Intrinsically stretchable and transparent thin-film transistors based on printable silver nanowires, carbon nanotubes and an elastomeric dielectric

Thin-film field-effect transistor is a fundamental component behind various mordern electronics. The development of stretchable electronics poses fundamental challenges in developing new electronic materials for stretchable thin-film transistors that are mechanically compliant and solution processable. Here we report the fabrication of transparent thin-film transistors that behave like an elastomer film. The entire fabrication is carried out by solution-based techniques, and the resulting devices exhibit a mobility of ∼30 cm2 V−1 s−1, on/off ratio of 103–104, switching current >100 μA, transconductance >50 μS and relative low operating voltages. The devices can be stretched by up to 50% strain and subjected to 500 cycles of repeated stretching to 20% strain without significant loss in electrical property. The thin-film transistors are also used to drive organic light-emitting diodes. The approach and results represent an important progress toward the development of stretchable active-matrix displays.

W earable and biomedical electronic applications demand advanced materials and electronic devices to simultaneously possess deformability, solid state, light weight, visual transparency and low processing cost [1][2][3][4][5] . Stretchable thin-film field-effect transistor (TFT) would become a fundamental building component enabling a variety of stretchable electronic devices including displays, sensor arrays, thin-film circuits and identification tags for control, data-processing and communication. We have recently reported intrinsically stretchable solid-state organic light-emitting devices (OLED) [6][7][8] and simple pixelated displays. For high-content information display, an intrinsically stretchable TFT array is also highly demanded to drive a display panel comprising a large number of pixels. Being transparent for the OLEDs and the TFTs would significantly expand the range of applications of the active-matrix displays and increase the aperture ratio of the display pixels. Developing a stretchable transparent TFT is thus a fundamental step toward this challenging yet exciting goal.
TFT is a three-terminal device comprised of source, drain and gate electrodes, a dielectric layer, a semiconductor layer and a substrate. Developing an intrinsically stretchable TFT entails development of all these materials with elastomeric stretchability. The requirements of visual transparency and printable process significantly increase the challenges in materials and process development. Traditionally, amorphous silicon and polysilicon are widely used as semiconductor channel for TFTs; evaporated metal films or indium tin oxide (ITO) and are preferred choices for the electrode materials; silicon oxide or high dielectric polymers are usually used for the dielectric materials. None of these materials is intrinsically stretchable, and most do not meet the requirements of being transparent and printable 9 . Advanced materials and processing techniques developed in recent years for flexible nanoelectronics employing metallic nanowires, semiconducting nanowires, single-walled carbon nanotubes (SWCNTs) and functional polymer-based dielectrics offer unique functionalities, promising performance and low-cost processing strategies that could be explored for the development of stretchable TFTs [10][11][12][13] .
To impart mechanical compliancy and robustness to these materials and the resulting TFTs, two general strategies have been employed to create TFTs capable of stretching without undergoing physical damage 13 . The first approach configures conventional brittle or rigid inorganics into specialized structural layouts and mechanical designs, such as wrinkled or buckled structures, to accommodate strains applied in the materials [14][15][16][17][18][19][20][21] . Non-stretchable device islands connected with stretchable conductors or wavy metal interconnects have realized high-performance integrated systems that consist of hundreds and even thousands of TFTs 14,15 . Kim et al. 17 introduced noncoplanar mesh designs and serpentine interconnects for the array of stretchable complementary metal-oxide-semiconductor integrated circuits, enabling them to accommodate even demanding configurations such as corkscrew twists with tight pitch and linear stretching to strains B140%. Kaltenbrunner et al. 18 fabricated organic field-effect transistors directly on ultrathin (1 mm) polymer foils which accommodate stretching up to 230% on pre-strained elastomers. A stretchable TFT combining stretchable graphene/SWCNT electrodes and a SWCNT network channel with a geometrically wrinkled Al 2 O 3 dielectric layer has also been reported through this approach 19 . These reported devices can achieve high electrical performance and impressive stretchability, but lack visual transparency, and the fabrication process is cumbersome and incompatible with fully solution-printed process. The second strategy employs intrinsically stretchable materials and components to obtain fully (or intrinsically) stretchable TFT [22][23][24][25] . Several intrinsically stretchable TFTs have been reported, including a graphene-based TFT employing graphene as both the semiconducting channel and the source/drain electrodes, and an ion gel as the gate 24 , and an organic-based TFT using poly(3-hexylthiophene) as the channel, carbon nanotubes as the source/drain, polyurethane (PU) as dielectric and liquid metal as gate 22 . However, these devices are not solid state and suffer from either low electrical performance, opacity or limited stretchability. Despite great efforts being exerted in this field, so far no one has succeeded in demonstrating a fully stretchable, solid state, transparent TFT suitable for stretchable thin-film displays.
