Multi-Gate Organic Neuron Transistors for Spatiotemporal Information Processing

Transcription

1 Multi-Gate Organic Neuron Transistors for Spatiotemporal Information Processing Chuan Qian, 1 Ling-an Kong 1, Junliang Yang 1, Yongli Gao 1,2, *, and Jia Sun, 1, * 1 Hunan Key Laboratory for Super Microstructure and Ultrafast Process, School of Physics and Electronics, Central South University, Changsha, Hunan , P. R. China 2 Department of Physics and Astronomy, University of Rochester, Rochester, NY 14627, USA Corresponding Author * ygao@csu.edu.cn (Y.L. Gao), jiasun@csu.edu.cn (J. Sun) 1

2 ABSTRACT Due to similar transmission characteristics with biological synaptic activities, neuromorphic behaviors simulated by organic electrochemical transistors (OECTs) is of great interest. In this letter, the fabrication and performance of multi-gate P3HT ECTs with ion-gel gating are reported. The neuromorphic behaviors, such as dendrite correlated excitatory postsynaptic current (EPSC), paired pulse facilitation (PPF) and modulation, were simulated in the OECTs. These behaviors were observed to depend on the degree of temporal correlation and distance between in-plane-gate and channel. More importantly, by using dendritic integration from two different gates, spatiotemporally correlated outputs were also emulated. The spatial orientations of the input pulse are defined and changing the orientation will result in a change of the EPSC amplitude. Our results provide a way to construct spatiotemporally neural network based on multi-gate OECTs. Keyword: organic semiconductors; electrochemical transistors; neuron devices; spatiotemporal synaptic response; orientation tuning. 2

3 Organic semiconductors offer unique features such as low cost, flexibility, and large-area process properties, and many electronic devices based on organic semiconductors have been developed, including organic light-emitting diodes, organic photovoltaics, and organic field-effect transistors. 1-6 In recent years, organic electrochemical transistors (OECTs) have attracted considerable interest in bioelectronics and sensing. 7, 8 OECTs are organic transistor devices that use an electrolyte as the dielectric material, and form extremely high specific capacitance of the electric-double-layer (EDL), typically larger than 1 F/cm 2 9, 10. As a result, OECTs can be operated at very low voltages (typically below 2.0 V) to derive a large number of charges in the channel. It is known that the synaptic activities are achieved via various ionic fluxes, 11, 12 and the OECTs transmit a signal based on the motion of ions. 13 It is therefore possible that the OECT can be used for emulating biological synaptic behaviors due to their similar transmission characteristics. In this context, a few endeavors have been reported that utilized OECTs in artificial synaptic devices Recently, Malliaras and his group have firstly demonstrated the fabrication of OECTs based artificial synapses by using liquid electrolyte gating. 14 Importantly, the function of spatial synaptic mapping and integration were achieved in these devices. 15 However, with liquid electrolyte, the electrolyte-ion-polarization-induced leakage current could distort the drain current in channel. Wentao Xu et al have reported organic nanowire synaptic transistors with ultralow energy consumption. 16 Most recently, our group has demonstrated that solid-state ion-gel laterally gated OECTs can be used for the simulation of biological 3

4 synapses. Although some important behaviors are successfully mimicked in artificial synapses, such as excitatory postsynaptic current (EPSC), paired pulse facilitation (PPF), synaptic filter and synaptic plasticity etc, an essential aspect of neuron network is still untouched that many synaptic inputs are transformed from dendrites. So, it is important to study spatiotemporal synaptic response and extract their global properties and configuration. In this letter, we demonstrate the fabrication of organic neuron transistors with a multi-gate dendritic input array. EPSC is shown to be dependent on the degree of temporal correlation and distance between the in-plane-gate and channel. More importantly, spatiotemporally correlated outputs were also emulated by using dendritic integration. The spatial orientations are defined by the geometry of dendritic array and changing the orientation will result in a change of the EPSC amplitude. 4

