A Large-Dynamic-Range Violet Phosphorus Heterostructure Optoelectronic Synapse for High-Complexity Neuromorphic Computing

A Large-Dynamic-Range Violet Phosphorus Heterostructure Optoelectronic Synapse for High-Complexity Neuromorphic Computing | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article A Large-Dynamic-Range Violet Phosphorus Heterostructure Optoelectronic Synapse for High-Complexity Neuromorphic Computing Xiaoxian Liu, Shuiyuan Wang, Ziye Di, Haoqi Wu, Chunsen Liu, Peng Zhou This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2441360/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Neuromorphic computing can efficiently handle data-intensive tasks and address the redundant data interaction required by traditional von Neumann architectures. Synaptic devices are essential components for neuromorphic computation. For high computational accuracy, synaptic devices need to retain good conductance linearity, but this leads to a limited dynamic range (10 ~ 100) and weight states, which impedes their processing of high-complexity tasks and restricts further advances in accuracy. Two-dimensional materials, such as transition metal disulfides and phosphorene, hold promise for the construction of synaptic devices with large dynamic ranges due to their strong light-matter interactions, while the stability of phosphorene remains an issue. Here, for the first time, we use the most stable violet phosphorene for device applications. The combination of violet phosphorene and molybdenum disulfide demonstrates an optoelectronic synapse with a record dynamic range of over 10 6 , benefiting from a significant threshold shift due to charge transfer and trapping in the heterostructure. Remarkable synaptic properties are demonstrated, including 128 distinguishable conductance states, electro-optical dependent plasticity, short-term paired-pulse facilitation, and long-term potentiation/depression. High-precision image classification with accuracies of 95.23% and 79.65% is achieved for MNIST and high-complexity Fashion-MNIST datasets, which is close to the ideal device (95.47%, 79.95%), indicating the potential of dynamic range and multi-states for optimizing accuracy. This work fills the device application gap of violet phosphorene and provides a strategy for building synaptic devices with large dynamic range to facilitate neuromorphic computing. Physical sciences/Materials science/Materials for devices/Electronic devices Physical sciences/Materials science/Nanoscale materials/Two-dimensional materials Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Emerging artificial intelligence applications such as image recognition, motion detection, and autonomous driving etc. require processing massive amounts of data, which poses a serious challenge to the von Neumann architecture that separates memory from computation 1 , 2 . Brain-inspired neuromorphic computing offers advantages in terms of energy efficiency and operational latency, bringing a promising approach to this challenge 3 . In the neuromorphic system, synaptic devices are considered as the core units due to their far greater number than neurons and their properties are essential to achieve efficient, high-precision computation 4 . Among all the characteristics, the dynamic range determines the capability of mapping device conductance to weights in neural network algorithms, which is crucial in neuromorphic computing, especially for inference and learning tasks 4 , 5 . A larger dynamic range means the potential to obtain more conductance states while maintaining linearity. It was reported that MNIST handwriting recognition accuracy improves significantly with increasing dynamic range below 30, and the desire for larger dynamic range is more pressing for complicated tasks 6 . However, the dynamic range of most memory devices in the analog switching regime is typically 10 ~ 100, unlike the high on/off ratios available in binary switching 4 . The limited dynamic range and conductance states constrains the accuracy, although it could be compensated by optimized linearity/symmetry. Due to their attractive characteristics, two-dimensional (2D) materials represented by molybdenum disulfide (MoS 2 ) appear to be promising candidates to build synapse devices, as their atomic-scale thickness could reduce operation voltage and energy consumption 7 , 8 . Furthermore, their strong resonant light absorption (> 20%) 9 and better electrostatics control make them suitable for building optoelectronic synapses with large dynamic ranges. In addition, 2D phosphorene such as black phosphorus have also been introduced to construct optoelectronic synapses, but their stability is still a problem 10 , 11 . Violet phosphorus (VP), considered to be the most stable allotrope of phosphorene, has recently been successfully synthesized 12 – 17 . With excellent optoelectronic properties, including a direct bandgap of ~ 2.54 eV 12 , extremely low dark current, and large light-to-dark ratio, VP becomes a potential alternative for optoelectronic synaptic devices. However, current research on VP has focused on material synthesis and characterization, while its application in devices has been left vacant. Here, we present a VP-MoS 2 heterostructure synapse that attempts the optoelectronic device application of VP. The device exhibits an ultra-high dark-to-light ratio due to the relatively large bandgap and strong light-matter interactions of VP. The optoelectronic coupling and charge transfer within VP-MoS 2 heterostructure lead to a strong threshold shift effect when weak light is applied. Thanks to this phenomenon, a dynamic range of over 10 6 with distinguishable conductance multi-states of 128 can be achieved at a light intensity of ~ 8 µW. Moreover, we demonstrated bionic synaptic behaviors with the heterostructure device, including short-term and long-term potentiation (LTP) using light stimulation and long-term depression (LTD) using electric stimulation. Compared with mainstream analog memories, our heterostructure device owns ultralow off-state current in synaptic behaviors (large dynamic range), which could minimize the deviation in the conductance-weight mapping (the lowest but not zero conductance state in the device is mapped into the absolute zero weight in algorithm, thus bringing deviation between device conductance and algorithm weight in every following states), resulting in improvement in learning accuracy. Using VP-MoS 2 device conductance mapping, we simulated MNIST handwriting digit and high-complexity Fashion-MNIST image classification tasks to evaluate and validate the impact of dynamic range on accuracy. The accuracy was able to reach 95.23% and 79.65%, respectively, which is comparable to the ideal device (95.47%, 79.95%). This work opens the way for the use of VP in optoelectronics and provides a feasible strategy for large-dynamic-range synapses, thus further contributing to the accuracy improvement in high-complexity neuromorphic computation. Results Figure 1 a shows the lattice structure of VP, which appears to be monoclinic with a space group of P 2/n , with crystallographic