Ultraflexible Organic Active Matrix Sensor Sheet for Tactile and Biosignal Monitoring

Flexible sensors are currently the subject of intensive research, as they allow cost‐effective and environmentally friendly production of large‐area, flexible, and when fabricated on ultrathin substrates, highly conformable devices. Among many intriguing applications, tactile and biosignal monitoring, where lightweight sensors with high wearing comfort are particularly interesting, is focused on here. The required spatiotemporal resolution of the signals is achieved by integrating the sensors in an active matrix configuration. Organic ferroelectric transducers of high uniformity, characterized, for example, by a sensitivity spread of only 1.5%, are combined with similarly uniform ultralow noise level organic thin film transistors operating below 5 V, showing, for example, a threshold voltage variation of just 0.13 V, in a 12 × 12 sensor array. The transistors transition frequency of up to 160 kHz (saturation range) and 17 kHz (linear range) allows for a high spatiotemporal resolution of ≈3 mm at a frame rate of 1400 fps. The thickness of only 2.8 µm renders the organic active matrix sensor sheet ultraflexible and therefore virtually imperceptible on the human skin. Real‐time monitoring of tactile modes in a subset of 8 × 3 pixels and of the pulse wave including heart rate and blood pressure using four sensors of the matrix is demonstrated.

were reported for temperature, [30,31] pressure, [32,33] magnetic fields, [34] fingerprint, [35] and proximity. [36,37] The use of Si nanomembranes and curvilinear designs for the conductors provide enough stretchability to enable in vivo recording of vital signals such as electrocardiograms directly on the heart [38,39] and electroencephalograms directly on the brain cortex, but needs very complicated Si-on-insulator processing. [40] Another, even more compelling strategy for high conformability is the design of intrinsically stretchable, optionally self-healable, organic materials for active-matrix tactile sensors. [41][42][43] However, this approach relies on very special materials that are not widely available and therefore it is not readily scalable. But, the job of ultra-high conformability can also be done in a strikingly simple form by fabricating the e-skin on ultrathin (< 10 µm) substrates, thus conferring the desired ultra-flexibility to the active-matrix sensors. [44,45] Accordingly, ultraflexible activematrix sensors were developed by a number of researchers for spatiotemporal mapping of pressure, [46] full color, [47] and diverse electrophysiological measurements like in vivo ECG (electrocardiogram) on the heart [48] or in vivo EEG (electroencephalography) on the brain. [49] The latter two cases are employing organic electrochemical transistors as multiplexers and sensors and are fabricated on a razor-thin 1.2 µm parylene-derivative substrate. We can further refine this concept towards higher sensitivity and lower noise levels by incorporating appropriate organic amplifier circuits on the matrix sheet, sometimes even in every individual pixel. [37,50] It goes without saying that all materials used for, that is, pseudo-CMOS amplifiers have to be organic to comply with the high mechanical stability and robustness imposed by the ultrathin substrate. Recently we could demonstrate the effectiveness, high reproducibility, and stability of organic active-matrix sensors with organic amplifiers for ECG [51] and magnetic field mapping. [52] The interface to external electronics is becoming more fluid as a recent report of a proximity sensor skin with integrated elements of external electronics illustrates. [37] In an interesting work, Baek et al. [53] demonstrated the spatiotemporal measurement of arterial pulse waves by a piezoresistive sensor sheet coupled to the source electrodes of the printed organic thin film transistor (OTFT) matrix. The matrix resolution and sensitivity were good enough to detect the position of the arterial line and extract the augmentation index accurately. We would like to note that this device is not ultraflexible, as it is composed of a number of separate foils, two micro-structured sensor foils, a spacer sheet, and the active matrix OTFT foil. Table S1, Supporting Information, summarizes the organic active matrix sensors reported in literature compared to our current work. Many of those sensory matrices are impressive in terms of spatial resolution (≥ 10 × 10) (see refs. [S1-S4, S6, S9-S11, S16], Supporting Information), but very few demonstrate a real-time recording of the sensor parameter's spatiotemporal distribution via multiplexing/scanning (see refs. [S1, S7, S8, S12, S16, S17], Supporting Information). Out of these only two are ultraflexible with rather decent resolution for magnetic field or electrophysiology sensing (see refs. [S8, S12], Supporting Information).
Here we present an ultraflexible 12 × 12 active-matrix sensor sheet, an imperceptible artificial e-skin, where piezoelectric sensors and switching transistors were monolithically fabricated on an only 1 µm thin polymer substrate. Our sensory matrix is the first to combine high spatial resolution with multiplexing and ultraflexibility for tactile and biosignal monitoring.
