V (Supplementary Fig. 3).Hence, the CNF lm does not exhibit any ferroelectric properties.It should be mentioned that all ferroelectric materials are piezoelectric but the reverse does not apply.Materials having non-centrosymmetric structure could possess piezoelectricity without being ferroelectric [27].The improved piezoelectric output voltage in poled CNF thin lms is likely due to a different mechanism from the ferroelectric PVDFs.As cellulose is known to be hygroscopic with a 10% to 14% water content in the atmospheres carrying a 40% to 50% RH, and as the applied poling voltage exceeds the electrochemical potential for electrolysis of water (1.2 -2V) [28,29], we assume that electrolysis of water occurs upon poling.In Fig. 2a and 2b, the poling current passing through the CNF lm when poled with a linearly increasing voltage at 10% RH and 40% RH were recorded, respectively.Constant scan rate of 5 V s -1 was used in all scans and the voltage range was increased for each, from 5 V to 110 V.The non-linear change in current upon applied voltage indicates that it is not capacitive in nature as it would have been expected for dielectric or insulating piezoelectric layers operated below the coercive eld.Moreover, the dramatically increasing current with the humidity level of the poling atmosphere suggests that the absorbed water in CNF lms undergoes an electrolysis reaction.The decreasing current at the same voltage after each successive scan at low humidity (10% RH) indicates the consumption of water in the lm step by step after each scan.The poling current for sample measured under ambient conditions (RH=40%) is high due to the large amount of absorbed water and reabsorption from atmosphere.
To understand the time scale of the water consumption in the CNF lm, a constant potential of 30 V is applied to the CNF thin lms for 30 s (Fig. 2c), the poling current passing through CNF lm during poling is recorded every 5s.For samples measured at all different humidity levels, we observed an exponential decay of the current that tend to stabilize after 25 s.The saturated current for samples poled at high humidity (RH=70%) is 85% of the original value, and the current for samples at low and room humidity decrease to 55% of their original values.Afterwards, the reabsorption of the water from the humid atmosphere is tracked by recording the current while applying a short pulse (t<1s, V=30V) every 50 s.As shown in Fig. 2d, the current increases with the exposure time indicating a regain of water.After 7.5 minutes, the current recovers almost to the same level as before poling for samples exposed at 40% and 70% RH, showing complete regaining of water and equilibrium with the atmosphere.On the contrary, in the dry condition (RH=10%), the normalized current remains at similar level as after poling for the measured time range, which indicates no obvious water regaining.The recapture of water in the CNF thin lms in room humidity after poling is also con rmed by FTIR spectroscopy when the OH vibrational stretching signal is tracked after 30 V poling for 30 s.As shown in Fig. 2e and 2f, the intensity of the OH stretching peak at 3340 cm -1 measured at the same location of the CNF lm increases with time, and saturates after 10 min.The water content probed by the FTIR signal increases with time due to the water absorption and the FTIR signal evolution follows the measured current evolution in Fig. 2d.The reabsorption of the water after poling, to a level similar found in the CNF lms before poling, proves that the change in the piezoelectric property is not due to the water loss from the lm.

Morphology changes upon electrochemical poling
We investigated the structural changes of CNF lm induced by electrochemical poling at different humidity levels by Atomic Force Microscopy (AFM).To access the surface of the CNF lm for probe tip scanning, a separate Au top electrode that can be detached from CNF after poling was designed.As shown in Fig. 3a and 3b, an Au thin lm (50 nm of Au on 5 nm of Cr) coated on a PDMS stamp is positioned on the CNF lm with slight pressure, which provides a tight and continuous contact with the CNF lm and enables effective electric poling (Fig. 3c).After poling at 110V for 30 s at xed humidity levels (10%, 40% or 70% RH, samples were kept in the humidity for 2 hours before poling), the top electrode was then removed and the surface of the lms were characterized with AFM (Fig. 3d).For the samples poled at low humidity (RH =10%), the morphology and phase images (Fig. 3g and 3h) resemble the non-poled lms (Fig. 3e and 3f) with only minor change.After poling at room humidity (40% RH), the bril morphology becomes blurred (Fig. 3i) compared to the non-poled lms.In the phase image of the poled CNF layer (Fig. 3j), nano-ring like structures with diameter around 10 nm to 15 nm can be observed.