5 Figure 1. (a) Simple schematic picture of a biological synaptic integration in a neuron. (b) Schematic illustrations of the P3HT-based neuron transistor with a multi-in-plane-gate structure. (c) EPSC triggered by six pre-synaptic spikes (-2.0 V, 90 ms) applied at the gate electrode 1 are shown as a function of time. Shown in Figure 1a is a simple picture of a biological synaptic integration in a neuron. 17 The pre-synaptic neuron can transmit many synaptic input signals from different dendrite to a synapse, which occurs via the motion of chemical ions from the pre-neuron to the post-neuron. Figure 1b displays the device structure of an ion-gel coupling P3HT neuron transistor with a multi-in-plane-gate structure on the 200 nm-sio 2 /Si (Si-Mat, Silicon Materials) substrate. The multi-dendritic input structure of neuron is thought to be very useful for enriching information transmission and computational ability. To mimic neuromorphic functions, the P3HT is considered as post-neuron, ion-gel as pre-neuron, in-plane Au electrodes as dendrite and conductance of channel as the synaptic weight. Polymer P3HT is purchased from Sigma-Aldrich, which was dissolved in dichlorobenzene (15 mg/ml). After being spin-coated, P3HT films were cured at 50 C for 2 hours. Thermally deposited Au electrodes through a micrometer sized carbon fiber shadow mask on the top of the P3HT film were used as the source, drain, and in-plane-gate, which define a channel with a length of 6 m and a width of 200 m. The preparation of the ion-gel films (P(VDF-HFP) + [EMI][TFSA]) is followed the previous method and the gels were then transferred to the device as the gate dielectric by employing cut and stick processes. 18 Electrical performance of the transistors and neuromorphic behaviors 5

6 were measured by a semiconductor parameter characterization system (Keithley 4200 SCS) in air and at room temperature. For organic electrochemical transistor, under gate electric field, the drain current is controlled by the injection of ions from an electrolyte into a semiconductor channel. 13 The transfer and output characteristic curves of the device are shown in Figure S1. When the gate voltage changed from 2.0 V to -2.0 V with a low source-drain voltage of -1.0 V, the I ds is increasing from A to A. To represent the synaptic strength in our organic neuromorphic devices, a potential spike signal is applied on the arbitrary gate electrode, which can trigger the motion of [TFSA] - ions from ion-gel to P3HT film and a peak EPSC in channel. 13 Figure 1c shows the typical temporal response of the organic neuron transistors to six -2.0 V pre-synaptic spikes (interval is about 3 s) applied on G 1 electrode with 100 ms duration. Because the devices are reliability and the P3HT layer is reversibly de-doped to its initial state under gate voltage back to 2.0 V, the peak EPSC is relatively stable, the average value of six peak EPSCs is about 7.89 A. All the synaptic simulations are measured with a constant potential V ds = -0.5 V applied on the source-drain electrodes. EPSCs recorded in response to the train of pulses (interval is about 50 ms) are shown in Figure S2. When the interval is shortened to 50 ms, the short-term plasticity (STP) and self-tuning are reproduced, which are discussed in biological system As the pulses applied, the peak EPSC is initially hypersensitive and subsequently adapts to the stimulation. After the end of the train of pulses, the EPSC back to the initial level. 6

7 Figure 2. (a) Spatiotemporal EPSC response in P3HT-based neuron transistors. A 3-D-scatter plot describing the EPSCs as a function of pulse duration and the distance between in-plane-gate and channel (dendrite to post-synapse). (b) Paired-pulse facilitation ratio as a function of interval t) and the distance between in-plane-gate and channel. The insert in (b) shows EPSCs triggered by paired pulses applied at the gate electrode 1 with pulse t of 50 ms. In a brain, spatiotemporal synaptic response in cortical neurons is a significant part in the formation of dynamic cell assemblies. 22 Information processing is spatiotemporal and the neuron network is very complicated, in which many synaptic 7