Each pixel of the 12 × 12 matrix consists of an organic thin film transistor based on DNTT (Dinaphtho [2,3b:2′,3′-f ]thieno [3,2-b]thiophene) or C 10 -DNTT (2,9-didecyldinaphtho [2,3-b:2′,3′-f ]thieno [3,2-b]thiophene) that is resistively coupled to a piezoelectric sensor element based on the ferroelectric copolymer poly(vinylidene fluoride: trifluoroethylene) P(VDF:TrFE). Recently, we have shown that such ultraflexible ferroelectric transducers are capable of multi-parameter sensing like the recording of pressure/strain, vibration, temperature as well as proximity and demonstrated wireless reading of the pulse wave and blood pressure on the neck and the wrist (see Scheme 1). [54] Moreover, the imperceptible sensor sheets can also be used for harvesting waste energy from, for example, biomechanical motions thus leading the way to self-powered imperceptible sensor patches (see Scheme 1). The ultraflexible organic 12 × 12 active matrix sensor sheet presented here allows for multi-parameter sensing such as mapping of pressure and touch with a high spatiotemporal resolution of about 3 mm at a frame rate of 1400 fps and is capable of pulse wave monitoring for simultaneous recording of blood pressure and pulse rate.

Results and Discussion
Scheme 1a provides a schematic overview of the ultraflexible ferroelectric active matrix (AM) sensor sheet. A detailed view of the layer stack of a single sensor pixel and a circuit diagram illustrating the principle of operation of the sensor cell are shown in the inset. A photograph of the ultraflexible sensor sheet and its integration with the flexible interposer sheet and the electronic module is provided in Figure S1, Supporting Information. Scheme 1b shows a photograph and a micrograph of the fabricated sensor-skin, while the photographs in Scheme 1c illustrate the lightweight and ultrathin nature of the AM sheet, making it virtually imperceptible on human skin.

Ultraflexible Ferroelectric P(VDF:TrFE) Polymer Sensor Arrays for Active Sensor Matrix
Ultraflexible ferroelectric sensor arrays were fabricated on a 1-µm thin parylene diX-SR (Daisan Kasei Co., Ltd.) substrate. During fabrication, the parylene substrate is fixed on a fluorinated supporting glass carrier which allowed the sensor array to be peeled off after all fabrication steps are completed, as shown schematically in Figure 1a. Notably, we can then apply the ultraflexible sensor sheet to another carrier of our choice, which allows us to equip objects with different shapes or materials with the sensor skin.
As shown in the inset of the photograph of Figure 1a, the capacitor-like sensor structure includes a spin-coated semicrystalline ferroelectric co-polymer poly(vinylidene fluoride: trifluoroethylene) (P(VDF:TrFE) 70:30 ) layer sandwiched between two thermally evaporated metal layers of Al. In order to activate its ferroelectric properties, we apply an electric field E to the ferroelectric layer that has to be larger than its coercive 2201333 (3 of 13) www.advelectronicmat.de field E c . This electric poling step aligns the crystalline domains and induces macroscopic polarization throughout the sample volume. From the poling hysteresis, which is the dielectric dis-placement D plotted over E, we extract the most important characteristics of the ferroelectric layer, namely its macroscopic remnant polarization P r , defined as P r = |D(E = 0)| and its coercive Scheme 1. Ultraflexible ferroelectric active matrix sensor sheet as an imperceptible multi-parameter sensing e-skin. a) Schematic design of the ultraflexible organic active matrix sensor sheet with the layer stack of an individual pixel. Each pixel consists of a piezoelectric transducer based on the ferroelectric copolymer P(VDF:TrFE) and is connected to an organic thin film transistor (OTFT) that switches the transducers during scanning. The pixels are multiresponsive to pressure changes ΔP, temperature changes ΔT, vibration, ultrasound (US), or mid-infrared (MIR) radiation. Transducers and OTFTs are supported by a 1 µm thin parylene substrate. For spatiotemporal matrix operation, we connect the freestanding ultraflexible sensor sheet with an interposer sheet (FLEX-PCB) that carries the electronics for signal conditioning (ADC) and wireless data transmission (Bluetooth Low Energy (BLE) chip). b) Photograph of an ultraflexible AM sensor sheet after fabrication (left) and detailed view of an individual pixel (right). c) Photographs of a freestanding ultrathin AM sensor film (left) and as attached to the human skin (right) to form a sensor-skin kind of augmented or electronic skin.

www.advelectronicmat.de
field E c, defined as E c = |E(D = 0)| ( Figure 1b). P r measures the transducer's sensitivity and is directly correlated to the piezoand pyroelectric coefficient, describing the ferroelectric copolymer's intrinsic piezo-and pyroelectric property, [55,56] while E c defines the field necessary to change the state of poling of the ferroelectric layer. A representative poling hysteresis of an individual transducer pixel is shown in Figure 1b; the displacement hysteresis (red) results from an integration of the poling current (black) recorded during the sweep of the electric field.
The poling current peaks around E c , when the dipoles in the ferroelectric domains change orientation. Figure 1d shows a histogram and Figure 1e is a 3D-plot of the remnant polarization over all 144 pixels in the sensor array. The uniformity of P r across the array is remarkable, as the standard deviation is only 0.9 mC m −2 , which is 1.5% of the average P r,avg = 61.9 mC m −2 .