The size of the rings is similar to the diameter of the nano brils; which indicates that those rings are the cross section of the brils pointing upwards.For lms poled at high humidity (70% RH), similar changes can be found with a blur morphology and nano-rings in the phase image (Fig. 3k and 3l).The obvious changes in morphology and phase images after poling at relatively high humidity (> 40%) indicate that the electrochemical poling current leads to the breaking of cellulose nano bers followed by the reorientation of small segments perpendicular to the surface.Note that the observed changes do not disappear several month after poling, however, they are not uniform on the whole CNF lms.As shown in Supplementary Fig. 4 and 5, different levels of changes can be found on the same sample after poling at relative humidity of 40% and 70%.This could be due to the variation of the CNF lms in thickness (up to hundreds of nanometers) resulting in a non-uniform electrochemical current during poling.Indeed, if a hard Au coated glass was used as top electrode, we could never see the morphology/phase changes because of rough CNF lms results a poor contact with the at and hard Au electrode.Moreover, those nano-rings and morphology changes are not found in control experiments.By simply heating those CNF lms at 250°C (before thermal degradation) and natural cool down, we could not observe such a change (Supplementary Fig. 6a and 6b).No change in morphology is found by applying the same strength of electric eld across a thin air gap as dielectric between the Au top electrode and the CNF lm (Supplementary Fig. 6c and 6d).Hence, we believe that the poling effect in the CNF lms is not due to any local Joule heating or electric eld, the key mechanism is likely related to "electrochemical poling", a phenomenon based on electrochemistry and electric eld in combination.
The morphology changes in electrochemical poled CNF thin lms can be evidenced by the dielectric characterization via impedance spectroscopy (details are discussed in Supplementary Note 2).The typical dynamic phenomena in CNF thin lms under a small AC current (from frequency range between 1×10 -1 -2×10 5 ) before poling, freshly after poling and 15 hours after poling are compared in Supplementary Fig. 7.The results show obvious changes in the impedance spectroscopy after poling and the change remains similar after 15 hours, which indicate that poling truly leads to an irreversible change of the dielectric properties.The Bode (Supplementary Fig. 8) and Nyquist plot (Supplementary Fig. 9) of the impedance spectroscopy of CNF thin lms poled in different humidity further con rmed the role of absorbed water in electrochemical poling and the induced morphology changes.

Piezoelectric device performance
The output voltage (V piezo ) of piezoelectric devices depends on the applied force (F), the longitudinal piezoelectric coe cient (d 33 ) and the capacitance (C) of the dielectric material according to: We notice that the capacitance at high frequencies (10 5 -10 6 Hz, RH=40%, Supplementary Fig. 8) that is critical for piezoelectric response [30], does not vary after poling at different voltages.Hence, we believe that the piezoelectric voltage measured upon poling is not due to the change in capacitance but mostly an effect of the d 33 coe cient of the lm.Note that the effect of humidity on the capacity value is expected as the introduction of water increases the average dielectric constant of the cellulose layer, however, this effect is vanishingly small (Supplementary Fig. 10).The control experiment of spin-coated CNF thin lm also con rms that the structure change of CNF is the key for the enhanced piezoelectric performance instead of water content (details can be found in Supplementary Note 3).
Piezoelectric Force Microscopy (PFM) has been widely used to con rm the reversed piezoelectric effect of a material, as well as to locally measure the piezoelectric coe cient d 33 (Supplementary Fig. 12a).A DC bias is applied between a conductive AFM tip in contact with the surface of the sample and the substrate, and the deformation of the thin lm (in thickness) can be detected by the cantilever.By exploiting PFM technique, we can accurately locate the regions on a CNF thin lm that experienced morphology change shown in Fig. 3, measure the d 33 and compare with that of the regions without visible morphology change.The de ection amplitude of the AFM cantilever (which can be directly correlates to the sample deformation through the de ection sensitivity of the tip, details can be found in experimental section) with applied bias is presented in Fig. 4. As shown in Fig. 4a, the deformation of region without morphology change is not obvious when applying different bias between the sample and the tip.For effectively poled area as shown in Fig. 4b, the lm deformation increases proportionally to the applied bias.The piezoelectric coe cient d 33 can be calculated from the average linear tting.As shown in Fig. 4c and 4d, the change of the de ection amplitude of effectively poled area is much larger than the noneffectively poled area of the same sample.The measured average d 33 is 46 pm V -1 for poled areas and only 10 pm V -1 for non-poled areas.Same operation of PFM was carried out for poled P(VDF-TrFE) thin lms, as shown in Supplementary Fig. 12b, the obtained d 33 is 46 pm V -1 , like previously reported values.