In the following, we report the successful development of such a device, using all solution-based printable processes. The work includes the following essential elements: (1) the stretchable TFT employs a silver nanowire (AgNW)-PU acrylate (PUA) composite as the stretchable transparent electrodes with high surface conductivity and low surface roughness, SWCNT network channel and a PU-co-polyethylene glycol (PU-co-PEG) elastomeric dielectric; the TFT has high mobility, modest ON/OFF current ratio (I ON/OFF ), high peak ON current (I ON ), large transconductance and relatively low operating voltages in unstretched state. (2) The TFT can be stretched by up to 50% while retaining a high mobility; at 20% strain, it can be stretched by 500 cycles; (3) the entire fabrication process is printable: the electrodes, semiconductor layer, dielectric layer and substrate are all processed from solutions at ambient conditions. (4) The TFT exhibits optical transmittance of 490% in the 450-1,100 nm wavelength range. (5) The TFT can drive an OLED in the full brightness range.

Results
Stretchable TFT fabricated by solution-based techniques. The structure and fabrication process of the stretchable transparent TFT are illustrated in Fig. 1a. The first key step of the process is to form the patterned stretchable source/drain electrode, as illustrated in Fig. 1b. It started with spray-printing a layer of AgNWs from dispersion in methanol through a shadow mask on glass substrate. The opening of the mask defines the source/drain electrode pattern, with a channel length (L) of 100 mm and channel width (W) of 5,000 mm. A stretchable polymer PUA is chosen to transfer and embed the AgNW source/drain electrode into the surface layer of a rubbery matrix, according to a previously reported approach 8 . The inset scanning electron microscopy (SEM) image in Fig. 1b shows that the edge of the AgNW pattern is fairly smooth, which is essential to the reproducibility and stability of the TFT devices. The sheet resistance of AgNW-PUA composite electrode can be controlled in a wide range with varied transmittance. The electrode with 15 O sq À 1 was chosen in this work for a compromised high conductivity, stretchability and optical transmittance 8 .
The second key step is to form a uniform channel material on the AgNW-PUA composite source/drain substrate which is hydrophobic. In this study, the active channel material was printed from drop-casting a certain amount of in-house formulated semiconductive SWCNT ink followed by drying and then rinsing with deionized water and isopropyl alcohol (IPA) alternatively. This semiconductive SWCNT ink was prepared by mixing Fluorosurfactant and propylene glycol with the 99 þ % semiconductive SWCNT solution (IsoNanotubes-S from NanoIntegris, Inc.) at an optimized ratio (detailed in Experimental part). The addition of Fluorosurfactant and propylene glycol enhances wetting and levelling control and of the SWCNT solution, and thus brings up uniform deposition of the SWCNT network on channel ( Supplementary Fig. 1). The density of the printed SWCNT depends on the amount of SWCNT ink cast on the substrate. Three densities with estimated values of B25, 13 and 6 tubes mm À 1 from three different casting amounts as shown in the SEM images from Fig. 1e were made and investigated in this study.