8 inputs are transformed from dendrites of pre-neuron to post-neuron. The outputs are sensitive to both the timing and location of inputs. 23, 24 The spike pulse duration and distances between in-plane-gate and channel (dendrite to post-synapse) will both contribute to a change in EPSC. Spatiotemporal synaptic response in P3HT-based neuron transistors is shown in Figure 2a, which is a 3-D-scatter plot describing the EPSCs as a function of pulse duration of 30 to 100 ms and the dendrite to post-synapse distance of 1.08 to 2.64 mm. Each point is the average value of six peak EPSCs. The point with 100 ms duration and 1.08 mm distance between dendrite and post-synapse is extracted from Figure 1c. With the duration increased, more and more ions move onto the interface of ion-gel/semiconductor, 13 and the average peak EPSC is increased from 1.79 A to 7.89 A. Moreover, with 100 ms duration and 2.64 mm distance, the average peak EPSC is just 2.33 A, 3.4 times smaller than the value with 100 ms duration and 1.08 mm distance. It is known that the ion-gel electrolyte resistivity and the area covered by ion-gel are constant values. If the gate-to-channel distance is larger, fewer ions will diffuse to the ion-gel/p3ht interface and the I ds in the channel will be relatively small. 15 The temporal response depends on the distance between in-plane-gate and channel. Increasing the distance from 1.08 to 2.64 mm decreased the slope of the spatiotemporal synaptic integration curve from about A/ms to A/ms, thus to evoke a large response, a longer pulse duration time is needed. The EPSCs depend both on spatial and temporal characteristics of dendritic inputs, which is important for neural network construction. 8

9 We emulated the paired-pulse facilitation (PPF) with the in-plane-gate and channel distance in the organic neuron transistors. The behavior of PPF is a kind of STP and very important for neural activities. 25, 26 The insert in Figure 2b shows EPSCs triggered by paired pulses applied at the gate electrode 1 with pulse t of 50 ms and the channel current is measured as a function of time. A 1 and A 2 are the peak EPSCs of the first and second spikes, respectively. A 2 is facilitated by the residual ions triggered by the first spike. The PPF ratio of two amplitudes (A 2 /A 1 ) as a function of interval t) and the dendrite to post-synapse distance is shown in Figure 2b. The pulse t and the dendrite to post-synapse distance for PPF measurement were ranged from 50 to 800 ms and 1.08 to 2.64 mm. A 2 /A 1 shows a distance-dependent weight, especially when the t is smaller than 200 ms. When t = 50 ms, the A 2 /A 1 reaches the top point in each line. The maximum value of 149.2% is obtained when distance is 1.08 mm. When the distance is increased to 2.64 mm, the value is reduced to 117.1%. With a nearer location away from channel, the residual ions triggered by dendrite in a short time are much more. With interval lengthened to 800 ms, the A 2 is almost equal to A 1 and most of ions triggered by the first spike are back to the ion-gel for three kinds of distance. 9

10 Figure 3. (a) Modulatory effect of V M and V pulse is plotted as a 3-D-scatter plot. (b) Dendrite to post-synapse distance-dependent EPSCs of the ion-gel gated organic neuron transistors with two in-plane spike inputs applied at the two gate electrodes with same dendrite to post-synapse distance, simultaneously. Under changing conditions, synaptic modulation provides flexibility to operate, which is a ubiquitous phenomenon in physiological systems. 27 The realization of dynamic synaptic behaviors by using modulation voltage (V M ) is very useful for understanding the principle of neural networks and guiding the innovation of neural circuits. 28, 29 Especially, Prof. Ren s group has demonstrated the dynamic synaptic plasticity by using modulation voltage of back gate. 29 For our organic neuron transistors, the output response can be tuned in a wide range by the V M of in-plane 10

11 gate. As showed in Figure 3a, modulatory effect of V M and V pulse is plotted as a 3-D-scatter plot. To mimic the modulation behavior, ten pre-synaptic spikes of V Pulse with duration time of 50 ms are applied on G 1 electrode and the constant modulation voltage (V M ) is applied on G 2 electrode, simultaneously. Each point in Figure 3a is the average of the ten peaks. Without modulation of V M and the V Pulse is set to -1 V, there are limited holes increased in the channel and the average of peak EPSCs is just about 1.9 A; when the V Pulse and V M are both set to -2 V, many ions move onto the interface of ion-gel/semiconductor to modulate the peak EPSC and the value is up to A. Applied more negative voltage on the modulatory terminal, a more obvious modulatory effect of V M and V pulse is observed and the average of peak EPSCs has an about eight times of magnitude increasing. Modulation may reflect a change in the relationship between spatiotemporal synaptic input and output. The influences of distance between in-plane gate and channel on the spatiotemporally correlated outputs from two different dendritic positions were also studied. Schematic image is shown in Figure S3: the two inputs are applied at the two gate electrodes with the same dendrite to post-synapse distance, simultaneously. Figure 3b shows the distance correlated average peak EPSCs of the integration of two in-plane pre-synaptic inputs. Pre-synaptic inputs applied at the gate electrode 1 and 2, the vertical distance between channel and gate electrode 1 or 2 is 1.08 mm, the average peak EPSC is about A; inputs applied at the gate electrode 5 and 6, the vertical distance between channel and gate electrode 5 or 6 is 2.64 mm, the average peak EPSC is about A. The peak EPSC shows a linear relationship with the distance between dendrite and 11