We now compare the piezoelectric performance of the poled CNF thin lm devices with that of P(VDF-TrFE) thin lms.The linearly increasing piezoelectric voltage with the applied force of poled CNF thin lm of the thickness of 0.86 µm and PVDF-TrFE thin lm of the thickness of 0.23 µm and 0.55 µm are shown in Supplementary Fig. 13.The similar linear tting of the output voltage with increasing force for the CNF and P(VDF-TrFE) thin lms shows typical piezoelectric response.The thicker P(VDF-TrFE) exhibits higher piezoelectric output voltage than the thin one due to the reduced capacitance.For CNF thin lms, the effect of the thickness to achieve optimum piezoelectric output voltage are investigated, and their piezoelectric performance is compared with P(VDF-TrFE) thin lms for benchmarking.As shown in Fig. 4e, the piezoelectric response of non-poled CNF lms is below 0.03 V when a 24 N force is applied.The response of the two thinnest CNF lms is higher than the thicker ones both before and after poling, which seems in opposite trend as for inorganic thin lms [31][32][33][34].This could possibly be due to the better alignment of the CNF bers at the surface of the at substrate.After poling, the piezoelectric response of CNF lms of all thicknesses increased compared to non-poled ones.For samples of thickness between 0.2 to 1 µm, the response of CNF lms after poling (with same force of 24 N) increase with the thickness.This is due to the decreasing capacitance with thickness (see Eq. 1), which is also observed for P(VDF-TrFE) thin lms (open green squares in Fig. 4e).It is important to note that the piezoelectric output voltage of CNF lms is similar to the PVDF based devices in the same conditions.After reaching the maximum value, the piezoelectric output voltage of CNF lms starts to decrease with thickness (from 1 to 1.4 µm).This could be attributed to non-effective poling of the thick lms.Thin polymer lms on a substrate are usually easy to deform under the eld due to competing thermodynamic and kinetic factors [35].The relatively thick CNF lms (around 2µm) were also poled with detachable PDMS electrodes, and no change in morphology and phase was then observed.Thick lms are more stable than the thin ones since the entangled bers hinder their reorientation during electric poling; which result in no obvious changes in the morphology and low piezoelectric responses.
In Fig. 4f, the d 33 of the CNF thin lms with different thickness are presented and compared with P(VDF-TrFE).The area capacitance of the P(VDF-TrFE) thin lm (0.55 µm) of 13.6 nF cm -2 can be obtained from the impedance spectroscopy where the phase angle is close to -90° at high frequency as shown in Supplementary Fig. 15.According to equation ( 1), the d 33 of this P(VDF-TrFE) lm is 44 pC m -1 based on the output voltage in Fig. 4e, which is similar to the PFM values and the previously reported value [36][37][38].Note that the unit of the direct and reversed piezoelectric coe cient are different, however they are equivalent.Though, due to the uneven morphology of the CNF lms, it is di cult to obtain the absolute value of the capacitance of those lms from the same impedance measurement (detailed discussion can be found in Supplementary Note 4).Hence, we cannot calculate the d 33 of CNF lms in the same way as for P(VDF-TrFE).Instead, previously reported dielectric constant of cellulose (ε r =10) [39][40][41] are used to calculate the piezoelectric coe cient of the CNF devices.As shown in Fig. 4f, most of the CNF samples have d 33 in the range between 35 to 55 pC m -1 , which is close to the value obtained from PFM (indicated by the green solid line).We can also see that the effect of electrochemical poling to the piezoelectric performance of CNF thin lms depends on the thickness, the maximum improvement of 13 times takes place when the thickness is around 0.7 µm.The results demonstrate that the piezoelectric voltage and d 33 of electrochemically poled CNF thin lms could reach similar value of P(VDF-TrFE) of similar thickness.