The third key aspect is to identify a stretchable and transparent dielectric layer with high dielectric constant and good mechanical compliance. For this purpose, PU-PEG copolymer with PU-co-PEG main chain and PEG side chain formed from siliconized urethane acrylate (CN990 from Sartomer Inc.), methoxy PEG monomethacrylate (SR550) and PEG diacrylate (SR259). The siliconized urethane acrylate oligomer was selected for its high stretchability and compatibility with the source/drain and gate composite electrode. The PEG diacrylate was selected for its high flexibility, low monomer viscosity which is important for printing, high dielectric constant 26,27 and semicrystalline structure of its homopolymer 26 . Methoxy PEG monomethacrylate was selected to introduce side chains with its high polarity and chain mobility at room temperature. The precursor solution was overcast and spin-printed on the SWCNT/AgNW-PUA source/drain followed by UV-curing to form a thin elastomeric dielectric overcoating and wrapping around the SWCNTs. The cross-sectional SEM image of a typical device (inset in Fig. 1a) confirmed the dielectric-wrapped SWCNT structure. Various weight ratios of CN990:SR550:SR259 were studied, and a ratio of 1:1:1 was found to give the optimal overall performance of the resulting elastomeric dielectric copolymer (EDC) in terms of dielectric constant (as high as  Table 1, the small loss factor (B0.011), almost overlap of the stress-strain curves during loading and unloading, and high resilience value suggest small hysteresis, fast recovery and good elasticity for the EDC (see Supplementary Note 1).
Finally, the resulting EDC/SWCNT/AgNW-PUA source/drain structure was laminated with another 15 O sq À 1 AgNW-PUA composite electrode as gate electrode on top of the dielectric layer to complete the TFT fabrication. Thanks to the amalgamation of all transparent components, the SWCNT-AgNW TFT arrays show high transparency, as displayed in Fig. 1c, where the element 'Ag' of the periodic table can be clearly seen through a TFT array consisting of six TFT devices (marked with brown dash lines) in the optical image. Figure 1d depicts that the optical transmittance of the array is 490% in the 450-1,100 nm wavelength range.
Electrical properties of the stretchable TFT. Electrical characteristics of a typical SWCNT-AgNW TFT with a density of SWCNT about 25 tubes mm À 1 are shown in Fig. 2a  were tested in air at room temperature. The SWCNT (with 99 þ % semiconductive SWCNTs) used in this work has a length range from 300 nm to 5 mm. Channel length of 420 mm is reported to be required to decrease the probability of percolative transport through the metallic nanotubes in order to achieve I ON/OFF 410 4 (refs 28,29). A channel length of 100 mm and channel width of 5,000 mm are thus selected, but not optimized, in the present work for all device testing. A family of output curves appears to be linear at low fields with gate voltage stepped from À 1 to À 5 V. The specific contact resistance between the semiconducting SWCNT network and AgNW electrodes is calculated to be 0.219 ± 0.043 O cm 2 through the transmission line model measurements as detailed in Supplementary Fig. 4 and Supplementary Note 2 (refs 30,31). The contact resistance between SWCNT and AgNW is higher than that between SWCNT and evaporated Ag electrode (0.02 O cm 2 ) 30,31 . The work function of the AgNW is measured to be B4.90 eV by Kelvin probe force microscopy, which is higher than the typical work function of Ag (4.73 eV), indicating surface of AgNW partially covered by Ag 2 O. This surface oxide might have a dominant effect on the interface characteristics with SWCNTs, and thus increase the contact resistance. However, this contact resistance between AgNW and SWCNTs is still negligibly small compared with the channel resistances because of the large channel length used in this work 32 . Clear current saturation at high drain voltages due to pinch-off is also evident. Moreover, notable increase of the conductance of the channel with the increasing gate voltage from 0 to À 5 V in 1 V steps manifests the effective gate modulation of the EDC.