In order to investigate the sensing properties of the ultraflexible piezoelectric sensors we perform temperature and mechanical stress tests. For the latter, transversal load tests are carried out on elastic and rigid carriers as shown in Figure S2, Supporting Information. For this purpose, sensor samples are loaded with trapezoid-shaped step forces with maximum levels ranging from 0.15 to 10 N. In Figure S2, Supporting Information, the data of one typical measurement on a soft and a rigid carrier are compared. For both carrier substrates, the charge response Q(t) obtained by numerical integration of the measured short-circuit current I(t), approximately corresponds to the shape of the applied force profile (see Figure S2, Supporting Information). From the linear dependence of ΔQ to ΔF we calculate the force sensitivity S F as For the rigid carrier, a 1-mm thick glass plate with an elastic modulus of Y ≈ 70 GPa, the sensitivity is S F ≈ 0.12 nC N −1 over a force range of 0.25 < ΔF < 10 N. For the elastic carrier, a 6-mm thick silicone rubber sheet (Y ≈ 1.45 MPa), we obtain a sensitivity of S F ≈ 1.9 nC N −1 over a force range of 0.15 < ΔF < 2.5 N.
As demonstrated in Figure 1c, for a given transversal load the sensitivity of ultraflexible sensor sheets attached to the more elastic silicone rubber carrier is more than 15 times higher than the sensitivity of samples on the rigid glass. We attribute the strongly enhanced charge signal to the stronger bending deformation of the ferroelectric layer on the more elastic rubber carrier, which transforms into a longitudinal tensile strain that in turn induces transversal stress via the Poisson's ratio, causing contributions from the d 31 and d 32 components of the piezoelectric tensor in addition to d 33 . In contrast, the deformation of the sensor mounted on a rigid carrier results in a simple uniaxial compression along the thickness direction inducing only transversal stress, which causes contributions to the charge signal only due to the d 33 component of the piezoelectric tensor. A more detailed explanation of the relationship between sensitivity and deformation mechanics of ultraflexible ferroelectric sensors on different carriers is given in Ref. [54]. There, we used finite element method simulations to prove that the sensitivity of ferroelectric co-polymer sensors is greatly enhanced by using ultraflexible substrates. In addition to the sensitivity boost, ultraflexible sensors show fast response and excellent mechanical stability.
The ability of the P(VDF:TrFE) co-polymer to generate electrical signals from deformation originates from the direct piezoelectric effect. External stress variations dσ 33 , dσ 11 (tensile stress components transversal and longitudinal to the film plane) induce changes in the dipole density across the sample volume which results in a change of the macroscopic polarization. The change in induced polarization has to be compensated by a change in the charge density at the sensor electrodes, resulting in a piezoelectric current I piezo .
(e) Figure 1. Performance of the ultraflexible ferroelectric sensor sheet. a) Photograph of an array with 144 ferroelectric P(VDF:TrFE) sensors on a 1-µm thin dix-SR (parylene) substrate and its handling scheme. During fabrication, the ultrathin films are fixed on a supporting glass carrier. Afterwards, the finalized devices can be easily peeled off the glass support and attached to various surfaces with different structures, shapes, curvatures or material properties. b) Representative D-E hysteresis curve of a ferroelectric transducer pixel measured during electrical poling at 1 Hz. E c denotes the coercive electric field and P r is the remnant polarization in the absence of an external field, that is, D (E = 0), which is the main figure-of-merit for the transducers. The hysteresis (red) is obtained by the integration of the poling current. c) Charge response of a single transducer pixel to dynamic transversal loads applied to the sensor pixel when attached to a rigid (glass, Young's modulus Y ≈ 70 GPa) or elastic (rubber, Y ≈ 1.45 MPa) carrier. The charge response ΔQ is plotted as a function of the peak forces ΔF. d,e) Typical distribution of the remnant polarization P r over all 144 transducers of a sensor array, represented as a (d) histogram and (e) 3D plot. The average polarization was 61.9 ± 0.9 mC m −2 .

www.advelectronicmat.de
Applying a temperature change ΔT to the P(VDF:TrFE) co-polymer induces a similar effect, known as pyroelectricity. Increasing or decreasing the temperature changes the thermal motions of the polymer chains and thus increases or decreases the macroscopic polarization accordingly. Again, this has to be compensated by a change in the charge density at the electrodes, which, under short-circuit conditions, is measured as a pyroelectric current I pyro . To investigate the pyroelectric behavior of the ultraflexible transducers we perform temperature tests on a precise heating plate in the range between 293 and 330 K. As shown in Figure S3, Supporting Information, the charge response of the ultraflexible ferroelectric transducer closely mimics the shape of the applied temperature profile. From this charge curve, a sensitivity S T ≈ 31 µC K −1 m −2 can be deduced. Temperature changes in the ultraflexible sensors can also be caused by thermal radiation emitted by moving objects, if they have a different temperature than the ambient temperature. Thereby, P(VDF:TrFE) transducers can act as proximity sensors detecting an approaching or moving person by their thermal radiation in the mid-infrared spectral range. Recently, we reported the successful use of printed ferroelectric polymer sensor arrays as proximity detectors. [36,37]

Ultraflexible Organic Thin Film Transistors for Active Sensor Matrix
Our active matrix sensor sheet is driven by organic thin film transistor which are integrated on a single substrate with the sensors as shown in Scheme 1a. In our sensor matrix concept, the electronics to address an individual pixel consists of one OTFT that switches the current generated by the transducer pixel in the source-to-ground path. No drain supply voltage is required, as the piezo-/pyroelectric transducer generates charges by itself upon mechanical or thermal excitation.