lattice constants of a = 9.210 Å, b = 9.128 Å, c = 21.893 Å, and β = 97.776° 12 . The VP structure generally shows a bi-tubular structure with one layer stacked on another vertically along the z-direction, as shown in Fig. 1 a lower panel. The Raman spectrum of VP-MoS 2 heterostructure excited by a 532 nm laser is shown in Fig. 1 b. The Raman shift peaks at 183, 211, and 278 cm - 1 represent for the variation modes of VP atoms while the peaks at 361, 379 and 476 cm - 1 represent for the stretching of the atom cages as a whole. The complex Raman spectrum indicates a large density of photon states in VP, which further contributes to the strong light-matter interaction and electron-phonon scatterring 18 . In addition, few-layered VP possesses a direct bandgap of ~ 2.54 eV 12 and usually behaves as an n-type semiconductor, which indicates its potential to build optoelectronic devices. On this basis, we built VP photo-transistors and measured their characteristics, as shown in Fig. 1 c,d. Due to its unique lattice and bandgap characteristics, VP-based photodevice shows an extremely low dark current (~ fA) and a high light-to-dark ratio (~ 10 5 ). In addition, the VP phototransistors were exposed to the ambient air without any encapsulation and could still exhibit an on/off ratio of ~ 100 and a light-to-dark ratio of ~ 10 3 after 14 days, while black phosphorous would completely lose on/off characteristics in 2 days 19 , indicating that VP is the most stable phosphorus, as shown in Fig. 1 d (original data see Supplementary Figure S1 ). However, as the photo-response of VP phototransistors is generally instantaneous (see Figure S2 in Supplementary Information), which is not suitable for synaptic applications, heterostructure was then introduced. The device schematic and process flow of our heterostructure device is shown in Fig. 1 e and Fig. 1 f, respectively. MoS 2 and VP were successively transferred to a 300 nm-thick silicon oxide substrate to form the vertically stacked heterostructure (Device I). Electron beam lithography (EBL) was used to form the electrode pattern and electron beam evaporation (EBE) deposits the source and drain electrodes. Top gate dielectric (30 nm HfO 2 ) was then deposited using atomic layered deposition (ALD) with 1 nm seed layer (SiO 2 ) using EBE, followed by top gate electrode deposition. Another MoS 2 device (Device II) without VP was built using the same material as the control. The VP stacked here on the top of MoS 2 works as a photogate, and the thickness of the dielectric and top gate have been carefully determined to ensure the transmission of light. Figure 1 g shows the scanning electron microscope (SEM) image of the heterostructure synaptic device. High-resolution scanning transmission electron microscopy (STEM) was used to characterize the device microstructure. Figure 1 h ~ j shows the cross-section HAADF STEM images of the VP-MoS 2 under different magnifications, which implies a layered van der Waals structure and is consistent with the lattice in Fig. 1 a. The STEM image in Fig. 1 j shows a single-layer VP thickness of ~ 2.2 nm, which is consistent with the theoretical value 12 . Figure 1 k shows the STEM image of the VP-MoS 2 heterostructure and the corresponding energy dispersive spectroscope (EDS) mapping, which exhibits a clean van der Waals interface between VP and MoS 2 layers. The heterostructure region was further investigated using atomic force microscope (AFM), which is provided in Figure S3 in the Supplementary Information. The thickness of VP and MoS 2 is measured by AFM to be ~ 6.66 nm and 5.85 nm, corresponding to 3 and 9 layers, respectively. The operating principle of our VP-MoS 2 synaptic device is described in Fig. 2 a. When a voltage is applied to the top gate, the device exhibits an n-type on/off switch, as shown by the dashed line in Fig. 2 a left panel. Subsequently, when a weak light stimulus is applied, a negative threshold shift of several V can be observed, leading to a “threshold window”, which is shown by the solid line in Fig. 2 a left panel. Different from the ordinary photocurrent effect where the off-state current increase obviously under illumination, here the photostimulation mainly act as a trigger to switch the on/off state instead of increasing the off-state current. Similar phenomena can be found in other heterostructures as well. By carefully selecting the operation point, the former off-state can be transformed to the on-state by light induction, resulting in a record large optical dynamic range (over 10 6 ), far better than most analog switching memories (10 ~ 100), as shown in Fig. 2 a right panel. It is worth noting that the operation point is typically negative, which means that our device could maintain a high response at a lower off-state current than most mainstream analog memories, leading to great improvement in dynamic range. To further investigate the principle of threshold shift, we construct an energy band model for the heterostructure device (Device I), as shown in Fig. 2 b. The theoretical bandgaps of MoS 2 and VP are about 1.9 eV 20 and 2.54 eV 12 , respectively. When stacked together without external voltage (flat band), VP and MoS 2 form a type-II heterostructure, with the bottom of the conductance band of MoS 2 higher than VP and the top of the valence band lower than VP. When a negative voltage is applied to the top gate, the device changes from the flat band to the programmed state. Photostimulation is then applied to the device in the programmed state to induce photogenerated electron-hole pairs. According to the energy band structure, electrons and holes move separately: holes move toward VP and are trapped by the potential well at the interface of VP and HfO 2 (the properties of VP-MoS 2 heterostructure device without the top gate dielectric are shown in Supplementary Figure S4 , where the separated photogenerated holes are not efficiently trapped, making the threshold shift rather weak and leading to a limited dynamic range); while electrons move toward MoS 2 , leading to an increase in transient channel current. Upon withdrawal of the light stimulus, the trapped holes contribute to the conductivity of the channel, thus resulting in a negative threshold shift. Through applying a positive voltage to the top gate, the band structure changes, leading to a release process of the trapped holes. As a result, the device is reset to the initial state. In this way, the heterostructure device shows the coexistence of optical potentiation and electrical inhibiting, which exactly mimic biological excitatory and inhibitory plasticity and can be mapped to artificial neural networks for neuromorphic computing. As a comparison, we construct another energy band model without VP (Device II) to further investigate the role played by VP in such a process, see Supplementary Figure S5 . Without the strong light-matter interaction with VP, MoS 2 produces limited electron-hole pairs. More importantly, in this case, the separation of electrons and holes cannot be captured efficiently, so only a transient photocurrent exists without a significant