From the transfer characteristics shown in Fig. 2b, one typical device exhibits a peak I ON value of 140 mA and peak transconductance (g m ) of 54.4 mS, corresponding to 28.0 mA mm À 1 (I ON /W) and 10.9 mS mm À 1 (g m /W) as normalized by the channel width at V D ¼ À 2.0 V. Well-defined ON-state and OFF-state currents were observed with a I ON/OFF of 7,000 at V D ¼ À 2.0 V. A gate-sweep-rate-dependent hysteresis was also observed for the SWCNT-AgNW TFT devices (see Supplementary Fig. 5), which is typical for SWCNT-based devices (see Supplementary Note 3) 33,34 . The mobility of the device is extracted using the following equation: where m is the device mobility, I D the drain current, V G gate voltage, L channel length, W channel width and C i specific capacitance of the elastomeric dielectric. C i was measured from parallel-plate test structures with an average value of B16.7 nF cm À 2 at 12 Hz with thickness being B1.1 mm. An unstretched device mobility of 32.5 cm 2 V À 1 s À 1 is thus calculated from the transfer characteristics in Fig. 2b. However, SWCNTs are sparsely distributed in the channel (Fig. 1e), and the parallel-plate capacitor model overestimates the gate capacitance and underestimates the mobility. Using the cylindrical model that takes into consideration of the effect of electrostatic coupling between nanotubes to estimate the gate capacitance 19,35,36 , a higher mobility of 51.6 cm 2 V À 1 s À 1 is obtained (see Supplementary Notes 4 and 5).
In 7 different batches for a total of 36 TFT devices fabricated, the yield was 85.7%. The I ON/OFF , peak I ON value, g m and mobility were measured in the ranges of 4,500-12,300, 8.6-23.0 mA, 32.3-72.1 mS and 19.3-43.1 cm 2 V À 1 s À 1 , respectively. Fig. 2d-f present the histograms of the statistical variations for these 36 devices in I ON/OFF , I ON /W, g m /W and mobility, respectively. The I ON /W and g m /W were measured at V D ¼ À 2.0 V, and the mobility was also extracted as described above at V D ¼ À 2.0 V. The average of I ON/OFF ratio, I ON /W and g m /W and mobility is 7,200±1,200, 30.0±8.93 mA mm À 1 , 11.2±3.40 mS mm À 1 and 27.0 ± 7.44 cm 2 V À 1 s À 1 , respectively. The variation in the device performance has been widely observed in printed SWCNT transistors 11 and could be improved by delicate control of the printing parameters 37 .
The device characteristics strongly depended on the density of the SWCNTs (see Supplementary Figs 6 and 7), which was controlled by the amount of SWCNT ink drop-casted on the substrates (Fig. 1e). Supplementary Figure 8 shows I ON /W, g m /W, µA) Number of devices I ON/OFF and mobility as a function of density of nanotubes for V D ¼ À 2.0 V. As the density of nanotubes decreases from 25 to 6 tubes mm À 1 , the average of I ON/OFF increases from 7,200 to 18,300. This increase of I ON/OFF ratio at low nanotube density is possible due to the decrease in the probability of percolative transport through the metallic nanotubes 29,38 . As long as the percolation threshold is smaller than the density of CNTs but larger than that of the metallic tubes, all percolation paths from source to drain must involve at least one semiconducting segment. Gate modulation of this 'week' connection of semiconducting SWCNTs ensures high I ON/OFF ratio 39,40 . At high SWCNT densities, the density of metallic SWCNTs can exceed percolation threshold and the network forms continuous paths across the device, which results in high off current (I OFF ) and low I ON/OFF . As a trade-off, low density of nanotubes leads to a decrease of I ON /W and g m /W. As a result of the direct dependence of mobility with g m , the average mobility also decreases from 27.0 to 9.22 cm 2 V À 1 s À 1 (parallel model) with the decrease of the density of SWCNT from 25 to 6 tubes mm À 1 (see Supplementary Note 5). In the present work, the SWCNT density for the TFT device was optimized to B25 tubes mm À 1 to have a high average mobility of 27.0 cm 2 V À 1 s À 1 , and an average I ON/OFF of 7,200, which meets the demand of an I ON/OFF ratio of at least B10 3 -10 4 for the application of digital logic circuits 28,39 .