The OTFTs are fabricated in a bottom-gate top contact architecture, as shown in Figure 2a, using two different organic semiconductors, namely DNTT and C 10 -DNTT. The gate dielectric is a bilayer of 10 nm thick anodized alumina (ε r ≈ 9) with a 25-27 nm spin-coated organic polymer layer (PNDPE, poly((±) endo,exo-bicyclo[2.2.1]hept-5-ene-2,3-dicarboxylic acid, diphenylester, ε r ≈ 2.3) on top, which has already demonstrated excellent dielectric and interfacial properties in high-performance DNTT-based OTFTs. [57][58][59] As shown in Figure S4, Supporting Information, the bilayer dielectric exhibits excellent stability with very little dielectric loss over the frequency range tested (100 Hz -1 MHz). The resulting stack with all materials and layer thicknesses and an optical micrograph of an individual OTFT is depicted in Figure 2a. Channel length L and width W of the transistors are 45 µm and 1.45 mm, respectively.
As already mentioned, the OTFTs in the active matrix pixel circuit operate without an externally supplied drain voltage. They are driven by the sensor signal with voltage levels ≤200 mV and sensor currents in the range of 5 µA. Therefore, only the OTFT performance in the linear regime is relevant for the operation of the active sensor matrix.
In Figure Figure 2c. The mobility reliability factor r proposed by Podzorov and co-workers, [60] is r = 64% (±2%) and r = 69% (±4%) for the DNTT and C 10 -DNTT thin film transistors, respectively. The r-factor is an indicator of the deviation from an ideal transistor behavior and provides information about real effective mobility.
Critical design parameters for the active matrix sensor arrays are the ON/OFF ratio and the OFF-current of the switching transistors. The off-current should be very low (<100 pA) and the on/off ratio must be large enough to allow an unambiguous distinction between on and off pixels. In all of the 144 pixels on our ultraflexible sensor sheets the ON/OFF ratio exceeds 10 5 (10 3 ) with an average of 3•10 6 (4•10 3 ) for the C 10 -DNTT (DNTT) OTFTs, whereas the OFF-currents are below 100 pA. All C 10 -DNTT (DNTT) transistors operate at voltages below 3 V with a  Table 1 and the distributions of the threshold voltage across the matrices are shown in Figure S5, Supporting Information. A representative transfer and output curve for one DNTT-and one C 10 -DNTT-based OTFT both from the matrix is shown in Figure S6, Supporting Information.
The fabricated OTFTs meet all the requirements to achieve good sensor pixel performance and matrix operation. These requirements are as follows: First, in the OFF-state the transistor's OFF-resistance should be very high, so that the sensor currents flow through the internal resistance of the sensor and not through the OTFT. This is achieved by the very low OFF-and gate leakage currents (I G < 100 pA) observed in all OTFTs of the matrix, as shown in Figure 2b for the OFF-current and Figure  S7a, Supporting Information, for the gate current.
Second, the ON-resistance of the sensors should be so low that most of the generated sensor currents flow through the transistor channel rather than through the sensor's internal resistance. These two requirements are prerequisites for a clear distinction between ON and OFF pixels. Third, the operation voltage V gs needed for ON switching the transistor should be very low to enable the matrix' application in energy-efficient, portable, and battery-powered wireless devices. Our OTFTs operate all at voltages below 3 V (meaning V thr and V on are < 3 V) and can therefore be powered by commercial thin film batteries (a common lithium-polymer battery supplies ≈ 3.7 V). The low operation voltage and low leakage current result in very low static power consumption of < 1 nW per sensor pixel. [36] To guarantee stable matrix operation the switching transistors should be stable against bias stress over a long period of time. As shown in Figure S8, Supporting Information, in electrical fatigue tests, C 10 -DNTT-based OTFTs show no effect on the drain current levels when switched ON and OFF continuously for more than 2.5 h (>5000 cycles). After 2.5 h we observe only a very small decrease in the ON and OFF currents. This comes from a reversible bias stress effect that causes a negative shift in the onset/threshold voltage of the OTFT (induced by charge trapping at the dielectric/semiconductor interface) inducing a reduction of the currents at a specific operation point. After the measurement, the negative shift in the threshold voltage stabilizes within a few minutes, and the ON/OFF levels recover to their initial values. In addition, we compare the transfer characteristics of all OTFTs on a given sensor sheet measured 1 month after fabrication with those measured 3 months after fabrication (see Figure S7b, Supporting Information). We observe no significant, and especially no detrimental, change in performance.
In addition, the OTFTs exhibit excellent mechanical stability, as displayed in Figure S9, Supporting Information. Bending tests were performed by attaching the OTFTs to a thin gold wire with a diameter of 80 µm only. The transfer curves of five OTFTs measured before and after bending show only a slight shift of the onset voltage from 0.71 (0.05 V) to 0.86 V (0.09) after bending, but no change in the gate leakage currents I gs or significant change in the drain currents I ds . This clearly shows that there is no significant and detrimental change in transistor performance by bending along a radius as small as 40 µm. It should not remain unmentioned that the ultraflexible ferroelectric polymer transducers also exhibit excellent mechanical stability, as demonstrated in one of our recent publications. [54] Finally, in applications where signals have to be monitored at high spatial resolution and at high frequency, in other words in applications with high spatiotemporal requirements, the operation speed (frame rate) of an active matrix sensor becomes important. For a given frame rate, the available readout time for a single pixel decreases with increasing resolution and the operating speed of the switching transistors must consequently increase. So the cutoff frequency f t of the switching transistors becomes a very important parameter, because it is the limiting factor for the operation speed of the active matrix and determines the maximum number of addressable sensor pixels.
