Open Access
Add to BibSonomy
Abstract
In the post-Moore era, the gradually saturated computational capability of conventional digital computers showing the opposite trend as the exponentially increasing data volumes imperatively required a platform or technology to break this bottleneck. Brain-inspired neuromorphic computing promises to inherently improve the efficiency of information processing and computation by means of the highly parallel hardware architecture to reduce global data transmission. Here, we demonstrate a compact device technology based on the barrier asymmetry to achieve zero-consumption self-powered synaptic devices. In order to tune the device behaviors, the typical chemical doping is used to tailor the asymmetry for energy harvesting. Finally, in our demonstrated devices, the open-circuit voltage (VOC) and power-conversion efficiency (PCE) can be modulated up to 0.77 V and 6%, respectively. Optimized photovoltaic features affords synaptic devices with an outstanding programming weight states, involving training facilitation, stimulus reinforce and consolidation. Based on self-powered system, this work further presents a highly available modulation scheme, which achieves excellent device behaviors while ensuring the zero-energy consumption.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
According to Moore's Law, the continuous reduction of transistor size can bring exponential improvement in computer efficiency [1]. Based on high-performance computer, the artificial neural network (ANN) algorithm can integrate and process massive information via the simulated mode of human brain, where the information is intelligently recognized and classified by the functional node connection and dynamic adjustment of internodal connection weight [2]. However, the software implementation of ANN still runs on the standard von Neumann architecture of separated logic and storage units, which can cause a lot of undesired efficiency redundancy [3,4]. The efficiency bottleneck will emerge with the exponential growth of information volumes. Inspired by the biological synaptic processing information in analog parallel mode [1,4], the developments of artificial synaptic devices with spatiotemporally concurrent learning and memory units have led to intense research. Like the human brain containing approximately 1015 synapses, neuromorphic computing demonstrated in hardware platforms needs integrate large-scale artificial synaptic devices to adaptively cognize and process massive information from the outside. Hence, building a simple, ultralow-energy artificial synapse device is an important step to realize the highly complex neuromorphic computing, that is considered as one of the most promising technologies to promote the development of computing system in the post-Moore era.
Thanks to the parallelism improvement of underlying hardware operation, the implementations of artificial synaptic device have demonstrated that the neuromorphic computing gains a great achievement in energy consumption compared to the digital computing [3,5]. In the existing literature it is shown that the elaborate design for device’s structure and working principle can typically open up the tantalizing opportunity to build ultralow-energy plastic synapse analog. Liu et al. demonstrated a monolayer MoS2/MoS2-xOy junction-based four-terminal artificial synapse that using electric or light gating to change the Schottky barrier and dynamically tune that by interface-based behaviors [6]. Demonstrated devices exhibited a low energy consumption of below 20 pJ per synaptic event, which is over 40 times lower than that of conventional complementary metal oxide semiconductor (CMOS) [7]. Xiong et al. demonstrated an electrochemical graphene-based synaptic device, through the Li+ concentration between layers in graphene to tune graphene conductance as analog weight, while maintaining a good energy efficiency of below 500 fJ per synaptic event [8]. The lateral symmetric back-to-back Schottky junctions (B-B SJ) based on molecular level-thick crystals have recently been demonstrated to function as a low-power optical synaptic device in our works [9]. Nanoscale distance between Schottky contact and trapping interface allows device for sensitive Schottky barrier lowering that accompanies the SiO2 interfacial trapping-based persistent photoconductivity. Furthermore, such free-standing physical interfaces responsible for resistance switching and history storage can allow the demonstration of a highly designable and tunable structure concept. Despite these achievements, most devices so far are orders of magnitude more consumption than the human brain (roughly 10 fJ per synaptic event [10]), and thus, the desired ultralow energy consumption synaptic device remaining challenges should be highlighted.