To investigate the electrical properties of the device under mechanical deformation, the transfer characteristics were further measured during the application of a uniaxial strain along either the channel length (Fig. 3a,b) or channel width (Fig. 3c,d). The I ON value is declined by 52 and 51% with the strain increase from 0 to 50% along the channel length and width directions respectively (see Supplementary Fig. 9). In the meanwhile, I OFF also exhibits a slight decrease under strain of up to 50%: the value is dropped by 17 and 46% with respect to the unstretched values along the channel length and width directions, respectively. The decrease in the source-drain channel current during stretching is mainly attributed to the increase in the contact resistance between the SWCNTs when stretched (see Supplementary Fig. 10) 19,41,42 . The mobility determined from the g m and capacitance data (see Supplementary Fig. 11) exhibits a decrease, from 32.7 and 28.2 cm 2 V À 1 s À 1 before stretching, to 16.2 and 15.0 cm 2 V À 1 s À 1 at 50% strain along channel length and width directions, respectively. It is suggested that the decrease in capacitance of EDC with increasing strain is the main cause leading to the decrease of mobility (see Supplementary Fig. 11). The chain mobility of PEG in the DEC would be restricted when tensile strain is applied on the EDC, which results in the decrease of dielectric constant (see Supplementary Note 6). The elongation of the channel length between the AgNW-PUA source and drain electrodes can be obviously observed from the magnified photographs of TFTs at specific strain applied along channel length direction (Fig. 3e). The device fails when the strain is over 50% because of the tearing of elastomeric dielectric. Furthermore, I ON /W, g m /W and I ON/OFF at strain of 50% along the channel length and width directions, which can be obtained from the transfer characteristics from Fig. 3a-c, are calculated to be 11.0 mA mm À 1 , 8.22 mS mm À 1 and 8,400 along the channel length direction, and 10.4 mA mm À 1 , 7.61 mS mm À 1 and 15,400 along the channel width direction, respectively. It is of noteworthy that the electrical performance of the stretchable TFTs at 50% strain is still comparable to most unstretchable TFT devices using SWCNTs as the channel material.
A fatigue test was subsequently performed by subjecting the TFT to 500 cycles of continuous stretch-relaxation between 0 and 20% strains along channel length or width direction. Figure 4 shows the electrical transfer characteristics, I ON and I OFF change during the continuous stretch-relaxation cycles. The I ON for both strain direction decreased lightly over the 500 stretch-relaxation cycles, and the I ON value remained fairly high value of B60 mA (initial value B140 mA) along channel length direction and B90 mA (initial value B160 mA) along channel width direction after 500 cycles (see Supplementary Fig. 9). In contrast, the I OFF values remain relative constant throughout the cycling measurements along either direction. The calculated I ON/OFF after 500 stretching-relaxing cycles along the channel length and width directions is B10,000 and 13,000, respectively. The mobility degraded relatively fast in the initial 200 cycles for both direction, but slowed down in the following 300 cycles. Mobility of 11.7 and 17.6 cm 2 V À 1 s À 1 were obtained from the transfer characteristics curves after 500 stretch-relaxation cycles along channel length and width direction respectively. The retention of the device mobility after repeated strain cycles is still much higher than data reported in literature for all-printed SWCNT-based transistors 37 . We speculate this good device performance for the device after subjecting repeated stretch-relaxation cycles is resulted from the unique structure where the EDC infiltrate into the SWCNT network and wraps around the nanotubes (Fig. 1a). The elastomeric dielectric wrapping around the SWCNT should constrain the junction disjoining and tube sliding, help distribute the stress more uniformly under tensile strain and thus increase the stretchability and stability of the SWCNT network under large strains 6,7,43,44 . Figure 5a,b display the normalized resistance (R/R 0 , the ratio of the instantaneous resistance at a specific tensile   ARTICLE strain to the initial resistance at zero strain) for the SWCNT coated on PUA substrate with and without the EDC overcoat during 1,000 cycles of tensile stretching-releasing between 0 and 20% strains. It can be seen that after a gradually increase in the initial 100 cycles, the resistance of the SWCNT Network with EDC overcoat becomes very stable in the subsequent 900 cycles. The resistance increase during the first 100 cycles can be attributed to the alignment and sliding of the SWCNT network during deformation and the viscoelasticity of the elastomeric dielectric 6,44 . Overall, the elastomeric dielectric helps preserve the SWCNT network during large deformation very well. This trend of resistance change for the SWCNT network is consistent with the trend of mobility change for the TFT during continuous stretching-relaxing cycles as shown in Fig. 4b,d. In contrast, it can be observed from Fig. 5b that the SWCNT network without EDC overcoat exhibits a gradual increase in resistant continuously over 1,000 cycles.