threshold change. To confirm the validity of our energy band model, we further use COMSOL Multiphysics to conduct finite element simulation based on the above theoretical analysis. More details about the simulation could be found in Section 6 in Supplementary Information. Figure 2 c depicts the distribution of holes in the heterostructure device under different conditions. When a positive voltage is applied at the top gate without light stimulation, the heterostructure device exhibits on current state with low hole densities, due to the fact that both VP and MoS 2 are n-type semiconductors. When the light stimulation was applied to the device, electron-hole pairs are generated and separated, with holes moving towards VP and trapped, resulting in the distribution shown in Fig. 2 c upper panel. In this case, although the concentration of carriers increases overall, the current of the device does not change much because the device is still in on-state. However, when a negative voltage is applied, the heterostructure device is set to off-state. When light stimulation is applied, the accumulation of holes in VP leads to threshold voltage shift, which turn the device from a non-conductive state into a conductive state, leading to a large dynamic range through light stimulation, as shown in Fig. 2 c lower panel. Experiments are implemented to verify these analyses. Figure 3 a upper panel shows the transfer curve of the VP-MoS 2 heterostructure device under varying illuminations, where the source-drain voltage (V ds ) is fixed at 1 V and the laser wavelength is 473 nm. Obvious threshold shift can be observed for illumination as low as 6 µW and expand with increasing intensity. The actual laser intensity may be lower, taking into account the influence of the top gate dielectric and electrodes. As the intensity increases to 20 µW, the threshold voltage has changed from about − 4 V to below − 8 V, which is desired for the large dynamic range. According to the transfer curve, when the top gate voltage (V tg ) is fixed in this range (-4 V to -8 V), a 20 µW optical stimulation could lead to a large light on/off ratio up to over 10 6 . In addition, we test the transfer curve of Device II as a control, as shown in Fig. 3 a lower panel . In this device, almost no change in threshold voltage is observed even when 30 mW of strong light is applied, which is in line with our previous analysis. To visualize the threshold shift of the VP-MoS 2 device, we collected the source-drain currents for different top gate voltages as well as for different light intensities, as shown by color mapping in Fig. 3 b (see original curves in Figure S8 in Supplementary Information). Considering 10 nA as the boundary between on and off switching, the dashed line could reflect the shift in threshold voltage, that is, the threshold changes from − 12 V to -16 V, which leaves a “threshold window” of ~ 4 V, sufficient for synaptic device operation. We chose an operation point of -4 V and test the output curve under varying light intensities, as shown in Fig. 3 c. At such an operation point, the output current increases to around 1 µA under 8 µW illumination, which is consistent with the transfer curve. Such a phenomenon has been repeatedly verified in different batches of devices with this heterostructure ( Figure S7 in Supplementary Information). Although the threshold voltage varies slightly with the thickness of 2D materials, this phenomenon could be reproduced in each device, which means that there is a stable strategy to improve the optical dynamic range. Next, we stimulate the device with a single laser spike and test its response, as shown in Fig. 3 d. When it comes to neuromorphic computing, the laser spike here could simulate presynaptic input, and the channel current is monitored as post-synaptic current (PSC). The stimuli are applied at different operation points (different V tg ). Although the base current of the device decreases with increasing V tg , the PSCs almost all reach the µA level, which is consistent with the on-state current of our heterostructure device. As a result, the excitement ratio (PSC/base current) increases rapidly as V tg increases, and can exceed 10 6 thanks to the extremely low dark current of VP. Similar tests are carried out using Device II (see Figure S9 in Supplementary Information). A comparison of the two devices is shown in Fig. 3 e, where the squares represent the performance of the VP-MoS 2 heterostructure device (Device I) and the circles represent the MoS 2 transistor (Device II). To eliminate the effect of different original threshold voltages, the x-axis has been unified as base current rather than top gate voltage. Regarding the PSC amplitude (left axis), the MoS 2 transistor shows high PSC amplitude only when the base current itself is relatively high, whereas VP-MoS 2 exhibits high PSC amplitude irrespective of the base current. This discrepancy leads to at least a 3-magnitude improvement in excitement ratio (right axis), representing a stronger synaptic response to the stimulus. We further explored the potential of VP-MoS 2 device for the simulation of synaptic plasticity and behavior. Dual laser pulses with different intervals were applied to the device to test the paired-pulse facilitation (PPF) characteristics, which is essential to simulate biologically short-term plasticity (STP). The inset in Fig. 4 a shows a output waveform, where the base current is around 100 fA, and A 1 and A 2 represent the amplitudes after the first and second laser pulses. The second pulse exhibits a stronger response than the first pulse, and quickly recovers to the base current, showing typical PPF characteristics. Waveforms based on different interval times are shown in Figure S10 in Supplementary Information. When the interval is as low as 150 ms, our device shows a fairly high PPF index (determined by A 2 /A 1 ) of up to 853%, and gradually recovers to 100% as the interval increases above 3000 ms. The dashed line here indicates the fitting curve at the experimental points, which obeys an exponential decay, consistent with the theoretical result 21 . Moreover, the long-term plasticity of the heterostructure synaptic device was explored by increasing the number of laser pulses. As shown in Fig. 4 b, 30 laser spikes with different intensities were applied to achieve a progressive excitatory PSC modulation, which simulated the LTP plasticity. The PSC gain (determined as A 30 /A 1 ) increases significantly as the base current A 0 decreases (increase in V tg ) as well as the increased laser power. Figure 4 c shows the PSC gain (A n /A 1 ) as a function of top gate voltage and pulse number. When the pulse number is as low as 5, the synapse device generally exhibits STP, as shown in Fig. 4 c left panel . In this case, the PSC increases linearly with the accumulation of laser spikes and falls back to the base current a few seconds after the removal of laser. The device shows a clear transition from STP to LTP as the pulse number increases, with an increase in PSC gain and retention time ( Figure S11 and Figure S12 in Supplementary