Here, we show how to obtain high-performance self-powered synapse based on B-B SJ in response to challenges in implementing low-powered neuromorphic devices. We specifically designed the asymmetric electrode pair upon B-B SJ to realize the self-powered behaviors through a needed built-in electric field among asymmetric Schottky contacts. Clearly, the analog functions were largely controlled by built-in electric field. Considering this, the doping modulation was used for increasing the work function (WF) difference between electrodes, and hence a larger field asymmetry at both contacts. The results revealed that the device performance could be dramatically improved by increasing the asymmetry of the contacts, for example, the crucial photovoltaic parameter of PCE could strikingly be increased up to 6%. In addition, it is clear that massive charge trapping inevitably occurred at oxide surface during the exciton diffusion, which is suggested as the intrinsic memristive mechanism. So, the key to improve the synaptic characteristics, including switching and retention, is increasing built-in electric field by doping modulation. Finally, the asymmetric B-B SJ device with a suitable doping condition has enabled the flexible synaptic plasticity upon self-powered mode, including short-term potentiation (STP), short-term-to-long-term forgetting and continuously updated weight states (>10 states). Clearly, such work demonstrates how to obtain a high-performance self-powered neuromorphic device which is a key step towards large-scale neuromorphic network hardware, not only as a zero-consumption processing unit, but also for greatly simplifying the external circuits in hardware.
2. Results and discussion
2.1 Device structure and film characteristics
In the biological visual-perception system, the rod-synapse acting as light-sensitive elements transduces optical information into neural impulses and, transmits them by weighting the synaptic connection between neuronal elements (pre- and post-synapse). In part, these neurobehaviors generally involves the changes in the quantity of the neurotransmitters to and from those neuronal elements [3,11]. To mimic this concept, we structured a lateral two-terminal device architecture of B-B SJ comprising two asymmetric electrodes (Dioctylbenzothienobenzothiophene (C8-BTBT) and poly(3,4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS)), semiconductor layer and SiO2/Si substrate (see Methods for detailed process). The PEDOT:PSS film electrode showed unique suitability in this device based on not only the high conductivity and transparency, but also highly tunable work function (WF). Schematic of a biological visual synapse and our asymmetric B-B SJ optical synapse are illustrated in Fig. 1(a), where the presynaptic membrane, Au/C8-BTBT Schottky interface, connects with a postsynaptic membrane of PEDOT:PSS/C8-BTBT Schottky interface via C8-BTBT channel. The interfacial conductance represents the synaptic weight, which can be controlled by photogenerated excitons from semiconductor photoexcited mode, representing optical signals transducing and transmitting. In this device, the two-layered (2 L) molecular crystals of C8-BTBT is preferentially considered, as the two-layered-based device has been proved to display more remarkable light-modulated Schottky barrier and flexible plasticity than in the case of other-layered-based devices in previous symmetric B-B SJ synapse. In Fig. 1(b), the optical micrograph reveals the highly uniform and precisely-layer defined features of grown C8-BTBT films over micrometer scale. The atomic force microscopy (AFM) demonstrates the nanoscale surface morphology of the grown films. The heights of first and second layers correspond to the ∼ 2.3 nm and ∼ 3.0 nm, roughly a molecular length, further the surface morphology displays an atomically smooth feature with a root-mean-square (RMS) roughness of ∼ 0.5 nm. To investigate the molecular packing structure, we used high-resolution AFM to precisely visualize molecular alignment, as shown in the right panel of Fig. 1(b), where the 2 L molecules orient with a classic herringbone-type packing with lattice constants of a = 6.50 ± 0.1 Å, b = 8.36 ± 0.2 Å, and θ = 87.7° ± 1.2°. In addition, two remarkable peaks at 2θ = 5.86 and 8.87 in the grazing incident X-ray diffraction (GIXRD) result (Fig. S1 in Supplement 1 supplemental document) are assigned to the reflections of C8-BTBT crystalline planes (002) and (003). These results full demonstrated the tightly packed crystalline characteristics of grown 2 L C8-BTBT films. This highly ordered molecule-stacking structure makes 2 L C8-BTBT film particularly suitable for high-performance electronics by virtue of efficient charge transport properties (the saturated hole mobility up to ∼ 12.7 cm2 V−1 s−1) [9].
Fig. 1. The self-powered mode characteristics of undoped device. (a) Schematic of the device and corresponding visual unit. (b) The morphology characteristics of C8-BTBT crystalline film. The left is optical micrograph, middle is AFM image of bilayer C8-BTBT, and right is High-resolution AFM image of bilayer C8-BTB. (c) Logarithmic I-V characteristics for various optical conditions. (d) Linear I-V characteristics in a very small voltage range closing to zero voltage. Inset shows the enlarged image near the origin. (e) The output power generated by self-powered device versus voltage under two optical conditions.