12 post-synapse, which meshes well with experimental finding in Figure 2a with single input. Figure 4. (a) EPSC triggered by eight pre-synaptic spikes (-2.0 V) applied at the gate electrode 0 and 5 are shown versus time with duration of 50 ms. (b) Schematic image of the ion-gel gated organic neuron transistors with two in-plane pre-synaptic inputs applied at the gate electrode 0 and another gate electrode x (x=1~6), simultaneously. The spatial orientations of the input pulse are defined by angle range from 0 to 180. The gate electrode 0 is defined as the common endpoint of each angle, the line between the gate electrode 0 and 5 as the common ray, the line between the gate electrode 0 and x as the other ray. (c) Polar diagram of the EPSCs for different spatial orientations of the input pulse. The neurons in primary visual cortex have strong tuning to a small set of stimuli and the neuronal responses can discriminate small changes in colors, spatial frequencies and visual orientations. 30 In our study, the spatiotemporal dendritic integration with correlated inputs can be illustrated by spatial orientations. Figure 4b shows the schematic image of the ion-gel gated organic neuron transistors with two in-plane pre-synaptic inputs applied at the gate electrode 0 and another gate electrode x (x=1~6), simultaneously. The spatial orientations of the input pulse are defined by angle range from 0 to 180. The gate electrode 0 is defined as the common endpoint of each angle, the line between the gate electrode 0 and 5 as the common ray, the line 12

13 between the gate electrode 0 and x as the other ray. EPSC triggered by eight pre-synaptic spikes (-2.0 V) applied at the gate electrode 0 and 5 are shown versus time with duration of 50 ms in Figure 4a. When the input spikes triggered simultaneously on two dendrites, the EPSC in the channel can be observed and the average value is about 1.38 A. Figure 4c shows the Polar diagram of the EPSCs for different spatial orientations of the input pulse. Each point is the average of the eight peaks and the average values are demonstrated in Table S1. When the orientation angle is 129.8, the maximum tuning response (2.22 A) is obtained. The average peak EPSC is gradually increased with the orientation angle changed from 0 to 50.2, or from 180 to 129.8, which is similar with the orientation tuning curve in the primary visual cortex. 31 The multi-gate organic neuron transistors have a symmetrical arrangement of gate array, but the average peak EPSC triggered simultaneously on the gate electrode 0 and x (2, 4, 6) is larger than the average peak EPSC triggered on the gate electrode 0 and corresponding x (1, 3, 5). The peak EPSC follow the empirical formula of I A/ d, where d is the distance from the gate to drain electrode; A is constant in this comparison, dependent on electrode area, electrolyte resistivity and pre-synaptic spike. 15 Thus increased amplitudes of peak EPSC are measured for the gates, which are closer to the drain electrode. In summary, the fabrication and performance of artificial neuron transistors based on multi-gate OECTs with ion-gel dielectrics are reported. The neuromorphic behaviors are dependent on the degree of temporal correlation and distance between in-plane-gate and channel. Furthermore, by using dendritic integration from two 13

14 different gates, spatiotemporally correlated inputs outputs from two different gates were also emulated. The spatial orientations of the input pulse are defined and changing the orientation will result in a change of the EPSC amplitude. Our results provide a way to construct complicated neural network based on spatiotemporal correlated multi-gate OECTs. See supplementary material for transfer and output characteristic curves of P3HT-based OFET, related characterizations, and setup. Acknowledgements This work was supported by the National Natural Science Foundation of China ( , ) and the Fundamental Research Funds for the Central Universities of Central South University (2016zzts018). Y. G. acknowledges support by National Science Foundation DMR and CBET