Information). In addition, a higher top gate voltage will also significantly increase the PSC gain, leading to a large dynamic range of over 10 6 ( Figure S12 in Supplementary Information). However, the extremely high dynamic range is achieved at the expense of linearity and retention time, which are also critical for neuromorphic computing. To balance all these key metrics, the operation point was carefully selected to achieve a large dynamic range (~ 10 6 ) with 30 conductance states, fair linearity (1.31) and retention time (~ 40 s), as shown in Fig. 4 d red curve (more details see Figure S13 in Supplementary Information), which could meet the requirements of high-precision neuromorphic computing 4 . A large dynamic range means the potential to contain more conductance states. To further investigate the capability of the heterostructure device to obtain distinguishable conductance multi-states, we applied more pulses for both potentiation and depression. Figure 4 d shows the normalized long-term synaptic plasticity using optical pulses for potentiation and electrical pulses for depression under different conditions. By applying 128 light pulses for stimulation, 128 stable, non-crossing, and distinguishable conductance states are generated. The original waveforms are shown in Figure S14 in Supplementary Information, with a base state of around 100 fA, and after 128 pulses of stimulation, the current rises to ~ µA, implying a dynamic range of over 10 6 , which indicates a strong capability to map synaptic device conductance to neural network weights. In addition, by applying over 200 pulses, we can obtain up to ~ 180 distinguishable conductance states (original waveforms shown in Figure S15 in Supplementary Information). Figure 4 e depicts the waveforms of each state extracted from Fig. 4 d, which are distinguishable and stable. The device also exhibits an extremely high A n /A 1 of ~ 10 4 , showing the potential to obtain more conductance states. However, limited by the instability of light stimulation and strong light response of the device, state intervals smaller than ~ nA hardly exist stably, limiting the further increase in the number of states. The dynamic ranges and number of conductance state statistics for synaptic devices based on emerging analog memories are shown in Fig. 4 f. The devices are divided into four categories according to working mechanisms: charge-trapping electrical devices 22 – 27 , ferroelectric devices 28 – 33 , electrical memristors 34 – 38 , and optoelectronic devices 10 , 39 – 43 , which are based on a variety of promising materials, including 2D, organic, oxide (e.g. indium gallium zinc oxide) and perovskite semiconductors. Thanks to the extremely low off-state current, our VP-MoS 2 heterostructure synapse exhibits a record high dynamic range of over 10 6 , as well as 128 distinguishable conductance multi-sates, far outperforming that of current mainstream analog memories, providing a new strategy for improving the dynamic range and multi-states in synaptic devices. Finally, we used the NeuroSim multilayer perceptron (MLP) neural network simulator 44 to validate the ability of VP-MoS 2 synaptic device with a large dynamic range for high-complexity image classification tasks. The neural network used is shown in Fig. 5 a, which consists of an input layer, a hidden layer and an output layer. Each neuron node in one layer is connected to each node in the following layer, forming a fully connected neural network. Neuron nodes are connected via synaptic devices, and device conductance represents network weights. W IH and W HO represents the weight matrix between the input and hidden layers and between the hidden and output layers, respectively. We have made the necessary modifications to the original network model in the NeuroSim simulator to make it suitable for our classification tasks. We performed the image classification based on two standard datasets: MNIST and Fashion-MNIST (an MNIST-like dataset with higher complexity) 45 . The input image data has been pre-processed into grayscale data of each pixel as the input layer (20×20 for MNIST and 28×28 for Fashion). The network contains 100 hidden neurons and 10 output neurons (referring to 10 kinds of labels, i.e. handwriting digits or objects), more details about the simulation could be found in Section 16–19 in Supplementary Information. Figure 5 b shows the distribution of weights before and after MNIST training, consisting of W IH and W HO . The initial weights for both matrices are set randomly from 7 states (0, ± 1, ±0.33, ± 0.66), and the weights after training are updated to 128 states, corresponding to the 128 conductance states of VP-MoS 2 device, indicating the update of the network weights. The classification accuracy as a function of training epochs is shown in Fig. 5 c. According to the simulation, our device eventually reach an accuracy of 95.23% for MNIST and 79.65% for the Fashion dataset, which is close to those of ideal devices (95.47% and 79.95%). As the original algorithm of the simulator is designed especially for the 20×20 MNIST dataset, the learning accuracy for the Fashion dataset is relatively lower but still could reflect the superiority of our device by comparing it with the ideal device. The detailed parameters and results of our simulation could be founded in Table S2 in Supplementary Information. It is worth mentioning that in the actual classification process, the conductance states of real devices are mapped into the synaptic weights in the algorithm, where the lowest conductance state is transferred into weight 0. However, due to the physical limitation of the device, the absolute 0 weight is unreachable, which would affect all the conductance-weight mapping and finally lead to accuracy deviation between the ideal condition and the physical device 5 . Compared with most mainstream analog memories, our heterostructure synaptic device owns a lower off-state current and could minimize such deviation, thus leading to negligible error from ideal cases. Therefore, we further investigated the dependence of classification accuracy on dynamic range. Figure 5 d shows the final classification accuracy for different dynamic ranges and the number of conductance states while keeping other parameters constant (consistent with the previous simulation for VP-MoS 2 heterostructure device). In these cases, the change in dynamic range occurs with a decrease in the off-state while keeping the on-state fixed. Simulation results show that for simple classification tasks like MNIST, accuracy is strongly suppressed when the dynamic range is below 10, and for dynamic range over 100, the increase of dynamic range could still lead to improvement in learning accuracy. In addition, the increase in states number will also improve learning accuracy significantly. In classification tasks with higher complexity like Fashion-MNIST, such dependence becomes more pronounced. In Fig. 5 e, we further investigated the learning performance of different synaptic devices in image classification tasks with different complexity. The x-axis shows the complexity