A flexible artificial intrinsic-synaptic tactile sensory organ

Imbuing bio-inspired sensory devices with intelligent functions of human sensory organs has been limited by challenges in emulating the preprocessing abilities of sensory organs such as reception, filtering, adaptation, and sensory memory at the device level itself. Merkel cells, which is a part of tactile sensory organs, form synapse-like connections with afferent neuron terminals referred to as Merkel cell-neurite complexes. Here, inspired by structure and intelligent functions of Merkel cell-neurite complexes, we report a flexible, artificial, intrinsic-synaptic tactile sensory organ that mimics synapse-like connections using an organic synaptic transistor with ferroelectric nanocomposite gate dielectric of barium titanate nanoparticles and poly(vinylidene fluoride-trifluoroethylene). Modulation of the post-synaptic current of the device induced by ferroelectric dipole switching due to triboelectric-capacitive coupling under finger touch allowed reception and slow adaptation. Modulation of synaptic weight by varying the nanocomposite composition of gate dielectric layer enabled tuning of filtering and sensory memory functions.

(0.375%, 0.75%, 0.9375%, 1.25%, and 1.88%). d, change ratio of PPR and increase in peak PSC and SW under a dynamic tensile strain of 1.25%. e, PSC with repetitive pulses around 100 s (pulse width of 0.1 s and amplitude of -10 V) before and after 10,000 and 100,000 cycles of exposure to a tensile strain of 1.25%. The cyclic bending test results indicate that there is a significant difference in the I PSC values between initial state and after compressive bending of 10,000 cycles (Supplementary Figure 9e). The I PSC values after 10,000 or 100,000 cycles are not much different even the number of bending cycles increases more than that between the initial and 10,000, which indicate that there is a stabilization stage similarly to other organic flexible devices. 1-5 AiS-TSO are not mechanically stretchable but flexible. There are some limitations in mimicking deformability of the skin for applications in electronic skin or soft robotics but its flexibility has many advantages compared to rigid devices. Here we touched the AiS-TSO varying the humidity condition with around 1 kPa pressure. As shown in the Supplementary Figure 19, the response was decreasing with humidity increasing. This phenomenon can be explained the humidity effect on triboelectricity. Since the polyimide film is hydrophilic, so water absorption is high when it is in the high humidity condition increasing surface conductivity from water layer 13 . The higher surface conductivity discharges the surface decreasing the effective triboelectric charges transfer between polyimide film and skin 14 . Here, we measured the memory strength of AiS-TSO in an order of touch, pixel 214(3) while the pixel 3 was skipped (not touched). The measurements were repeated three times. All three experiments showed the same tendency that the untouched pixel 3 has the smallest memory strength and the memory strength was increased from the firstly touched (pixel 2) to lastly touched device (pixel 4). The analogy between an MCNC and AiS-TSO is explained for each step of signal propagation described in Step (i): Tactile stimulation is applied.

Supplementary
Step (ii): In an MCNC, Merkel cells, which are mechanical sensory epithelial cells that have mechanically activated Piezo-2 ion channels, activate a SA-I neuronal afferent. Upon tactile stimulation, Ca 2 + enters the cell via Piezo-2 ion channels and increases membrane potential 15,16 . In our AiS-TSO, tactile stimulation generates a triboelectric potential due to influx of electrons from triboelectrification between the finger and receptive polyimide layer, which corresponds to an artificial Merkel cell membrane. Similar to the MCNC, the AiS-TSO mechanism uses mechanical energy to generate a potential without any external energy.
Step (iii): Merkel cells have a high membrane resistance; therefore, the Ca 2+ ion current generated by influx of Ca 2+ through the Piezo-2 channels produces a large sustained depolarization of the Merkel cell 17 . The ability to produce sustained depolarization via Piezo-2 channels is responsible for the two-receptor-site mechanism where inactivation of the Piezo-2 channel results in SA firing 17,18 . By analogy, the receptive polyimide layer of our AiS-TSO, which corresponds to an artificial Merkel cell membrane, has an ability to store electrons from triboelectrification and therefore functions as a capacitor to generate a SA output signal.
The rate of increase in potential of the polyimide layer decreases gradually, which means that the response to the stimulus gradually decreases, and SA reception is achieved.
Step (iv): In an MCNC, Merkel cells act as pre-synaptic neurons and release serotonic neurotransmitters to activate receptors of A afferents for excitatory signals 19 . Similarly, the potential generated in the receptive polyimide layer of the AiS-TSO switches the dipoles of the BT NP/P(VDF-TrFE) nanocomposite ferroelectric layer above the coercive field. Polarization changes due to dipole switching in the AiS-TSO correspond to neurotransmitter release in the MCNC.
Step (v): In an MCNC, the amount of neurotransmitter that accumulates depends on the degree of tactile stimulation, resulting in synaptic plasticity, which, in turn causes generation of an action potential in the A afferent to transduce SA-I signaling 18,19 . In our AiS-TSO, the amount of permanent polarization depends on the degree of stimulation as this modulates the drain current, i.e., post synaptic current (I PSC ) and can be considered the synaptic weight. Indeed, the retention time of synaptic weight in a biological MCNC is much shorter than that in the AiS-TSO although the exact principles of sensory memory function in an MCNC have not yet been elucidated. In our AiS-TSO, short-term and long-term plasticity (STP and LTP) can form because of the ferroelectric properties of the gate dielectric layer, which makes it possible to create short- which are related to the SRDP and SDDP of biological synapses, respectively.