The bandwidth of OTFTs characterized by the cutoff frequency f t can either be improved by increasing the semiconductor mobility or by down-scaling critical device parameters such as the channel length L and the overlapping length between S/D electrodes and gate electrode (L ov,total ). However, the contact resistance R c can limit the high-frequency operation of transistors as well. When reducing the channel length, the channel resistance R ch decreases proportionally-thus the contact resistance R c , unaffected by L, becomes more and more dominant. A detailed discussion of this short channel effect and its implications on the effective mobility µ eff is found in Figure S10, Figure S11, Supporting Information, section: Downscaling of OTFTs To verify the dynamic performance of the organic switching transistors we investigate 5-stage unipolar ring oscillators (RO)

www.advelectronicmat.de
with an output buffer stage that were fabricated on the same 1 µm-thin parylene substrate as the active matrix. A microscope image of such a ring-oscillator is shown in the upper part of Figure 3a, while its equivalent circuit is found in Figure S12a, Supporting Information. All OTFTs in the ring oscillators have a channel length of either L = 12, 36, or 45 µm and use C 10 -DNTT or DNTT as the organic semiconductor. In the lower part of Figure 3a the signal-propagation delay times (τ) and the related stage frequencies (f s ) of the ROs are plotted as a function of the operation voltage V DD for two different L values in each case.
The fastest ring-oscillator is based on C 10 -DNTT and has a channel length of L = 12 µm. As shown in Figure S12b, Supporting Information, this RO oscillates at a frequency of f osc = 3.4 kHz for an operation voltage V DD = 3 V, at 6.5 kHz for 5 V and at 10.5 kHz for 7 V. The signal-propagation delay (τ) is calculated from the oscillation frequency f osc and the number of stages m by τ = 1/(2(m+1) f osc ). From the relationship τ = (2f s ) −1 we can derive transistor stage frequencies f s = 25, 42, and 68 kHz for τ = 23, 12, and 7.5 µs, which are the values plotted in Figure 3a. In comparison, the fasted DNTT-based RO shows ≈ 50% smaller stage frequencies than the best C 10 -DNTT-based RO with f s = 12 kHz at V DD = 3 V, 20 kHz at 5 V, 29 kHz at 7 V, and 45 kHz at 10 V. This is attributed to the smaller charge carrier mobility of DNTT. For both, DNTT and C 10 -DNTT ROs, the reduction of the channel length from 45 to 12 µm (DNTT) and 36 to 12 µm (C 10 -DNTT) increase the stage frequency level by about one order of magnitude. The measured stage frequencies (and details w.r.t. the calculation), are summarized in Table 2 and Table S2, Supporting Information, for supply voltages of V DD = 3 and 5 V.
To get a better estimation of the switching speed of a single ultraflexible OTFT we have determined the transition frequency f T (in the saturation as f T,sat and linear regime as f T,lin ) by measuring the unity-gain bandwidth as shown and described in Figure 3b and in Figure S13, Supporting Information. [61,62] f T is determined by the crossover point at which the modulated channel current I p-p_d becomes equal to the parasitic gate current I p-p_g for a constant source-drain voltage V ds . The measured f T,sat values of 28 and 62 kHz for 3 V operation agree very well with the calculated f t,sat values of 24 and 67 kHz for OTFTs with a channel length of 12 µm based on DNTT and C 10 -DNTT, respectively, as shown in Table 2. In this context, the theoretical maximum of the speed in the saturation regime can be estimated by calculating the cutoff frequency f t via Equation (2), as explained in Section 4; the results are listed in Table 2. The measured frequency values for the higher operation voltages and longer channel lengths also show a good agreement with the calculated ones as listed in Table 2 and Table S2, Supporting Information. It is worth mentioning that for 5 V operation a remarkable transition frequency f T,sat of 160 kHz is measured for a C 10 -DNTT transistor with a channel length of 12 µm. However, the OTFTs in the active matrix pixel circuit operate in the linear regime because V ds is supplied by the low sensor signals (<200 mV). Thus, the OTFT transition frequency values in the linear regime are more relevant for the maximum operation frequency of the active matrix than the values in the saturation regime. For the small channel length OTFTs (L = 12 µm) f T,lin exceeds 5 kHz with a maximum value of 17 kHz and for the higher channel length devices (L = 36 µm and L = 45 µm) f T,lin exceeds 1 kHz. That allows the operation of a 12 × 12 sensor matrix with a speed of up to 1400 frames per second at a channel length of 12 µm. Please note that the precise monitoring of biosignals like the human pulse wave needs a switching speed in the kHz range (frame rate > 1000 fps, equivalent to a response time < 1000 µs).