2.2 Electrical properties and working mechanisms
Figure 1(c) shows the current versus voltage (I-V) curve of asymmetric B-B SJ device based on such molecular crystals displaying a typical rectifying characteristic in the dark condition, which demonstrates the presence of Schottky barrier between electrodes and C8-BTBT. Meanwhile, the slightly left-shifting I-V curves as the increase of illumination strength can also be observed, indicating an optically tunable open-circuit voltage (VOC). The Voc as a critical parameter of photovoltaics verifies that there is a built-in electric field needed to be offset during the voltage scanning and thereby inducing the movement of I-V curve under illumination [12]. From Fig. 1(d and e), we see a reasonable effect of illumination strength on photovoltaic properties, in which the VOC and maximal output power (Pmax) should of course increase with that. The detailed data is summarized in Table 1, in which the calculated PCE is 8 × 10−6 and 2.5 × 10−5 under 162 and 374 µW/cm2, respectively. The low PCE can be explained by the weak built-in electric field resulting from the slight asymmetry. Overall, the results of photovoltaic properties from this asymmetric B-B SJ device demonstrates a self-powered driving option, and yet, more efficient driving performance evidently require a larger barrier asymmetry. Considering this requirement, we have modified the WF of PEDOT:PSS by typical polyethylenimine (PEI) doping to increase the field asymmetry [10,13], thus gaining a global enhancement of synaptic performance.
Table 1. The detailed summary of important photovoltaic-parameter values at various doping concentrations.
The whole output of synaptic behavior of B-B SJ device directly relies on the dynamic modulation of Schottky barrier via optical programming. During self-powered operation, the back-and-forth changing of Fermi level (EF) at each Schottky contact is described as an origin of the charge movement [14], as demonstrated in Fig. 2(a). Notably, the asymmetric electrode pair with different WF can make Schottky contacts different band structures. Due to the mismatch between Fermi level of C8-BTBT (4.26 eV) and WFs of two electrodes (5.2 eV of Au, 5.18 eV of PEDOT:PSS) [15,16], the junction potential will drive the electrons to drift from HOMO of C8-BTBT to each electrode side, and hence resulting in near the C8-BTBT interface lowered the Femi level. During this process, the electrons at C8-BTBT interface are gradually depleted and can thus manipulate the formation of space charge zone to block the electrons at electrode side. Eventually, following the electron accumulation at each electrode side, the edge bands of C8-BTBT will be downward bended to form two high hole barriers (viewed as high-resistance state, HRS), likewise the Femi level also displays downward trend to align with the WF of each side. Due to larger difference between WF of Au and Fermi level of C8-BTBT, more hole accumulation leads to a stronger electric field than PEDOT:PSS side, thus, the lateral built-in electric field in C8-BTBT is formed with a direction of Au side to PEDOT:PSS side.
Fig. 2. The operating mechanism analysis of undoped device. (a) Schematic illustration of band diagram and charge movement during switching the light. (b) The built device model in Silvaco ATLAS software. (c) The electric field distribution across the C8-BTBT layer under two electrodes. (d) The quantized electric field distribution extracted along the white line 1 and 2 in (c).
To better understand switching mechanism, we utilized a modeling technique of Silvaco ATLAS software to demonstrate the electrostatic field inside the C8-BTBT layer at contacts. Figure 2(b) shows the written device structure with some real device parameters, such as materials, geometrical dimension and work functions. As shown in Fig. 2(c), the closer distance to top contact exhibits a stronger electric field, attributing to the residual hole accumulation. Moreover, note that the simulated electric field under Au electrode indeed is stronger than that under PEDOT:PSS electrode, because higher WF can reach equilibrium only carrying out more electron drifting, resembling a larger hole capacitor. By a cutline tool along the white line 1 and 2, the difference in both-side junction fields is observed as shown in Fig. 2(d), thus, the lateral built-in electric field (from Au to PEDOT:PSS) would be the result of this difference. Furthermore, the field still maintains 35000 V/cm2 at 5.3 nm depth (that is the C8-BTBT thickness) that indicates the space charge region going throughout the C8-BTBT layer.