Proposed mechanism
As illustrated in Fig. 5a, before poling, CNF bers composed of alternating amorphous and crystalline regions (enlarged illustration and chemical structures of CNF ber can be found in Fig. 5d and 5g) are lying relatively parallel to the electrodes.The adsorbed water molecules in the CNF lm are mostly bonded to the amorphous region of the cellulose bers between the crystalline regions (CNC), or on the surface through hydrogen bond with hydroxyl groups [42,43] (the water distribution here).At high level of humidity, water molecules can accumulate between the nano bers [44].During the electrochemical poling (Fig. 5b), the ionic conductivity leads to electrochemical reactions as well as perhaps other additional phenomena, such as formation of a space charge polarization like what is found in PVDF lms poled in humid environments [45].The grounded Si electrode is expected to induce half-reaction of O 2 reduction in presence of absorbed water.To describe that in more details, we remind that the oxygen reduction reaction on a non-catalytic surface in neutral environment leads to the production of hydrogen peroxide through the following reduction reaction O 2 + H 2 O + 2e -→ HO 2 -+ OH -and the acid equilibrium HO 2 -+ H + → H 2 O 2 .Therefore, the neutral environment turns rapidly basic upon this electrochemical reaction.Note that both the basic pH and the H 2 O 2 are known to degrade the cellulose [46].The half-reaction taking place at the positively bias gold electrode could be the oxidation of cellulose (Fig. 5h and 5i) as well as the oxidation of water.Interestingly, the oxidation of cellulose has been reported to be electrocatalytic with Au nanoparticles and on Au electrodes in alkali medium triggering the formation of carbonyl groups on the cellulose that increases its solubility [47,48].The breaking of the CNF chains into lower molecular weight fractions could also increase the solubility [49].The formation of carboxylic group in the cellulose lm can be evidenced from the growing characteristic C=O peak at 1730 cm -1 in Supplementary Fig. 17 obtained after poling at higher voltages.The dissolution of cellulose in the alkali solution is known to promote hydrolysis that cleaves the glycosidic linkage of the polymer chains (Fig. 5i) [50,51].Further electrocatalytic oxidation paths lead to the formation of gluconate also carrying C=O bonds [47].
Hence, the poling is accompanied with an electrochemical reaction, and that is why we proposed the name "electrochemical poling".The electrochemical reaction leads to a consumption of the water, and the creation of an alkali medium; which fragilizes the cellulose nano brils (Fig. 5e).Finally, the oxidation of cellulose at the gold electrode is promoted by the poling current, which degrades the cellulose by breaking the chains of the biopolymer (Fig. 5f).We believe that the amorphous regions of the nano brils, softened by the absorbed water molecules [52], that are in contact with Au become the centers of the electrochemical degradation of the cellulose nano brils.As a result, those surface nano brils are cleaved at the level of the amorphous regions; and nanocrystals of cellulose are detached.Finally, the high electric eld tends to orient preferentially the nanocrystals perpendicular to the Au electrode (Fig. 5c and   5f).The observation of such a surface in the AFM phase images with nano-rings corresponding to the diameter of the bers supports the hypothesis.Furthermore, grazing incidence small angle x-ray scattering (GISAXS) experiments performed on non-poled and 110V-poled CNF thin lms also reveal the poling induced structural changes in these lms.As shown in Supplementary Fig. 18, the scattering feature changes observed in the q z and q x linecuts, in conjunction with the information derived by AFM, support the hypothesis of a re-orientation of lying-down bers, with respect to the substrate, towards standing-up bers upon poling.This electrochemical cleavage of the cellulose nano brils is not complete in dry condition (10% RH) due to the lack of absorbed water, hence only partially degraded at bers were observed (without total cleavage, Fig. 5e), while nicely oriented nanocrystals are formed at the surface at and above 40% RH (upon total cleavage, Fig. 5f).The new morphology of compact nano-rings at the interface between the CNF bulk and the Au surface is expected to create a new region of different ionic conductivity, this is supposedly the new ionic transport contribution found by impedance spectroscopy (reduced diameter of the rst semicircle in Nyquist plot in Supplementary Fig. 9).Indeed, the reorientation of ber-like llers perpendicular to the electrode surface has been previously reported to reduce the resistance [53].

Discussion
In this work, we have shown that the electrochemical poling (electric poling in humid environment) is a new method to enhance the piezoelectric coe cient of pure CNF thin lms, and the complete device could easily provide an output voltage of 100 mV with 20 N of applied force.For the rst time, piezoelectric devices composed of pure nanocellulose are demonstrated with piezoelectric output voltage and piezoelectric coe cient d 33 similar to PVDF-TrFE based devices.Combining the electrical changes observed from impedance spectroscopy and morphology change in AFM images, we proposed that the improvement of the piezoelectric performance is due to the breaking and re-orientation of the cellulose bers under electric poling process that induces non centrosymmetric arrangement of the CNC.The humidity level was identi ed to have important effect on the poling process and affect the result of piezoelectric performance.The investigation of the thickness dependence shows that the electrochemical poling enhancement is only effective for thin CNF lms, which limit the output voltage compared to PVDF based devices of scalable thickness.Piezoelectric sensors based on thin lms of pure cellulose are sensitive, non-toxic, biodegradable solution processible, scalable on large area.Those features bring nanocellulose on the map of green and e cient piezoelectric materials with potential impact in selfpowered sensing, actuation, wearable and implantable health monitoring etc. with great advantage in wearable and implantable (bio-) electronic applications.