When we think about using our active sensor matrix as an e-health patch monitoring biosignals, frequency-dependent OTFT noise (1/f noise or flicker noise) becomes a critical issue. 1/f noise occurs in all electronic devices, therefore also in transistors. Since 1/f noise is inversely proportional to frequency, it is particularly critical in low-frequency applications like biosignal monitoring (biosignals have a frequency domain in the range of several tens of mHz to 1 kHz). Therefore, the 1/f noise in OTFT-based sensor devices should be very low to avoid signal degradation and ensure a high signal-to-noise ratio.
The 1/f noise is strongly affected by shallow traps at the dielectric/semiconductor interface. The origin of the shallow traps is a disorder at the gate dielectric surface and structural defects in the organic semiconductor. [63] For the DNTT-based OTFTs, noise measurements revealed an average noise level as low as www.advelectronicmat.de 2 × 10 −8 µm 2 Hz −1 at 10 Hz (see Figure S14, Supporting Information), which is the lowest 1/f noise level among OTFTs reported to date. [63] The low noise can be attributed to the very low trap density at the dielectric/semiconductor interface, which is usually observed in PNDPE transistors. From the measured subthreshold swing the upper limit of density of interfacial trap states is N ss,max = 5.9 × 10 11 (±3.6•10 11 ) cm −2 eV −1 as extracted according to the method reported by Rolland et al. [64,65]

Operation of the Ultraflexible Active Matrix Sensor Sheet
Our ultraflexible active matrix sensor sheet contains 12 rows and 12 columns, has a pixel size of 3.8 × 2.9 mm 2 , a total active sensing area of 35 × 45 mm 2 , and is processed on a 1-µm thin parylene substrate as shown in Scheme 1. An equivalent circuit of the AM-sheet operated with a wireless data acquisition module is shown in Figure 4a.
The AM-sensor is driven line-by-line so that during one frame time all gate lines WL x are sequentially selected, while the BL x columns connected to the drain of the OTFTs are read in parallel. During the on-time t L a certain gate line is selected by applying a negative voltage that reduces the channel resistance of all OTFTs in the line (meaning it switches the OTFTs on) and the sensor signal is transferred through the channel of the transistors to the analog-to-digital converter's (ADC) input channels of the module. The sensor signal is read out during t r at the end of t L to eliminate any parasitic signals that may be generated by the switching process. During the rest of the frame time, the gate line voltage is high (V gs, WLx = 2.5 V) in order to block the transistor channel for sensor signals. It has to be mentioned that the wireless module is powered by a battery (V Batt = 5 V). Thus, a negative voltage V gs, WLx = −2.5 V is realized by applying 0 V to the gate line and a bias voltage of V Bias = 2.5 V to the source-path of the OTFTs.
In Videos S1-S4, Supporting Information, we demonstrate dynamic tactile sensing in real-time with the ultraflexible AM sensor sheet for different modes of touch like tapping and rolling by a fingertip. We have to note that 8 × 3 pixels were the largest part of the 12 × 12 matrix that could be read out, as the number of functional ADC channels was limited, unfortunately. Figure 4b shows the voltage response of a fingertip touch on different positions of the 8 × 3 matrix section for a frame rate of 10.42 Hz (limited by the available ADC wireless module). Moreover, in Figure S15, Supporting Information, we contrast the voltage response signals of the touch phase (blue colored) and the release phase (red colored). The AM-sheet clearly identifies the fingertip touch events with excellent spatial resolution and a sufficient frame rate, and delivers a response signal that correlates with the strength of the touch.
Another simple and instructive demonstration of an active matrix in operation is shown in Figure S16, Supporting Information. Here, we connect a 2 × 2 organic transistor matrix with four ferroelectric sensors in cantilever configuration and drive it at a low frequency of 0.05 Hz. The rows are switched sequentially between ON/OFF (V gs,WL = −/+2.5 V) while sensor currents I d,BL1,2 are measured in parallel with low resistance ampere meters. By means of the matrix, we can clearly distinguish two simultaneous input events (sensor vibration) without any crosstalk.
For a better understanding of the high-frequency behavior of an ultraflexible AM-pixel, we operate a single sensor pixel at a frequency of 2 kHz. An equivalent circuit of the test setup and a sketch of the pixel operation concept is shown in Figure 5. The AM pixel is alternately switched to the ON and OFF state by varying the gate voltage between +4 and −4 V. As displayed in Figure 5a the sensor signal is read out at the end of the ON-state in order to avoid the switching peaks. In Figure 5b the voltage response to a fingertip during the touch and release phase is shown. The magnification of the measurement demonstrates that large signals are generated during switching the pixel on and off with amplitudes much higher than the signals from the sensors. The origin of these parasitic signals is the charging and discharging of the pixel capacitances which are strongly frequency dependent. However, by placing the sensor read-out time t r at the end of the ON-state the parasitic signals can be completely avoided Table 2. Overview of calculated and measured frequencies of DNTT-and C 10 -DNTT-based organic transistors and 5-stage ring oscillators. The channel length L is 12 µm for all OTFTs. The bilayer gate dielectric comprises a 15 nm thin AlO x layer (ε r = 9) and an ultrathin 25-27 nm PNDPE polymer layer (ε r = 2.3) on top.