When a 365 nm light is applied, seeing the middle panel of Fig. 2(a), the C8-BTBT absorbs the photon energy to generate the electron-hole pairs, in which the electron will undoubtedly flow to the Au/C8-BTBT interface, while the hole flows in the opposite direction to accumulate at PEDOT:PSS/C8-BTBT interface under built-in electric field. These directionally accumulated photogenerated electron (hole) can thus result in the Fermi level at Au (PEDOT:PSS) side reducing (elevating). At this moment, the junction potential will again acts as a ‘charge pump’ to drive electrons (hole) through the depletion zone interface to Au (PEDOT:PSS) side until reach a new thermodynamically equilibrium, as shown in the right panel of Fig. 2(a). In this process, the barrier width will become thinner and the band of C8-BTBT at PEDOT:PSS side bends towards lower hole energy level (upward bended), and hence lower tunneling and thermionic emission barriers (low-resistance state, LRS). Therefore, the switching is accomplished by a change in barrier, and whole output current is achieved by the overflow of the accumulated photogenerated holes (electrons) to Au (PEDOT:PSS). Besides, the photoconductivity retension was controlled by capacitive charging at SiO2 interface as the distortion-induced trappings of Si-O band were inevitable under various electrostatic factors [17,18]. Therefore, these results demonstrated above full proof the physical accuracy of self-powered memristive mechanism.
2.3 Modulation and optimization
The Schottky barrier is well known to be sensitive to the electrode work function, making the work function modification as an effective approach for the contact electric field modulation [18,19]. We began by using the PEI, a typical PEDOT:PSS’s dopant, to tune its work function, where the amino groups in PEI spontaneously grafted onto the sulfonic acid group in PSS to reduce surface dipole and then the work function [13]. Thus, the measurement of work function should of course be required to estimate the barrier asymmetry. The work function can be calculated using a previously reported measurement method of UPS spectra and an energy Eq. (1) [20]:
where hν is the energy of He(I) irradiation (21.22 eV) and Ecutoff is secondary cutoff edge. The UPS results of PEDOT:PSS films with various doping concentrations are shown in Fig. 3(a), in which the films doped by 0, 1, 1.5, 2.6 wt% display a typical peak-liked distribution of secondary cutoff edge with a relative maximum of 16.54 eV at 1 wt%. The calculated WFs of 5.18, 4.71, 4.66, 5.1 eV corresponds to the 0, 1, 1.5, 2.6 wt% doping, indicative of a likely maximal asymmetry at 1.5 wt%, as shown in the Fig. 3(b). This doping saturation phenomenon was consistent with what have been previously reported in other organic semiconductor system, that is ascribed to the increase in energetic disorder causing the dedoping issues [21]. Seeing Fig. S2 in Supplement 1, the model estimates suggest the completely different junction field at various doped PEDOT:PSS contacts. Then, corresponding field distribution extracted in such devices significantly reveals doping-controlled junction potential as shown in Fig. 3(c). We furthermore compared the lateral field drop at 1 nm distance. The results are shown in Fig. 3(d), from which the changing trend of field difference is evident. Notably, This difference obtained a relatively maximum at 1.5 wt% doping concentration, consistent with the work function estimates. The I-V characteristics of these doped devices are shown in Fig. S3 (seeing Supplement 1), from which several important parameters can be extracted. In Fig. 3(e), the VOC varies with the doping concentrations showing a similar trend as |ΔWF| (given as: |WFPEDOT:PSS-5.2|), just as expected. Another trend that the higher illumination intensity has a larger VOC is also observed. Figure 3(f) shows the calculated PCE vs doping concentration, displaying a similar trend observed in the VOC-modulation (Fig. 3(c)). The detailed photovoltaic-parameter values have been summarized in Table 1. Note that the photovoltaic parameters (e.g. VOC and PCE) are relatively high when the doping concentration is 1.5 wt%. This remains consistent with the larger asymmetry of work function. These results suggested the effective doping modulation of work function for maximizing the energy harvesting. The optimal photovoltaic performance involves a VOC of 0.77 V, Pmax of 8.4 × 10−11, FF of 25% and PCE of 6%.Fig. 3. The modulation of device behaviors by PEI doping. (a) The result of UPS measurements for various doped device. Inset is a partly detailed UPS image. (b) The secondary cutoff edge extracted from (a) and work functions of doped PEDOT:PSS electrodes versus doping concentration. (c) The quantized electric field distribution under electrodes in various doped devices. (d) the lateral field drop at 1 nm distance versus doping concentration extracted from (c). (e,f) The VOC and PCE values for four devices with various doping concentration under two different illumination strength. The |ΔWF| can be given as: $|{\Delta \textrm{WF}} |\textrm{ = W}{\textrm{F}_{\textrm{PEDOT:PSS}}}\textrm{ - 5}\textrm{.2}$.