OLED driven by stretchable TFT. To prove that the stretchable TFTs is applicable for stretchable active-matrix OLED displays, a prototype of white-light OLED driven by the stretchable SWCNT-AgNW TFT was demonstrated. Figure 6a,b present the output and transfer characteristics for the stretchable TFT used in the OLED control circuit. The TFT is able to source a current 150 mA at V G ¼ À 8 V and V D ¼ À 2.0 V with an I ON/OFF B1.0 Â 10 4 , which can meet the demand for driving and controlling the OLED. The white-light OLED used in this study has a sandwiched configuration of glass/ITO/PEDOT:PSS (40 nm)/white emissive polymer layer (75 nm)/CsF (1 nm)/Al (100 nm) (see Supplementary Note 7). In the drive circuit (inset of Fig. 6c), the connection was made by putting a probe onto the drain electrode of the device, and the probe is connected to a clamp by a conductive wire. The clamp is then fixed on the cathode of the OLED. A negative voltage was applied to anode of the OLED. Figure 6c displays a set of OLED current (I OLED ) versus drain supply (V DD ) characteristic curves for the OLED circuit under various input gate voltage V G . Good diode behaviour with a clear cutoff region and triode region could be clearly observed, indicating a good control from the stretchable TFT over the OLED. The cutoff voltage of V DD is approximate À 2.7 V, which is in accordance with the threshold voltage (V th , also called turn-on voltage) of the OLED (see Supplementary  Fig. 12) 45 . Moreover, from Fig. 6d, the current flow through the OLED can be modulated by changing the input V G by a factor of B1,000. This modulation could lead to the control of the OLED luminance. The inset optical photographs in Fig. 6d demonstrate the OLED operated with V G ¼ À 8 V (upper), À 3 V (middle), 0 V (lower), respectively, obviously showing the OLED brightness modulation. When V G ¼ À 8 V, the OLED (with lighting area of 2.0 Â 2.5 mm) is on and the light intensity is measured to be 218 cd m À 2 ; when V G ¼ 0 V, the OLED is off and the luminance is measured to be lower than 0.01 cd m À 2 , the detection limit. Thus, the modulation for the OLED luminance is larger than 2 Â 10 4 . The OLED control circuit is further characterized by measuring the I OLED with the stretchable TFT under 0, 20 and 30% strains applied along the channel length direction respectively. From the I OLED -V DD characteristic in Fig. 7a-c, field-effect transistor-like behaviour with various curves corresponding to various values of input V G is still clearly observed even when the TFT was stretched up to 20 and 30% strains respectively. The I OLED flow through the OLED shows a mild decrease trend with the increasing strain. Figure 7d-f show the I OLED versus V G curves with the stretchable TFT under specific strains. A close examination of the curves reveals that the V G modulation factor is B1,000, 750 and 650 for the TFT under strain of 0, 20 and 30%, respectively. The decrease of the modulation factor is attributed to the decrease of I ON , in consistency with the discussion above (Fig. 3). Figure 8 further shows the light-emitting properties of the OLED circuit at different V G when V DD is À 4 V. The brightness of the OLED integrated with TFT under different strains of 0, 20 and 30% is shown for comparison. When the TFT was stretched to 20% strain, the brightness decreased from 196 cd m À 2 (initial value at 0% strain) to 120 cd m À 2 at V G ¼ À 8 V, which is still sufficient for display applications. When stretched up to 30%, the brightness declined by B68% to 63 cd m À 2 .