Supplementary Note 3
To demonstrate that ferroelectric characteristics of gate dielectric layer in Fe-OFET mainly contribute to synaptic properties of the device, we also fabricated the OFET with non-ferroelectric PVP (polyvinylpyrrolidone) as a gate dielectric layer. The measured capacitance of BT NP(20wt%)/P(VDF-TrFE) was much higher (~21 nF/cm 2 , dielectric constant = 13.81) than PVP (~5.2 nF/cm 2 , dielectric constant = 3.25) in the MIM structure with the same insulator thicknesses as those in the FET structure, which results in the higher on-state current level in the transfer characteristics (Supplementary Figure 3a) in Fe-OFET with ferroelectric nanocomposite than that in OFET device with PVP device. The change of PSC (∆PSC) of the device with PVP (~1x10 -11 A) during V rec biasing was much smaller than that of the device with BT NP(20 wt%)/P(VDF-TrFE) (~1x10 -9 A) during V rec biasing (Supplementary Figure 3b). However, Fe-OFET with ferroelectric gate dielectric material has larger hysteresis in the transfer curve (Supplementary Figure 3a) and larger change in PSC after finishing V rec pulsing (Supplementary Figure 3b) compared to OFET with PVP gate insulator. These results are originated from internal field generated in partial polarization switching in ferroelectric material, which results in the generation of synaptic weight (SW) in the Fe-OFET.
On the other hand, negligible hysteresis and residual ∆PSC after finishing V rec pulsing in the OFET with non-ferroelectric PVP indicate that charge trapping does not significantly affect synaptic properties as much as ferroelectric effect event though charge trapping of organic semiconductors has been reported. 20,21 On the other hand, it can be argued that synaptic property in our Fe-OFET device is mainly related to ferroelectricity of nanocomposite gate dielectric layer.

Supplementary Note 4
To demonstrate the repeatability of Fe-OFET as a synaptic device, we applied the V rec of increasing and decreasing amplitude of pulses and number of pulses. As shown in Supplementary Figure 6, PSC response was almost the same when we applied the V rec in increasing or decreasing amplitude and pulse number.
Since we applied each pulse after full recovery to the state at the previous pulse, PSCs were only minimally affected by the previously formed polarization. Therefore, we found that if we need the repeatability of the device, we can apply the pulse after the full recovery. Here the repeatability in SW was good when we increase and decrease the amplitude of pulses and the number of pulses.

Supplementary Note 5
To However, the coercive field value was not much different from those in the previous reports and the tendency of increasing in polarization was observed when the BT NPs are included in nanocomposite, similarly to the results reported from the previous reports [22][23][24][25][26] . Also, when the range of applied voltage was varied to a smaller range, there was observed partial polarization in minor loop. We could demonstrate that usage of smaller range of voltage such as -10 V to generate SW by controlling the partial polarization with pulse rate, number and duration time varied. Therefore, polarization switching can be controlled according to the range of applied voltage 10,11,[27][28][29] , which implies that SW in Fe-OFET will depend on amplitude, duration time, rate and number of the V rec pulses similar to other Fe-RAM devices using minor loop of ferroelectric materials. 23,[30][31][32][33][34] Of course, the device has smaller retention time for polarization of minor loop [30][31][32][33] . However, the retention time can be controlled by pulse duration time, number or frequency for generating SW and controlling the STP and LTP behaviors, as shown in Fig. 2.