of different datasets, which are defined as the average information entropy of the images (See Section 19 in Supplementary Information) and the y-axis shows the error between ideal cases (software implementation) and physical devices. In those cases, most synaptic devices show relatively high error (mostly over 2%) due to their relatively low dynamic range and fewer state numbers 22 , 26 , 30 , 38 , 43 , 46 – 50 . By combining large dynamic ranges and sufficient state numbers, our VP-MoS 2 synaptic device shows negligible error from ideal cases, even for the high-complexity Fashion-MNIST classification task. Discussion In conclusion, we have achieved the first device application of the most stable phosphene VP and proposed the VP-MoS 2 heterostructure for optoelectronic synapse. VP exhibits extremely low dark currents and high dark-to-light ratios and restricts photogenerated holes by forming the type-II heterostructure with MoS 2 , leading to a strong shift in threshold voltage. This phenomenon was exploited to realize optically induced synaptic plasticities with large dynamic ranges. By carefully selecting the operation point, key metrics including dynamic range, linearity and retention time could be balanced. A record dynamic range (over 10 6 ) with a high number of conductance states (128) was obtained, which is superior to current synaptic devices. Furthermore, there is still room for improvement in these indicators. For example, the light-matter interaction could be enhanced by optimizing the design of the top gate dielectric. And the retention time is limited by the charge-trapping capability of the interface, which can be further improved by introducing additional physical fields. The larger dynamic range allows for more levels of conductance states, making it suitable for high-complexity tasks and further improving accuracy. We explored the potential of VP-MoS 2 synaptic device for MNIST and high-complexity Fashion-MNIST image classification through NeuroSim simulator and achieved accuracies comparable to that of ideal devices. The results indicate the significance of a large dynamic range for accuracy optimization. This work demonstrates the potential of VP as a unique optoelectronic material for device application and provides a viable strategy for high-precision neuromorphic computing with large-dynamic-range synaptic devices. Methods Fabrication of the heterostructure synaptic device. Few-layer MoS 2 and VP were exfoliated and transferred to a silicon substrate with 300nm-thick SiO 2 , forming the heterostructure. The electrode pattern was defined by EBL and Cr/Au metal was deposited by EBE to form the source and drain. For the top gate dielectric, 2 nm SiO 2 was deposited as a seed layer using EBE, followed by ALD to form 30 nm HfO 2 as the dielectric layer. Then the top gate electrodes were formed again through EBL and EBE. Finite element simulation of heterostructure device. The simulation process is realized based on COMSOL Multiphysics. We combined the semiconductor module and electromagnetic waves module to simulate the generation, migration, and trapped process of photo-generate carriers. The material characteristics of VP and MoS 2 are set according to former research. The vertical dimension of the device is set to 10 nm, which is negligible compared with a channel length of 3 µm. The top gate dielectric (HfO 2 ) is placed at the top of VP, where the Shockley-Read-Hall model is used to simulate the trap-assisted surface recombination process. Through the Newton iteration method, the carrier concentration and corresponding port current are calculated under different voltage and illumination conditions. Characterization and measurement of the heterostructure synaptic device. The surface morphology of the heterostructure device was characterized by AFM, showing the thickness of VP and MoS 2 are about 6.66, and 5.85 nm, respectively. To examine the constructed van der Waals heterojunction interface, a cross-sectional analysis was performed using HAADF STEM technology with EDS elements mapping analysis. In addition, the 2D layered materials were characterized by Raman spectroscopy. VP showed strong peaks near 361, 379, 476 cm - 1 , and MoS 2 showed peaks at 383, 410 cm - 1 . The Cascade probe station equipped with Keysight B1500A semiconductor analyzer was used to characterize the electrical properties of the heterostructure device and the simulation of synaptic plasticity under the ambient environment. The optical stimulation is applied through a 473 nm blue light laser and the corresponding light intensity is measured by a light intensity meter. Simulation of image classification based on the fully connected neural network. The simulation process is realized based on the NeuroSim MLP simulator. A complete epoch includes loading the data set, defining a fully connected network model, training the network, testing the network and calculating the classification accuracy. The training process includes feed-forward and back-propagation processes. To execute the simulation using the Fashion-MNIST dataset, the topology of the network has been modified, and the learning rate has been carefully changed to optimize the accuracy. Data availability The data that support the findings of this study are available from the corresponding authors upon reasonable request. Code availability The codes used for neural network simulation are available from the corresponding authors upon reasonable request. Declarations Acknowledgements This work was supported by the National Key Research and Development Program of China (2021YFA1200500), National Natural Science Foundation of China (61925402 and 62090032), Science and Technology Commission of Shanghai Municipality (19JC1416600), and China Postdoctoral Science Foundation (2022M720032). Author contributions X. Liu and S. Wang co-wrote the manuscript. Z. Di, H. Wu and C. Liu helped in material characterization. P. Zhou conceived the idea and supervised the work. All authors provided suggestions for revisions and improvements to the work. Competing interests The authors declare no competing interests. Additional information Correspondence and requests for materials should be addressed to P Zhou or S. Wang. References Wang S. Y., Chen X. Z., Huang X. H., Zhang D. W.&Zhou P. Neuromorphic Engineering for Hardware Computational Acceleration and Biomimetic Perception Motion Integration. Adv Intell Syst-Ger 2 , (2020) Yoo H. J. Intelligence on Silicon: From Deep-Neural-Network Accelerators to Brain Mimicking AI-SoCs. Isscc Dig Tech Pap I 62 , 20–28 (2019) Roy K., Jaiswal A.&Panda P. Towards spike-based machine intelligence with neuromorphic computing. Nature 575 , 607–617 (2019) Zhang W. Q., et al. Neuro-inspired computing chips. Nature Electronics 3 , 371–382 (2020) Choi S., Yang J.&Wang G. Emerging Memristive Artificial Synapses and Neurons for Energy-Efficient Neuromorphic Computing. Advanced Materials 32 , (2020) Choi S., et al. SiGe epitaxial memory for neuromorphic computing with reproducible high performance based on engineered dislocations. Nature Materials 17 , 335-+ (2018) Wang S. Y., Liu X. X.