OSC: organic semiconductor; b) V op …operation voltage of the RO (V DD ) and the single organic transistors; c) f s …stage frequencies (min-max values) of at least three 5 stagering oscillators with an output buffer stage. It is determined by the relation τ = (2f s ) −1 , where τ is the stage delay calculated from the oscillation frequency f osc of the RO by τ = (2(m+1)f osc ) −1 with m denoting the number of stages; d) f t… calculated cutoff frequency (see Equation (2) in Section 4 for details); e) f T …transition frequency in the saturation (f T,sat ) and the linear regime (f T,lin ) determined by measuring the unity-gain bandwidth. V ds is a constant voltage (shown in brackets), while the gate voltage is an ac input signal with a dc offset (V gs = −V op + sin(ωt)•1 V). At the frequency of transition f T , the amplitudes of the input and output ac currents are the same (see Figure 3b). www.advelectronicmat.de enabling a clean measurement of the sensor signal (see V read in Figure 5b).
To demonstrate the potential of ultraflexible sensor sheets in bio-medical applications, we mounted a ferroelectric sensor array to a human above the radial artery to monitor the propagation of the pulse wave (see Figure 6). Measurements by two pixels are shown in Figure 6, and by four pixels in Video S5, Supporting Information. In principle, we can derive important cardiovascular parameters like heart rate, heart rate variability, augmentation index, pulse wave velocity (PWV), or even blood pressure from these measurements. [54] The pulse wave signal of a 32-year-old female subject at rest, shown in Figure 6, reveals a pulse rate of 60 min −1 . In addition, the pulse wave velocity can be determined by measuring the signal delay Δt for a given sensor distance Δx. In this example, PWV was ≈6.9 m s −1 . From the PWV, a blood pressure of 85 mm Hg was estimated, according to the formula suggested by Ma et al., [54,66] which is a typical value for a 30 -40 year old female.
Since our active sensor sheet is ultraflexible, it can be easily integrated as an imperceptible sensor in a wireless healthcare patch to allow a precise real-time pulse wave mapping with high spatiotemporal resolution, in a comfortable, non-invasive way.

Conclusion
In this study, we demonstrated an ultraflexible active matrix sensor sheet fabricated on a polymer substrate as thin as 1 µm. The addressing of the multi-stimuli responsive organic ferroelectric sensor pixels was accomplished by organic thin film transistors based on DNTT or C 10 -DNTT organic semiconductors. The sensor sheet with a total thickness of only 2.8 µm has 12 × 12 pixels of 2.9 × 3.8 mm 2 each. All of the 144 OTFTs and sensors were functional with remarkable performance uniformity across the matrix (σ(P r ) = 1.5%, σ(V tr ) = 0.13 V) and high bias stress stability. The sensors had an average remanent polarization P r,avg = 61.9 mC m −2 , resulting in a high transversal force sensitivity of up to S F ≈ 1.9 nC N −1 and a temperature sensitivity of S T ≈ 31 µC K −1 m −2 . For all ultraflexible OTFT pixels, the ON/OFF ratio exceeded 10 5 which allowed for an unambiguous distinction between the ON and OFF states of the pixels.
The dynamic behavior of the OTFT was tested in 5-stage ROs, showing a minimum stage delay of only 23 µs at a supply voltage of 3 V. Measurements of the transition frequencies showed values of up to 160 kHz (saturation range) and 17 kHz (linear range), which agreed very well with the theoretical values and allowed us to operate the sensor matrix with a speed of up to 1400 fps. A further increase of the frame rate can be realized by decreasing the critical OTFT parameters, in particular the channel length L, by high-resolution shadow masks. On the one hand, this would increase the speed and decrease the power consumption and on the other hand reduce the sensor pixel footprint and increase resolution. Limits for an upscale of the technology presented here are set by the size of the available evaporation equipment.
An additional advantage, especially for bio-medical applications, is the extremely low 1/f noise level of the transistors of only 2•10 −8 µm 2 Hz −1 at 10 Hz, which can be attributed to their very low interface trap density.
We successfully demonstrated the real-time operation of our ultraflexible AM sensor sheet with a subset of 8 × 3 pixels for detection of different modes of touch (tapping and rolling by a fingertip) without any crosstalk and with a few millimeter spatial resolution. The sensitivity and the operation speed of the AM sensor sheet are sufficiently high for the monitoring of low bio-signals with good spatial and temporal resolution.

Experimental Section
Ultraflexible Active Matrix Sensor Sheet Fabrication: The active matrix was fabricated according to the pixel concept with one organic thin film transistor and one ferroelectric transducer, as shown in Scheme 1. The layer setup of the pixel is displayed in the inset of Scheme 1a and details of the fabrication steps are displayed in Figure S17  www.advelectronicmat.de layers (Figure 1a). Bottom and top electrodes were formed by thermal evaporation of a 90-110 nm thick Al layer through a shadow mask at a rate of 1-2 nm s −1 in a high vacuum system at a pressure of 10 −4 Pa. After the ferroelectric layer was spin-coated (MS-A1500 Opticool by Mikasa), samples were annealed for 5 min on a hot plate at 130 °C, followed by 1 h of annealing in a vacuum oven also at 130 °C. Directly after the annealing step, the samples were taken out of the vacuum oven and cooled down to RT. On top of the transducer layer stack a 300 nm thick passivation layer was applied by CVD and on top of this passivation layer the OTFTs were fabricated. The metal electrodes were thermally evaporated through a shadow mask (60 nm of Al for the gate electrode and 50 nm of Au for the S/D electrodes). As the gate dielectric, a bilayer of ≈10 or ≈15 nm thick alumina (formed by anodization) [ [3,2-b]thiophene (C 10 -DNTT) layer for the p-type OTFTs was deposited under high vacuum conditions on the gate dielectric layer. Before Au-electrode evaporation, holes were drilled with a green laser (marker system Keyence, T-Centric MD-T1000) to connect the source of the OTFTs to the top electrode of the transducer. It has to be mentioned here, first, that the poling step of the transducers was performed before the transducers were connected with the OTFTs, so as to avoid any damage to the OTFT from the high voltage during the poling step, and second, that the line wise connected transducer top electrodes were disconnected after the poling by laser-cutting.