Supplementary Note 6
In Supplementary Figure 8, we checked the PSC values when applying -30 V of V rec which is above the coercive voltage of gate insulating layer in Fe-OFET. Comparing with the results obtained with -10 V of V rec , the results show the fast saturation in PSC but poor characteristics in terms of linearity of PSC increasing rate. On the other hand, in case of retention time, the time taken to drop below 15% of the maximum PSC change at -30 V was longer (~1814 min) than that at -10 V (~68 min). Also, the maximum PSC increasing rate was also about 2 times at -30 V of V rec (~18) larger than that at -10 V of V rec (~11). This is related to remnant polarization formed by applying bias pulses above coercive field and, therefore, much longer retention time was obtained compared to that by applying bias pulses in minor loops of P-E curve.
But as we mentioned before, we can generate and adjust SW even with -10 V of V rec pulses. Therefore, we judged that it was not necessary to apply a voltage above the coercive field, which would decrease the linearity of the PSC increasing rate and the SW window, and cause the fatigue in ferroelectric material resulting in device breakdown. Also, we used the triboelectric-capacitive coupling effect by touching the polyimide substrate to generating of SW on AiS-TSO, and the measurement of synaptic characteristics of Fe-OFET was conducted as electrical demonstration of AiS-TSO. We could not generate triboelectric voltage output above coercive field by touching. So, we determined that using -10 V magnitude of V rec is appropriate to demonstrate the generation of the synaptic weight originated from ferroelectric dielectric layer. Furthermore, we focus on the fact that our sensory organs have synaptic-like functions that can be used for sensory memory before they are processed in the brain, which are not intended to implement semipermanent memory. Therefore, we believe that using -10 V of V rec in the minor loop region is enough to demonstrate synaptic function by mimicking the Merkel cell neurite complex. LTP in synapse is very broad [35][36][37] , ranging from minutes to decades, and we can control the STP (seconds) and LTP (hours) of this device by controlling the duration, number and frequency of stimuli.

Supplementary Note 7
I Supplementary Figure 9, we compared PSC values for the device with poled ferroelectric gate dielectric layer. We conducted poling process by applying the bias of -30 V between gate electrode and drain electrode for 30 min. The channel layer acts as a poling electrode. As shown in Supplementary Figure 9a, output characteristics showed increased drain current without saturation, which is different from saturation behavior of unpoled device. No saturation in the poled device is attributed to internal field generated by remnant polarization of gate insulating layer. This result corresponds to our previous investigation about poled Fe-OFET. 5,38 The generated internal field in the poled device acts as negative bias, which enhances accumulation of holes in p-type organic semiconductor channel (Supplementary Figure 9b). PSC change and synaptic properties with V rec pulsing were much different for the unpoled and poled device. During pulsing of -10 V V rec is applied to the poled device, the change of PSC is negligibly small since the dipoles in the ferroelectric layer are already aligned (Supplementary Figure 9c). So, the dipoles are difficult to be switched further in the same direction with negative V rec pulsing since they are already fully switched by gate biasing of -30 V. Therefore, the synaptic weight of the poled device under negative V rec pulsing is negligible because the polarization was already saturated by poling process. During positive V rec pulsing, the change of PSC in poled device is much smaller than that in unpoled device. The positive V rec pulsing will try to rotate dipoles in the opposite direction to the poled direction of the poled ferroelectric layer resulting in a slight decrease in the PSC because partial switching of dipoles can be more difficult by applying the field opposite to aligned direction of dipoles compared to switching of randomly oriented dipoles (Supplementary Figure 9d). Under positive V rec pulsing of the poled device, therefore, a smaller synaptic weight value than that of the unpoled device was also observed. In our approach, partial polarization in ferroelectric gate insulating layer in the unpoled device is used for generating synaptic weight by applying negative or positive V rec pulses. The partial polarization behavior is closely related to mechanism of ferroelectric memory devices. 4,23-27 Therefore, we conclude that the poling process was disadvantageous in generating synaptic weights with potentiation and depression. In our work, the unpoled devices were used for all other measurements to utilize the change in partial polarization as synaptic plasticity.

Supplementary Note 8
When the device was touched with a bare hand, gloved hand, aluminum foil, stainless steel foil, or label tape paper, the PSC increased due to the electron affinity of polyimide is higher than that of these materials. This is because triboelectric charges (electrons) accumulated on the polyimide, which generated a negative potential on the polyimide and, in turn, an excitatory PSC [44][45][46][47] . When the device was touched with PEN or PVC film, triboelectric electrons moved from the polyimide to the touching materials, generating a positive potential on the polyimide and, in turn, an inhibitory PSC [44][45][46][47] . Furthermore, when we touched the device by polyimide film, the device response and change of PSC were very small because the triboelectric effect was the smallest (Supplementary Figure 15). This phenomenon indicates that triboelectric-capacitive coupling is the dominant mechanism in AiS-TSO.