CNF lms and device preparation
Nano brillated cellulose (NFC) was produced at Innventia AB, Sweden.0.5 wt% CNF dispersion in water was used for the thin lm fabrication.The CNF are carboxymethylated and has a degree of substitution of 0.1.CNF dispersion in DI water was used to prepare CNF lms on doped Si wafers.0.5 wt% CNF solution was poured dropwise on a Si wafer within a PDMS cavity to de ne the area of the CNF lms.Hence, the average thickness of the CNF lm can be tuned by dropping different amount of solution.The lms were then dried in an oven at 60°C for 3 h.
For AFM and GISAXS characterizations, the prepared CNF lms were directly used as the non-poled samples.To perform the poling, PDMS stamps coated with Cr and Au were used as top electrodes that can be removed after poling.Potentials of 110V was applied between the Au coated PDMS and the doped Si wafer for 30 s at room temperature, the top electrode was removed immediately after poling.For impedance and piezoelectric characterization, the top electrode was fabricated by evaporating Cr (5 nm) and Au (50 nm) directly on the top surface of the CNF lms through home-cut plastic shadow mask.

Material characterization
A keithley analyzer (2400) was used to pole the CNF lms with different voltage.The impedance measurement was carried out with impedance spectrometer (Alpha high-resolution dielectric analyzer, Novocontrol Technologies GmbH, Hundsangen, Germany) with an AC voltage of 10 mV in the frequency range of 0.1 Hz to 1 MHz.A dimension 3100 was used for the AFM characterization.The images were recorded in tapping mode with tips of 60 kHz resonance frequency.AFM ICON was used for PFM characterization, the Model of the conductive tip is SCM-PIT-V2 with spring constant of 3 N m −1 and standard de ection sensitivity of 85 nm V −1 (which was calibrated for each measurement).
The thickness of the CNF lms was measured by using a surface pro ler Dektak3ST (Veeco) across scratches on the lms.Roughness of the CNF lms were measured by using optical pro lometer PLu neox (Sensofar).
The piezoelectric response from the devices were tested at room temperature and recorded with National Instrument (NI) systems.The applied force was controlled by tuning the amplitude of the shaker, and the frequency was maintained 3 Hz.The distance between the probe and the sample surface was kept at 1 mm for all the tests.The effective contact area is 1.13 cm 2 .The force was monitored using a force sensor (208C01) placed under the sample.The logging system is NI CDAQ 9174 chassis with NI 9263 voltage output and NI 9239 DAC modules interfaced with a computer using LabVIEW software.
Grazing incidence small-angle x-ray scattering experiments (GISAXS) were performed at the NCD-SWEET beamline of the ALBA Synchrotron, located in Cerdanyola del Vallès, Spain.The wavelength of the X-rays, λ, was 0.9998 Å (12.4 keV), the sample-to-detector distance was 2.538 m and, the angle of incidence, α i , was set at 0.15°.The diffracted intensity was recorded using a Pilatus 1M Dectris detector.Data were normalized by the incident photon ux and the acquisition time.The scattering vectors in the directions perpendicular and parallel to the substrate, q z and q x respectively, are de ned as follows: q z = 2π/λ [sin(α f ) + sin(α i )] and q y = 2π/λ [sin(2θ f ) cos(α f )], where 2θ f and α f are the exit (scattering) angles at the horizontal and vertical directions.

Piezoelectric generator fabrication and characterization
To induce the piezoelectric property in CNF lms we the lms with different poling voltage applied between the top and bottom electrode.At different humidity, the piezoelectric response of the device was recorded rst before poling.The poling for each voltage last for 30 s, and afterwards the piezo response was recorded using the same electrical connection as before poling.The piezoelectric response of the poled device was measured directly after the poling nished, and the data was recorded when the signal was stable (the signal is often high at the beginning of the measurement).For the study in high and low humidity levels, the devices were rst kept in xed humidity for 2 hours and the non-poled characterization was carried out in room humidity immediately afterwards.Then the devices were returned to the same humidity for another 2 hours before the poling was conducted in the same humidity.
The piezoelectric characterization of poled device was carried out right after poling in room humidity.

Figure 2 The
Figure 2