From an incomplete synaptic test, the relatively good doping concentration could fast be picked out by comparing the facilitation, retention and state number of various doping devices. In the test, all devices display photoresponse but post-stimulation retention is clearly different as shown in Fig. 4(a), and the current noise of devices is very low (∼ 2 pA) (Supplement 1, Fig. S4). This state retention as an inherently key for non-volatile neuromorphic devices plays an indispensable role in history-related programming. So, a long state-retention time is very desirable [22]. Clearly, the retention time reaches a relative maximum in the device doped by 1.5 wt%, and a little reduction is observed at 1 wt% doping, while no doping and 2.6 wt% doping of devices does almost not have retention. Further comparison of plastic functions is made by extracted double-pulses facilitation, as given by B/A, and state-loss ratio, as given by (A-C)/A from this test. In Fig. 4(b), both the facilitation ratio and loss ratio display a similar trend that observed in the post-stimulation retention, where the 1.5 wt% device gains relatively clear state enhancement (∼ 1.18) and low state-loss ratio (∼ 0.3). The most likely mechanism for dynamic modulation of these two key parameters should involve the effects of built-in electric field on charge trapping. The field-assisted exciton separation and diffusion always accompany the massive charge trapping, that means larger built-in electric field has more trapping events. So, these parameters changing trends are consistent with the built-in electric field change. In addition, the number of distinct conductance states as one of the most important criteria towards neuromorphic computing has also been considered. Note that, there is almost no enhanced conductance states in undoped and doped by 2.6 wt% devices, and only a few enhanced states corresponding to first few pulses at 1 wt% doping, instead, the device at 1.5 wt% doping gives continuously updating conductance states over the programming course (> 10 states). Therefore, considering the non-volatile state retention in this test, the device doped by 1.5 wt% is particularly noteworthy and thus undoubtedly required a further simulation scheme.
Fig. 4. The comparison of synaptic behaviors of various doped device. (a) The photo-programming synaptic behaviors comparison among four devices. (b) The facilitation and loss ratio extracted from (a) of four devices.
2.4 Synaptic STP, PPF, and short-term-to-long-term forgetting
The device scheme to change synaptic weight via width or height of Schottky barrier is expected as a vivid emulation of the connection strength between pre- and post-synapse, and state retention can be further manipulated by spatiotemporal coupling of interface trapping. Based on results above, the detailed analog tests have been carried out on the 1.5 wt% doping of devices, including STP and short-term-to-long-term forgetting [23]. As illustrated in Fig. 5(a), the essentially self-powered Schottky device exhibits an expected merit of a very low resting current, that is down to fA-level noise for highly sensitive switching. Even at zero-bias condition, the EPSC clearly presents spike-timing and spike-intensity-dependent plasticity for a single stimulation. The longer or stronger stimulation can result in a higher output feedback, analog to the biological stimulus reinforce. On the basis of the trapping effect, the maintained interface conductance can electrically hold the post-stimulated current gradual decay rather than transient recovery. The bottom of Fig. 5(a) displays the photocurrent (Iph) as a nonlinear function of illumination power (P), which demonstrates the central role of trapping-induced photogating effect in photocurrent (Iph) [24]. The Iph can be expressed as: ${\textrm{I}_{\textrm{ph}}} \propto {\textrm{P}^{\textrm{0}\textrm{.45}}}.$ In addition, the interval effect is also a required factor of the spike-time-dependent plasticity (STDP) for neuromorphic applications, estimated by paired-pulse facilitation (PPF). PPF, an EPSC dynamical facilitation, correlates the enhancement of the second EPSC level evoked by spike with the time interval [25], and is thus regarded as a fundamental way to code temporal information in the biological system [26]. In this case,
Fig. 5. The synaptic behaviors of device doped by 1.5 wt%. (a) The short-term plasticity upon various single pulses. The EPSC behaviors programmed under different pulse intensity (upper panel), the EPSC behaviors programmed under different pulse lasting time (middle panel), and photocurrent is plotted and fitted as a function of light power (lower panel). (b) The typical PPF behavior. The PPF ratio is fitted by a double-exponent form. (c) The post-stimulation current decay under different pulse number. The decay is fitted by a power form. Inset is learning degree and forgetting rate versus pulse number.