&Zhou P. The Road for 2D Semiconductors in the Silicon Age. Advanced Materials (2022) Wang S. Y., Zhang D. W.&Zhou P. Two-dimensional materials for synaptic electronics and neuromorphic systems. Sci Bull 64 , 1056–1066 (2019) An J. R., et al. Perspectives of 2D Materials for Optoelectronic Integration. Adv Funct Mater 32 , (2022) Tian H., et al. Anisotropic Black Phosphorus Synaptic Device for Neuromorphic Applications. Advanced Materials 28 , 4991–4997 (2016) Ahmed T., et al. Optically Stimulated Artificial Synapse Based on Layered Black Phosphorus. Small 15 , (2019) Zhang L. H., et al. Structure and Properties of Violet Phosphorus and Its Phosphorene Exfoliation. Angew Chem Int Edit 59 , 1074–1080 (2020) Zhang B., et al. Cross structured two-dimensional violet phosphorene with extremely high deformation resistance. J Mater Chem A 9 , 13855–13860 (2021) Zhao R. Z., et al. Violet phosphorus quantum dots. J Mater Chem A 10 , 245–250 (2021) Zhang L. H., et al. Fast Identification of the Crystallographic Orientation of Violet Phosphorus Nanoflakes with Preferred In-Plane Cleavage Edge Orientation. Adv Funct Mater 32 , (2022) Ricciardulli A. G., Wang Y., Yang S.&Samori P. Two-Dimensional Violet Phosphorus: A p-Type Semiconductor for (Opto)electronics. J Am Chem Soc 144 , 3660–3666 (2022) Baumer F., et al. Synthesis, Characterization, and Device Application of Antimony-Substituted Violet Phosphorus: A Layered Material. Acs Nano 11 , 4105–4113 (2017) Li Y. Y., et al. Impurity-Induced Robust Trionic Effect in Layered Violet Phosphorus. Advanced Optical Materials 10 , (2022) Guo Z. N., et al. Metal-Ion-Modified Black Phosphorus with Enhanced Stability and Transistor Performance. Advanced Materials 29 , (2017) Mak K. F., Lee C., Hone J., Shan J.&Heinz T. F. Atomically Thin MoS2: A New Direct-Gap Semiconductor. Physical Review Letters 105 , (2010) Zucker R. S.&Regehr W. G. Short-term synaptic plasticity. Annu Rev Physiol 64 , 355–405 (2002) Kim S., et al. Pattern Recognition Using Carbon Nanotube Synaptic Transistors with an Adjustable Weight Update Protocol. Acs Nano 11 , 2814–2822 (2017) Ren Y., et al. Gate-Tunable Synaptic Plasticity through Controlled Polarity of Charge Trapping in Fullerene Composites. Adv Funct Mater 28 , (2018) Kumar M., Ban D. K., Kim S. M., Kim J.&Wong C. P. Vertically Aligned WS2 Layers for High-Performing Memristors and Artificial Synapses. Advanced Electronic Materials 5 , (2019) Wang S. Y., et al. A MoS2/PTCDA Hybrid Heterojunction Synapse with Efficient Photoelectric Dual Modulation and Versatility. Advanced Materials 31 , (2019) Noh G., et al. Large Memory Window of van der Waals Heterostructure Devices Based on MOCVD-Grown 2D Layered Ge4Se9. Advanced Materials (2022) Shim H., et al. An elastic and reconfigurable synaptic transistor based on a stretchable bilayer semiconductor. Nature Electronics (2022) Seo M., et al. First Demonstration of a Logic-Process Compatible Junctionless Ferroelectric FinFET Synapse for Neuromorphic Applications. Ieee Electr Device L 39 , 1445–1448 (2018) Jang S., et al. Ultrathin Conformable Organic Artificial Synapse for Wearable Intelligent Device Applications. Acs Appl Mater Inter 11 , 1071–1080 (2019) Kim M. K.&Lee J. S. Ferroelectric Analog Synaptic Transistors. Nano Letters 19 , 2044–2050 (2019) Li B. C., et al. An Electronic Synapse Based on 2D Ferroelectric CuInP2S6. Advanced Electronic Materials 6 , (2020) Wang S. Y., et al. Two-dimensional ferroelectric channel transistors integrating ultra-fast memory and neural computing. Nature Communications 12 , (2021) Chung W., Si M. W.&Ye P. D. D. First Demonstration of Ge Ferroelectric Nanowire FET as Synaptic Device for Online Learning in Neural Network with High Number of Conductance State and G(max)/G(min). Int El Devices Meet (2018) Xu R. J., et al. Vertical MoS2 Double-Layer Memristor with Electrochemical Metallization as an Atomic-Scale Synapse with Switching Thresholds Approaching 100 mV. Nano Letters 19 , 2411–2417 (2019) Ren Y., et al. Phosphorene nano-heterostructure based memristors with broadband response synaptic plasticity. Journal of Materials Chemistry C 6 , (2018) Xu Z. W., et al. Ultrathin electronic synapse having high temporal/spatial uniformity and an Al2O3/graphene quantum dots/Al2O3 sandwich structure for neuromorphic computing. Npg Asia Mater 11 , (2019) Sangwan V. K., et al. Multi-terminal memtransistors from polycrystalline monolayer molybdenum disulfide. Nature 554 , 500-+ (2018) Pereira M. E., et al. Flexible Active Crossbar Arrays Using Amorphous Oxide Semiconductor Technology toward Artificial Neural Networks Hardware. Advanced Electronic Materials (2022) Wang J. X., et al. Deep-ultraviolet-triggered neuromorphic functions in In-Zn-O phototransistors. Applied Physics Letters 113 , (2018) Wu Q. T., et al. Photoelectric Plasticity in Oxide Thin Film Transistors with Tunable Synaptic Functions. Advanced Electronic Materials 4 , (2018) Wang Y., et al. Photonic Synapses Based on Inorganic Perovskite Quantum Dots for Neuromorphic Computing. Advanced Materials 30 , (2018) Zhang Z. H., et al. All-in-one two-dimensional retinomorphic hardware device for motion detection and recognition. Nature Nanotechnology 17 , 27-+ (2022) Liu K. Q., et al. An optoelectronic synapse based on alpha-In2Se3 with controllable temporal dynamics for multimode and multiscale reservoir computing. Nature Electronics (2022) Chen P. Y., Peng X. C.&Yu S. M. NeuroSim plus: An Integrated Device-to-Algorithm Framework for Benchmarking Synaptic Devices and Array Architectures. 2017 Ieee International Electron Devices Meeting (Iedm) (2017) Xiao H, Rasul K, Vollgraf R. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747 (2017). Shim H., et al. An elastic and reconfigurable synaptic transistor based on a stretchable bilayer semiconductor. Nature Electronics 5 , 660–671 (2022) Joshi V., et al. Accurate deep neural network inference using computational phase-change memory. Nature Communications 11 , (2020) Luo Z., et al. High-precision and linear weight updates by subnanosecond pulses in ferroelectric tunnel junction for neuro-inspired computing. Nature Communications 13 , (2022) Seo S., et al. An Optogenetics-Inspired Flexible van der Waals Optoelectronic Synapse and its Application to a Convolutional Neural Network. Advanced Materials 33 , (2021) Cao R. R., et al. Compact artificial neuron based on anti-ferroelectric transistor. Nature Communications 13 , (2022) Additional Declarations There is NO Competing Interest. Supplementary Files VPSIV6.pdf Supplementary Information Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-2441360","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":170285519,"identity":"c765009a-35cd-4f2d-9d67-bd31844962cc","order_by":0,"name":"Xiaoxian Liu","email":"","orcid":"","institution":"Fudan