Active Matrix Sensor Sheet Electrical Characterization: The active matrix sensor sheets were electrically tested by a wireless module including a bluetooth chip, two ADC chips for measuring 16 channels simultaneously (bit-lines), and 16 channels to apply modulated signals at the word-lines. The maximum frame rate for the 12 × 12 matrix was limited to about 10 Hz by the available wireless module.
Electrical Characterization of OTFTs and Ring-Oscillators: For the electrical measurements of OTFTs and ring-oscillators, a Figure 5. Read-out of a single active matrix sensor pixel. a) Equivalent circuit of an active matrix sensor pixel consisting of an OTFT and a transducer connected to the source electrode. When the WL is addressed by changing the gate voltage from +4 to −4 V, the p-type transistor is switched ON and the transducer gets in low-ohmic contact with the BL during the on-time t L . So, the piezoelectric charge accumulated at the transducer is discharged across the 10 kΩ resistor inducing the voltage signal V meas . b) The time graphs illustrate the detailed read-out mechanism of a sensor pixel. At every transition of the switching pulse (V gs , red line), a voltage peak occurs on V meas (grey line) due to parasitic capacitance combined with discharge via the pull-down resistor. The significant part of the signal is the offset voltage level during the READ phase towards the end of the ON-state, which is caused solely by a piezoelectric charge generation during touch or release of the transducer. Therefore, only this part of the signal is captured as a read-out voltage V read (blue) during t r (green line) and related to the force/pressure input by touch.

www.advelectronicmat.de
semiconductor parameter analyzer (Keysight, B1500A), a manual probe station (Cascade Microtech MPS 150) in a dark shield box, and a digital oscilloscope (Keysight DSO-X 3014A) were used. For the measurements of thin-film capacitors, a chemical impedance analyzer (E4980A, Keysight Technologies, Inc., Santa Rosa, California, USA) was employed. All electrical measurements were conducted under ambient conditions.
Electrical Poling and Hysteresis Measurements of the Ferroelectric P(VDF:TrFE) Transducers: For poling, a programmable signal generator (B2912A Precision Source/Measure Unit, 2 ch, 10 fA system) was used. Sinusoidal voltages with amplitudes up to 140 V and a frequency of 1 Hz were applied to the samples while simultaneously recording the current flows. Poling was done line by line (12 poling steps for a 12 × 12 matrix) or pixel by pixel. By numerically integrating the poling currents, displacement-field (D-E) hysteresis loops were plotted (Figure 1b). For sufficient poling, a maximum electric field of 100 V µm −1 was applied, which was about twice the coercive field strength of the P(VDF:TrFE) films.
Transversal Load Tests: A Shimadzu SZ-SX (1-500 N) with customized equipment was used for transversal load tests. Current signals were measured with a precise source measuring unit (Keysight B2912A, 2 ch -10 fA system). For transversal pressure tests, a teflon stamp transferred the compression force over a contact area of 4 mm 2 onto the transducer, while force levels were monitored with a calibrated load cell. The force profile was trapezoidal with amplitudes between 0.15 and 10 N. As a carrier, either glass or silicone rubber was used.
Transition Frequency f T Measurement of an OTFT: The transition frequencies f T,lin and f T,sat of OTFTs were determined by measuring the unity-gain bandwidth as explained in Refs. [61,62]. The details of the setup used for these measurements can be found in Figure S13, Supporting Information.
Human Research Participants: To demonstrate the pulse wave monitoring by the ultrathin organic sensing sheet, one healthy female (age 32) was tested. Informed consent was obtained from the participant. All experiments regarding pulse wave monitoring complied with guidelines by the Osaka University Research Ethics Committee.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author. Figure 6. Imperceptible e-health patch. The top left inset shows the scheme of the e-health patch, whereby an ultraflexible ferroelectric sensor array attached to the skin imperceptibly senses the human pulse wave and its propagation. Cardiovascular parameters (heart rate, heart rate variability, pulse wave velocity, blood pressure) can be extracted from these biosignals, allowing conclusions to be drawn about cardiovascular health. The potential of ferroelectric sensors to monitor the radial arterial blood flow is demonstrated by two sensors selected from a ferroelectric sensor array attached to the wrist. The pulse wave signals of sensor pixels S 4 and S 2 (bottom right inset) clearly show the propagation of the pulse wave signal.