Discussion
In summary, a high-performance, transparent and intrinsically elastomeric TFT have been fabricated using a solution-based printable process. The device is made of all stretchable materials, including AgNW-PUA composite conductor as the source, drain, and gate, a SWCNT network channel and PU-co-PEG elastomeric dielectric. The SWCNT network is infiltrated by the elastomeric dielectric, and wrapping the nanotubes without interrupting the nanotube-nanotube junction is beneficial for large-strain stretchability without significantly diminishing electric performance. The elastomeric TFT retains fairly high electrical performance at strains as high as 50% and can survive a large number of stretching cycles. The stretchable TFT is successfully adapted to switch a white-light OLED from totally dark to 196 cd m À 2 or display brightness. The TFT layout and fabrication process could be adapted to fabricate a large of TFT to achieve stretchable active-matrix OLED display.  Preparation of elastomeric dielectric. The precursor of DEC was formulated by mixing the components CN990, SR259, SR550 and photoinitiator, in an optimized weight ratio of 1:1:1:0.015. CN990 was chosen for the high stretchability and good bonding of its homopolymer with the AgNW-PUA source/drain and gate composite electrode. SR259 and SR550 were chosen for the good flexibility and high dielectric constant of their homopolymer and low viscosity of their monomer (25 and 19 c.p.s. @ 25 o C, respectively). SR550 can serve as a polar side chain to further improve the dielectric constant.
Fabrication of fully transparent and stretchable transistor. The device fabrication started with printing SWCNT on the AgNW-PUA composite source/drain electrode. First, SWCNT ink was prepared by mixing 99% þ Semiconducting -SWCNT aqueous solutions (used as-received from Nanointegris Inc.), DI-water, propylene glycol and Fluorosurfactant at the ratio of 25:5:5:1, followed by untrasonicating for 20 min. Subsequently, a certain amount of SWCNT ink was uniformly drop-casted on the conductive surface of AgNW-PUA source/drain composite electrode, followed by heating up to 90 o C on a hot plate to dry the SWCNT network. SWCNT densities on the source/drain channel was controlled by the amount of SWCNT ink being 6, 15 and 30 ml on 0.5 Â 1.5 cm 2 AgNW-PUA composite source/drain electrode to obtain SWCNT densities of B6, B13 and B25 tubes mm À 1 , respectively. The SWCNT/AgNW-PUA source/drain was rinsed with Deionized water and IPA alternatively to remove the sodium dodecyl sulphate surfactant and Fluorosurfactant, followed by drying in vacuum for 1 h. SWCNT was patterned by O 2 plasma etching to the source/drain electrode. Then, the elastomeric dielectric precursor was overcoated on the SWCNT/AgNW-PUA source/drain substrate, spin-printed at 8,000 r.p.m. for 1 min and then UV-cured in nitrogen gas. Transparent tape (3M) was used to define the openings for the source/drain electrode and peeled off after polymerizing the elastomeric dielectric. Finally, the resulting elastomeric dielectric/SWCNT/AgNW-PUA source/drain structure was laminated with another 15 O sq À 1 AgNW-PUA composite electrode as gate on top of the dielectric layer to complete the device fabrication.
Characterization method. Cyclic strain tests were performed on a motorized linear stage with built-in controller (Zaber Technologies Inc.). A Keithley 2000 digital multimeter was used tomonitor the resistance change. Strain and resistance data were recorded with a custom-made LabViewcode. The transmittance spectra were recorded utilizing a Shimadzu UV-1700 spectrophotometer. SEM was performed on a JEOL JSM-6701F scanning electron microscope. The dielectric constant and capacitance of the elastomeric dielectric were measured using LCR meter (GW INSTEK LCR-819.) the by fabricating 12 ITO-elastomeric dielectricevaporated alumina capacitors and averaging the unit area capacitance of all 12 capacitors. Kelvin probe force microscopy was conducted with a Bruker Dimension Icon Scanning Probe Microscope to measure the work function of AgNW. Transistor electrical characterization was performed with two Keithley 2400 source meter. The measurement sequences were controlled by a custom-made LabViewcode. All the transistor measurements were tested under ambient atmospheric conditions.