15 REFERENCES 1. S. Reineke, F. Lindner, G. Schwartz, N. Seidler, K. Walzer, B. Lüssem and K. Leo, Nature 459, 234 (2009). 2. W. Lee, D. Kim, J. Rivnay, N. Matsuhisa, T. Lonjaret, T. Yokota, H. Yawo, M. Sekino, G. G. Malliaras and T. Someya, Adv. Mater. 28, 9722 (2016). 3. C. Qian, J. Sun, L. Zhang, H. Huang, J. Yang and Y. Gao, J. Phys. Chem. C 119, (2015). 4. C. Qian, J. Sun, L. Zhang, H. Xie, H. Huang, J. Yang and Y. Gao, Synth. Met. 210, 336 (2015). 5. J. Xiong, B. Yang, J. Yuan, L. Fan, X. Hu, H. Xie, L. Lyu, R. Cui, Y. Zou, C. Zhou, D. Niu, Y. Gao and J. Yang, Org. Electron. 17, 253 (2015). 6. W. L. Huang, B. C. Yang, J. Sun, B. Liu, J. L. Yang, Y. P. Zou, J. Xiong, C. H. Zhou and Y. L. Gao, Org. Electron. 15, 1050 (2014). 7. S. P. White, K. D. Dorfman and C. D. Frisbie, Anal. Chem. 87, 1861 (2015). 8. S. P. White, S. Sreevatsan, C. D. Frisbie and K. D. Dorfman, ACS Sensors 1, 1213 (2016). 9. O. Larsson, E. Said, M. Berggren and X. Crispin, Adv. Funct. Mater. 19, 3334 (2009). 10. S. H. Kim, K. Hong, W. Xie, K. H. Lee, S. Zhang, T. P. Lodge and C. D. Frisbie, Adv. Mater. 25, 1822 (2013). 11. R. S. Zucker, Neurosci. 12, 13 (1989). 12. L. A. Kong, J. Sun, C. Qian, G. Gou, Y. He, J. Yang and Y. Gao, Org. Electron. 39, 64 (2016). 13. C. Qian, J. Sun, L.-a. Kong, G. Gou, J. Yang, J. He, Y. Gao and Q. Wan, ACS Appl. Mater. Interfaces 8, (2016). 14. P. Gkoupidenis, N. Schaefer, B. Garlan and G. G. Malliaras, Adv. Mater. 27, 7176 (2015). 15. P. Gkoupidenis, D. A. Koutsouras, T. Lonjaret, J. A. Fairfield and G. G. Malliaras, Sci. Rep. 6, (2016). 16. W. Xu, S. Y. Min, H. Hwang and T. W. Lee, Sci. Adv. 2, e (2016). 17. D. Kuzum, S. Yu and H. P. Wong, Nanotechnology 24, (2013). 18. K. H. Lee, M. S. Kang, S. Zhang, Y. Gu, T. P. Lodge and C. D. Frisbie, Adv. Mater. 24, 4457 (2012). 19. M. V. Tsodyks and H. Markram, Proc. Natl. Acad. Sci. USA 94, 719 (1997). 20. J. A. Varela, K. Sen, J. Gibson, J. Fost, L. Abbott and S. B. Nelson, J. Neurosci. 17, 7926 (1997). 21. G. G. Turrigiano, Cell 135, 422 (2008). 22. R. Azouz, J. Neurophysiol. 94, 2785 (2005). 23. G. Ariav, A. Polsky and J. Schiller, J. Neurosci. 23, 7750 (2003). 24. A. Polsky, B. W. Mel and J. Schiller, Nat. Neurosci. 7, 621 (2004). 25. D. V. Buonomano, J. Neurosci. 20, 1129 (2000). 26. D. V. Buonomano and W. Maass, Nat. Rev. Neurosci. 10, 113 (2009). 27. S. Martin, P. Grimwood and R. Morris, Annu. Rev. Neurosci. 23, 649 (2000). 28. C. Wan, L. Q. Zhu, Y. H. Liu, P. Feng, L. P. Liu, H. L. Cao, P. Xiao, Y. Shi and Q. Wan, Ad. Mater. 28, 3557 (2016). 29. H. Tian, W. Mi, X. Wang, H. Zhao, Q. Xie, C. Li, Y. Li, Y. Yang, and T. Ren, Nano Lett. 15, 8013 (2015). 30. D. H. Hubel, T. N. Wiesel and M. P. Stryker, J. Comp. Neurol. 177, 361 (1978). 31. R. Shapley, M. Hawken and D. L. Ringach, Neuron 38, 689 (2003). 15

16

17

18

19