University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Xiaoxian","middleName":"","lastName":"Liu","suffix":""},{"id":170285520,"identity":"7b619f45-d29c-41d5-a94b-03dd9de153c4","order_by":1,"name":"Shuiyuan Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA80lEQVRIiWNgGAWjYPACifp+5gMMEmD2AeK0WDDObEsgTUsF44ZjxGoxOH728MufbRLMxsd4DG/83MEgx3cjgfFzAT4tZ/LSrHnbJNjMjvEYW/aeYTCWvJHALD0DjxazAzlmxoxtEjxm93vMJHjbGBI33EhgY+bBp+X8GzNDoMMkjNt4zCT/tjHUE9ZyI8f4AdBhBgZsPGbSQFsSDAhpsb/xxoyZ55xEgsQxtmJr2TYJw5lnHjZL49Mi2Z9j/PFHWV0Cfxvzxptv22zk+Y4nH/yMTwsQsEkwssE5oKhhbMCvgYGB+QPDH0JqRsEoGAWjYEQDAGseSb3wo9ndAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-1979-7430","institution":"Fudan University","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Shuiyuan","middleName":"","lastName":"Wang","suffix":""},{"id":170285521,"identity":"c16109ba-fcd3-49db-b191-84242e734f01","order_by":2,"name":"Ziye Di","email":"","orcid":"","institution":"Fudan University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Ziye","middleName":"","lastName":"Di","suffix":""},{"id":170285522,"identity":"9bb0f390-b7cf-43b1-ad2e-caee8fc57136","order_by":3,"name":"Haoqi Wu","email":"","orcid":"","institution":"Fudan University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Haoqi","middleName":"","lastName":"Wu","suffix":""},{"id":170285523,"identity":"7572d085-b8be-4f22-9a79-621a50edac55","order_by":4,"name":"Chunsen Liu","email":"","orcid":"https://orcid.org/0000-0003-0842-7503","institution":"Fudan University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Chunsen","middleName":"","lastName":"Liu","suffix":""},{"id":170285524,"identity":"0bddb1f5-20b6-4498-b720-21718041abaa","order_by":5,"name":"Peng Zhou","email":"","orcid":"https://orcid.org/0000-0002-7301-1013","institution":"Fudan University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Zhou","suffix":""}],"badges":[],"createdAt":"2023-01-04 06:15:34","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2441360/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2441360/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":32125024,"identity":"a5ab6cea-4752-49b4-a7e9-a22856a771fe","added_by":"auto","created_at":"2023-01-27 15:28:45","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":227727,"visible":true,"origin":"","legend":"\u003cp\u003eStructure and characterization of VP and fabrication of heterostructure synapse. a, Schematic of the crystal structure of VP. The upper panel and lower panel represent the cross-section and top view of the VP lattice, respectively. The violet spheres represent phosphorus atoms while gold beams indicate the configuration of atoms. b, Raman spectrum of the heterostructure device with excitation laser of 532 nm. The blue, violet and red curve represent the Raman peak of MoS\u003csub\u003e2 \u003c/sub\u003e(383 cm\u003csup\u003e-1\u003c/sup\u003e and 410 cm\u003csup\u003e-1\u003c/sup\u003e), VP (183, 211, 278, 361, 379 and 476 cm\u003csup\u003e-1\u003c/sup\u003e), and silicon (525 cm\u003csup\u003e-1\u003c/sup\u003e) respectively. c, Transfer and output curves of VP phototransistor in darkness and under illumination. d, The current of VP phototransistor over 14 days. e, Schematic of the synaptic devices, which consists of Device I (right, VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure) and Device II (left, MoS\u003csub\u003e2\u003c/sub\u003e). f, Fabrication flow of the synaptic device, including transferring, source/drain forming, top dielectric deposition and top gate forming. g, SEM image of the synaptic device. Scale bar: 5 μm. h~j, HAADF STEM image of the synaptic device under different magnifications. The STEM image shows a clean van der Waals interface and the VP lattice structure is consistent with that in a. k, Cross-section STEM image of the VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure with corresponding EDS mapping. Scale bar: 2 nm.\u003c/p\u003e","description":"","filename":"floatimage11.jpg","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/47afeb0d8584c6ad34f1db78.jpg"},{"id":32125943,"identity":"e84f17e9-0138-4cf4-935d-ac008aab92ee","added_by":"auto","created_at":"2023-01-27 15:36:45","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":96779,"visible":true,"origin":"","legend":"\u003cp\u003eEnergy band analysis and COMSOL finite element simulation of VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic device. a, Schematic of the strategy to obtain large dynamic ranges. When light stimulation is applied, a threshold shift is induced, creating a “threshold window”. By selecting the operation point within this window, a dynamic range as large as 10\u003csup\u003e6\u003c/sup\u003e could be achieved with weak light. b, Schematic of the energy bands in different states during the optical potentiation and electrical depression process. When a negative voltage is applied on the top gate, a potential well is formed at the interface of VP and MoS\u003csub\u003e2\u003c/sub\u003e. Then electron-hole pairs are generated by light stimulation and as the holes are trapped by the potential well, the threshold voltage drifts negatively. c, Hole distribution under different conditions in the heterostructure based on COMSOL simulation. The dash lines show the isoline of the hole density.\u003c/p\u003e","description":"","filename":"floatimage2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/a0d0fb71d30488ed7c18d58e.jpg"},{"id":32125031,"identity":"b8c29b8a-89cf-4f9a-be5f-9052f06b8b5b","added_by":"auto","created_at":"2023-01-27 15:28:45","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":73761,"visible":true,"origin":"","legend":"\u003cp\u003eThreshold drift and optoelectronic characteristics of the VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic device. a, Transfer curve of the two structured devices under different laser illuminations. V\u003csub\u003eds\u003c/sub\u003e is fixed at 1 V and the laser wavelength is 473 nm. The heterostructure device shows a considerable threshold shift compared to the MoS\u003csub\u003e2\u003c/sub\u003e transistor. b, Current mapping under different V\u003csub\u003etg\u003c/sub\u003e and laser intensities. The dashed line indicates the trend of threshold voltage shift. c, Output curve of the synapse device, where V\u003csub\u003etg\u003c/sub\u003e is fixed at -8 V. d, Output waveforms at different V\u003csub\u003etg\u003c/sub\u003e, when a single laser pulse is applied. The PSC can reach ~1 μA regardless of V\u003csub\u003etg\u003c/sub\u003e. e, Performance comparison between the two structured devices. The VP-MoS\u003csub\u003e2 \u003c/sub\u003edevice could maintain a higher excitement ratio and PSC amplitude than the MoS\u003csub\u003e2 \u003c/sub\u003edevice.