Another required analogue functions mainly involving short-to-long-term plasticity, which can be described by repeated stimulation leading to longer conductance retention, has also been mimicked on our synapse. As shown in Fig. 5(c), a higher learning degree reflected in higher EPSC can be obtained by repeated stimulation, while displaying a slower post-stimulation current decay (that is, forgetting rate). For forgetting course, we used a power-law Wickelgren biological forgetting model to further quantify, which can be written as:
where I is the memory strength, t is the decay time, λ is the state of long-term memory at t = 0 (i.e., degree of learning), β is a scale parameter and ψ is the forgetting rate. Compared to exponential form, such power law can demonstrate the gradually slowing forgetting rate with time, that is referred to as consolidation. The fitting results are shown in Fig. 5(c), from which the similarity towards biological behavior is observed. Physically, such power-form current decay (I∼tβ) from LRS to HRS indicates the so-called Curie-von-Schweidler behavior, which is typical for capacitive charging, and it is commonly observed for charge trapping under bias in dielectrics [18]. The inset of Fig. 5(c) illustrates that the higher learning degree obtained by continuously repetitive ‘training’ has a slower forgetting rate, so it is described as an analogue consolidation. Therefore, our synaptic device can well be analogue to the biological characteristics in various forms of plasticity even at zero external voltage. In addition, considering that the device stability is no doubt important properties, we have tested device’s changing trends of single-pulse spike current (Iph) and double-pulses facilitation over time to estimate the device stability of photoresponse and synaptic functions, as shown in Supplement 1, Fig. S5. The device still exhibits slight attenuation, where the attenuation ratio of Iph and facilitation is 1.8% and 4.9% over 3 hours, respectively. This attenuation is thought to be the result of the gradual degradation of two-layered crystalline films. So, the improvement of film stability should be an important topic to promote the further applications of our devices towards advanced photo-computing integrated systems and multifunctional artificial intelligent fields.3. Conclusion
We have demonstrated a highly tunable self-powered synaptic device, in which the doped film electrode of PEDOT: PSS was used to modulate the asymmetry of work functions and thus manipulate the built-in electric field. When there was no doping, the built-in electric field was extremely weak due to the slightly asymmetric work functions between Au and PEDOT: PSS electrodes, and thus made the device poor synapse behaviors. In this work, the PEI doping modification has allowed the demonstration of an efficient photovoltaic-modulated method. When the PEDOT:PSS electrode was doped by 1.5%, the device yielded a remarkable VOC of 0.77 V and a high PCE of 6%. Based on optimized self-powered capacity, such synaptic devices, at applied zero electric field, could program various analog characteristics, such as typical PPF behaviors, stimulus reinforce and repetitive learning consolidation, etc. This work provides a zero-consumption synaptic device scheme which should promise to greatly accelerate the progress of neuromorphic hardware platform.
4. Experimental section/methods
4.1 Growth of ultrathin molecular crystals of C8-BTBT for synaptic device
A piece of Si substrate (roughly ∼1cm2) grown 100-nm-thick layer of SiO2 was cleaned by a standard cleaning process. Then, molecular layer-defined C8-BTBT molecular crystals were easily prepared on this substrate via the floating-coffee-ring-driven assembly [28]. The detailed grown procedure involved following several steps. The grown solvent was first allocated by mixing anisole and p-anisaldehyde (0.5 wt%). The C8-BTBT was then dissolved in this mixed solvent to obtain a grown solution. Subsequently, 3 µL solution was dropped onto the pre-cleaned SiO2/Si substrate and then using a pump to vent the air. The resulting high-quality molecular crystals were grown along pulling track of the liquid drop.
4.2 Fabrication of a synaptic device
The detailed fabrication of the synaptic device was accomplished following a two-step procedure. The PEDOT:PSS film was first prepared onto a hydrophobic modified SiO2/Si substrate for a easily exfoliated interface. To achieve this, a piece of pre-clean SiO2/Si substrate was sunk in an OTS solution (17 vol% in n-hexane) for 20 mins to achieve a hydrophobic modification, and then the modified substrate was ultrasonically cleaned in n-hexane, absolute ethanol and distilled water for 10 minutes each to remove residual OTS. Next, the PEDOT:PSS film covered substrate was obtained by using a pipette gun to suck PH1000 (1.2 wt% of PEDOT:PSS) onto the hydrophobic substrate and then being dried in glove box for 20-30 minutes. After these steps, easily exfoliated PEDOT:PSS film was achieved. The various-doped PEDOT:PSS films were obtained by bulk doping method, in which corresponding PEI is directly added into PEDOT:PSS solution, then performed the same steps demonstrated above to obtain films. The second step was to tailor same-sized Au and PEDOT:PSS electrodes (roughly ∼ 60 × 80 µm2), then the microprobe-assisted technology was used to transfer both electrodes onto the as-grown C8-BTBT with the channel width and length of approximately 60 µm and 10 µm. Notably, the two electrodes should be attached onto the same molecular layer. A micrometer-sized asymmetric device was fabricated.