Supplementary Note 9
For further investigation about the mechanism of our AiS-TSO, we characterized the response of PSC depending on touching object, temperature and bending strain. First, we touched the AiS-TSO with polyimide (PI) film and finger varying the temperature. As shown in Supplementary Figure 17a, the PSC change (ΔI PSC / I PSC , i )of AiS-TSO to finger touch was much larger by three orders of magnitude than that to touching with PI at the same temperature (~27 o C). In both cases, the ΔI PSC / I PSC , i was increased with the temperature increased. These results indicate that the main mechanism of AiS-TSO is triboelectriccapacitive coupling effect even though there is a slight change in the ΔI PSC / I PSC , i with the temperature increased.
For further investigation effect of temperature change on the response of PSC, we fabricated the OFET device using PVP (polyvinylpyrrolidone) as gate dielectric layer which has no pyroelectricity compared its response to the AiS-TSO. As shown in Supplementary Figure 17b, when we touch the both devices by finger using BT NP(20wt%)/P(VDF-TrFE) and PVP gate dielectrics, we could observe the increase of PSC (I PSC ) in both devices with the temperature increased. From those results, it can be confirmed that the response of AiS-TSO is not originated from pyroelectricity. Since the OFET with PVP gate dielectric shows an increase in the I PSC with the temperature increased, an increase in I PSC may be attributed to increase in channel conductance due to thermal generation of carriers. Furthermore, pyroelectric effect is expected to be negligible because we didn't carry out any poling process for generating pyroelectricity. From those data, we could confirm that the pyroelectric effect in AiS-TSO is negligible compared to triboelectric-capacitive effect which is the main mechanism of touch response.
In order to investigate the response of AiS-TSO with BT NP(20 wt%)/P(VDF-TrFE) gate dielectric to mechanical strain, we also measured ΔI PSC / I PSC , i of the device to tensile and compressive bending strains and compared to that to finger touch. As shown in Supplementary Figure 17c,  could not generate synaptic weight. These results also indicate that main mechanism of synaptic weight generation is triboelectric-capacitive coupling effect.

Supplementary Note 10
Firstly, working principle of POSFET is much different from that of AiS-TSO. POSFET touch sensor utilizes the piezoelectric response of the piezoelectric gate dielectric while AiS-TSO does not utilize the piezoelectric effect of the gate dielectric. In POSFET, touch stimuli induce i) the displacement of polarized piezoelectric material and ii) change the electric field in piezoelectric material, which modulates the carrier density in the channel and, in turn, the drain current (Supplementary Figure 18a). Here, the piezoelectric layer upon pressurizing induces change in dipole alignment resulting in change in effective gate electric field [49][50][51] . Therefore, POSFET needs intended poling process for generating the saturated remnant polarization, P r , as much as possible to generate piezoelectric voltage enough to modulate the drain current. In AiS-TSO, on the other hand, i) triboelectrification between skin and polyimide substrate (described as receptive part in the manuscript) generates triboelectric charges and ii) coupled capacitive effect in the receptive part induces the partial dipole switching in ferroelectric material and, in turn, change in the drain current (Supplementary Figure 18b). Therefore, mechanisms of generating and transduction of energy are different, in which the POSFET sensor uses piezoelectric effect while our AiS-TSO uses triboelectric-capacitive coupling effect between skin and receptive part. Those mechanisms have studied by theoretical analyses on triboelectric effect 45,[52][53][54] or tribotronics 45,55,56 although AiS-TSO has uniqueness in intrinsic-synaptic function and structure mimicking MCNCs.
Secondly, the functions of sensor have differences. Both sensors have a common function with energy transducer from mechanical to electrical as a device of mimicking mechanoreceptors, but AiS-TSO adds intrinsic synaptic functions and enhances the functionalities for mimicking mechanoreceptors (mimicking synaptic functions of Merkel cell neurite complex). Differently from POSFET, we do not carry out intended poling process to generate P r in ferroelectric material. Instead, we induce modulation of polarization with touch stimuli causing dipole switching depending on nature of stimuli. Therefore, due to the characteristics of ferroelectric material, the conductivity changes in the channel changes are inherently endowed with information of touch stimuli. In conclusion, we could induce SW through the modulation of polarization switching in ferroelectric layer under varying stimuli of touch resulting in inherent change in post-synaptic current and, in turn, modulation of SW. Thus, AiS-TSO has the advantages of a simple structure and manufacturing process, and unlike other mechanoreceptor-mimetic sensors whose only detection function has been reported, AiS-TSO has an intrinsic synaptic function that mimics the Merkel cell neurite complex.