4.3 Electrical measurements
Electrical measurements were carried out with a semiconductor parameter analyser (Agilent B1500) in a closed-cycle cryogenic probe station with a vacuum condition of 10−5 Torr. For lateral two-terminal diodes, the current-voltage (I-V) scanning measurement was performed by connecting metal probes to two electrodes to supply voltage sweeps. Emulations of the self-powered synaptic device were measured based on this test platform upon zero bias in bi-channel. The OLED light source was fixed into the probe station and linked to an optical pulse generator to produce various pattern pulses.
Funding
National Natural Science Youth Foundation (62106111); Wuxi University Research Start-up Fund for Introduced Talents (2021r011, 2021r012).
Acknowledgments
We gratefully acknowledge the National Natural Science Foundation of China and WuXi University for help with the laboratory construction and operation.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
References
1. C. D. Danesh, C. M. Shaffer, D. Nathan, and R. Shenoy, “Synaptic resistors for concurrent inference and learning with high energy efficiency,” Adv. Mater. 31(18), 1808032 (2019). [CrossRef]
2. Y. Wu and J. Feng, “Development and Application of Artificial Neural Network,” Wireless. Pers. Commun. 102(2), 1645–1656 (2018). [CrossRef]
3. M. L. Schneider, C. A. Donnelly, S. E. Russek, B. Baek, and M. R. Pufall, “Ultralow power artificial synapses using nanotextured magnetic Josephson junctions,” Sci. Adv. 4(1), 1701329 (2018). [CrossRef]
4. B. C. Jang, S. Kim, S. Y. Yang, J. Park, and J.-H. Cha, “Polymer analog memristive synapse with atomic-scale conductive filament for flexible nuromorphic computing system,” Nano Lett. 19(2), 839–849 (2019). [CrossRef]
5. J. Mao, L. Zhou, X. Zhu, Y. Zhou, and S. Han, “Photonic memristor for future computing: a perspective,” Adv. Opt. Mater. 7(22), 1900766 (2019). [CrossRef]
6. X. Wang, B. Wang, Q. Zhang, Y. Sun, and E. Wang, “Grain-boundary engineering of monolayer MoS2 for energy-efficient lateral synaptic devices,” Adv. Mater. 33(32), 2102435 (2021). [CrossRef]
7. G. Indiveri, E. Chicca, and R. Douglas, “A VLSI array of low-power spiking neurons and bistable synapses with spike-timing dependent plasticity,” IEEE Trans. Neural Netw. 17(1), 211–221 (2006). [CrossRef]
8. M. T. Sharbati, Y. Du, J. Torres, N. D. Ardolino, and M. Yun, “Low-power, electrochemically tunable graphene synapses for neuromorphic computing,” Adv. Mater. 30(36), 1802353 (2018). [CrossRef]
9. C. Yang, J. Qian, S. Jiang, H. Wang, and Q. Wang, “An optically modulated organic schottky-barrier planar-diode-based artificial synapse,” Adv. Opt. Mater. 8(13), 2000153 (2020). [CrossRef]
10. Y. van de Burgt, E. Lubberman, E. J. Fuller, S. T. Keene, and G. C. Faria, “A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing,” Nat. Mater. 16(4), 414–418 (2017). [CrossRef]
11. B. L. Sinnen, A. B. Bowen, J. S. Forte, B. G. Hiester, and K. C. Crosby, “Optogenetic control of synaptic composition and function,” Neuron 93(3), 646–660.e5 (2017). [CrossRef]
12. Z. Hao, H. Wang, S. Jiang, J. Qian, and X. Xu, “Retina-inspired self-powered artificial optoelectronic synapses with selective detection in organic asymmetric heterojunctions,” Adv. Sci. 9(7), 2103494 (2022). [CrossRef]
13. K. M. Kim, S. Ahn, W. Jang, S. Park, and O. O. Park, “Work function optimization of vacuum free top-electrode by PEDOT:PSS/PEI interaction for efficient semi-transparent perovskite solar cells,” Sol. Energy Mater. Sol. Cells 176, 435–440 (2018). [CrossRef]
14. L. Zeng, Q. Chen, Z. Zhang, D. Wu, and H. Yuan, “Multilayered PdSe2/perovskite schottky junction for fast, self-powered, polarization-sensitive, broadband photodetectors, and image sensor application,” Adv. Sci. 6(19), 1901134 (2019). [CrossRef]
15. L. Lyu, D. Niu, H. Xie, Y. Zhao, N. Gao, H. Zhang, Y. Zhang, P. Liu, and Y. Gao, “The correlations of the electronic structure and film growth of 2, 7-diocty [1] benzothieno [3, 2-b] benzothiophene (C8-BTBT) on SiO2,” Phys. Chem. Chem. Phys. 19(2), 1669–1676 (2017). [CrossRef]
16. D. Dumin and S. Rhodes, “Oxide wearout, breakdown, and reliability,” Int. J. Hi. Spe. Ele. Syst. 11(03), 617–718 (2001). [CrossRef]
17. E. Miranda, C. Mahata, T. Das, and C. K. Maiti, “An extension of the Curie-von Schweidler law for the leakage current decay in MIS structures including progressive breakdown,” Microelectron. Reliab. 51(9-11), 1535–1539 (2011). [CrossRef]
18. E. Mikheev, B. D. Hoskins, D. B. Strukov, and S. Stemmer, “Resistive switching and its suppression in Pt/Nb:SrTiO3 junctions,” Nat. Commun. 5(1), 3990 (2014). [CrossRef]
19. S. Heinze, J. Tersoff, R. Martel, V. Derycke, J. Appenzeller, and P. Avouris, “Carbon nanotubes as schottky barrier transistors,” Phys. Rev. Lett. 89(10), 106801 (2002). [CrossRef]
20. J. H. Jeon, D. H. Wang, H. Park, J. H. Park, and O. O. Park, “Stamping transfer of a quantum dot interlayer for organic photovoltaic cells,” Langmuir 28(25), 9893–9898 (2012). [CrossRef]
21. T. Zhang, “Ion-modulated radical doping of spiro-OMeTAD for more efficient and stable perovskite solar cells,” Science 377(6605), 495–501 (2022). [CrossRef]
22. Y. van de Burgt, A. Melianas, S. T. Keene, G. Malliaras, and A. Salleo, “Organic electronics for neuromorphic computing,” Nat. Electron. 1(7), 386–397 (2018). [CrossRef]
23. T. Ohno, T. Hasegawa, T. Tsuruoka, K. Terabe, J. K. Gimzewski, and M. Aono, “Short-term plasticity and long-term potentiation mimicked in single inorganic synapses,” Nat. Mater. 10(8), 591–595 (2011). [CrossRef]
24. J. O. Island, S. I. Blanter, M. Buscema, H. S. J. van der Zant, and A. Castellanos-Gomez, “Gate controlled photocurrent generation mechanisms in high-gain In2Se3 phototransistors,” Nano Lett. 15(12), 7853–7858 (2015). [CrossRef]
25. R. S. Zucker and W. G. Regehr, “Short-term synaptic plasticity,” Annu. Rev. Physiol. 64(1), 355–405 (2002). [CrossRef]
26. Y.-X. Hou, Y. Li, Z.-C. Zhang, J.-Q. Li, D.-H. Qi, X.-D. Chen, J.-J. Wang, B. -W. Yao, M. -X. Yu, T. -B. Lu, and J. Zhang, “Large-scale and flexible optical synapses for neuromorphic computing and integrated visible information sensing memory processing,” ACS Nano 15(1), 1497–1508 (2021). [CrossRef]
27. Y. H. Liu, L. Q. Zhu, P. Feng, Y. Shi, and Q. Wan, “Freestanding artificial synapses based on laterally proton-coupled transistors on chitosan membranes,” Adv. Mater. 27(37), 5599–5604 (2015). [CrossRef]
28. Q. Wang, Y. Li, X. Wang, and Y. Shi, “2D single-crystalline molecular semiconductors with precise layer definition achieved by floating-coffee-ring-driven assembly,” Adv. Funct. Mater. 26(19), 3191–3198 (2016). [CrossRef]