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/78355cf77f4a7e8ca16c2727.png"},{"id":32125028,"identity":"c12a4059-9223-4e1d-9460-5d3c7eb05b30","added_by":"auto","created_at":"2023-01-27 15:28:45","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":186744,"visible":true,"origin":"","legend":"\u003cp\u003eShort- and long-term synaptic plasticity in VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure synapse. a, PPF indexes at different intervals. The inserted image shows a typical waveform of the PPF characteristics. The dashed line indicates the fitting curve at the experimental points. b, Variation of PSC gain for different optical intensities and base current (A\u003csub\u003e0\u003c/sub\u003e) during 30 pulses of stimulation. The frequency of the laser spikes is fixed at 2 Hz with a 473 nm laser. c, PSC gain as a function of top gate voltage and pulse numbers. The left panel shows a typical waveform of short-term plasticity, which gradually changes into long-term plasticity as the number of pulses increases. d, Normalized LTP and LTD with 30,128 and ~180 conductance states. Optical and electrical spikes are applied for the potentiation and depression processes, respectively. The waveforms used are shown in the insert. e, Specific waveforms of every state extracted from d, which indicates distinguishable and stable conductance states. f, Dynamic range and conductance state statistics for emerging artificial synapses, including charge trapping, ferroelectric, memristors, and optoelectronic devices. The VP-MoS\u003csub\u003e2\u003c/sub\u003e device shows a record high dynamic range of ~10\u003csup\u003e6\u003c/sup\u003e and 128 multi-states, far better than the mainstream analog memories.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/d23d5c8cba4ae03600267ce7.png"},{"id":32125032,"identity":"89ffd1b7-57ff-405c-8847-acf1e39f61c7","added_by":"auto","created_at":"2023-01-27 15:28:45","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":198187,"visible":true,"origin":"","legend":"\u003cp\u003eImage classification of MNIST and Fashion-MNIST datasets using VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic devices. a, Schematic of the neural network model used for image classification simulation. b, Simulated weight distributions before and after training, including W\u003csub\u003eIH\u003c/sub\u003e (40000 synapses) and W\u003csub\u003eHO\u003c/sub\u003e (1000 synapses). c, The learning accuracy as a function of training epochs for both datasets. The red, blue and violet lines represent the ideal device and the large-dynamic-range VP-MoS\u003csub\u003e2\u003c/sub\u003e device with 128 and 30 conductance states, respectively. Thanks to the extremely large dynamic range, the heterostructure device with 128 states shows negligible distinction from the ideal device in both classification tasks. d, Dependence of classification accuracy on different dynamic ranges and state numbers. Dynamic ranges and state numbers affect accuracy over a wide range, especially when the dynamic range is below 50, where accuracy is severely degraded. The infinity dynamic range is achieved by setting the off-state current to 0 (ideal condition). e, The learning error of synaptic devices in classification tasks with different complexity. The dataset complexity is defined as the average information entropy of images. For tasks with higher complexity, synaptic devices generally show a larger error rate, while VP-MoS\u003csub\u003e2 \u003c/sub\u003edevice could maintain a low error rate thanks to the large dynamic range and sufficient conductance states.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/5079583cb9ebb8527d5e64b9.png"},{"id":32125945,"identity":"b9841658-3560-44f3-ae87-0d930cf2429e","added_by":"auto","created_at":"2023-01-27 15:36:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1020359,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/a1ef7f27-639b-4805-877c-bcd5319496d4.pdf"},{"id":32125944,"identity":"c061d771-4f28-431d-8eeb-6f40a4c5f9c9","added_by":"auto","created_at":"2023-01-27 15:36:45","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1150916,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Information\u003c/p\u003e","description":"","filename":"VPSIV6.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2441360/v1/a5bef65fe9eb7772ef3af866.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"A Large-Dynamic-Range Violet Phosphorus Heterostructure Optoelectronic Synapse for High-Complexity Neuromorphic Computing","fulltext":[{"header":"Introduction","content":"\u003cp\u003eEmerging artificial intelligence applications such as image recognition, motion detection, and autonomous driving etc. require processing massive amounts of data, which poses a serious challenge to the von Neumann architecture that separates memory from computation\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Brain-inspired neuromorphic computing offers advantages in terms of energy efficiency and operational latency, bringing a promising approach to this challenge\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. In the neuromorphic system, synaptic devices are considered as the core units due to their far greater number than neurons and their properties are essential to achieve efficient, high-precision computation\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Among all the characteristics, the dynamic range determines the capability of mapping device conductance to weights in neural network algorithms, which is crucial in neuromorphic computing, especially for inference and learning tasks\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. A larger dynamic range means the potential to obtain more conductance states while maintaining linearity. It was reported that MNIST handwriting recognition accuracy improves significantly with increasing dynamic range below 30, and the desire for larger dynamic range is more pressing for complicated tasks\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. However, the dynamic range of most memory devices in the analog switching regime is typically 10\u0026thinsp;~\u0026thinsp;100, unlike the high on/off ratios available in binary switching\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. The limited dynamic range and conductance states constrains the accuracy, although it could be compensated by optimized linearity/symmetry.\u003c/p\u003e \u003cp\u003eDue to their attractive characteristics, two-dimensional (2D) materials represented by molybdenum disulfide (MoS\u003csub\u003e2\u003c/sub\u003e) appear to be promising candidates to build synapse devices, as their atomic-scale thickness could reduce operation voltage and energy consumption\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Furthermore, their strong resonant light absorption (\u0026gt;\u0026thinsp;20%)\u003csup\u003e9\u003c/sup\u003e and better electrostatics control make them suitable for building optoelectronic synapses with large dynamic ranges. In addition, 2D phosphorene such as black phosphorus have also been introduced to construct optoelectronic synapses, but their stability is still a problem\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Violet phosphorus (VP), considered to be the most stable allotrope of phosphorene, has recently been successfully synthesized\u003csup\u003e\u003cspan additionalcitationids=\"CR13 CR14 CR15 CR16\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. With excellent optoelectronic properties, including a direct bandgap of ~\u0026thinsp;2.54 eV\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, extremely low dark current, and large light-to-dark ratio, VP becomes a potential alternative for optoelectronic synaptic devices. However, current research on VP has focused on material synthesis and characterization, while its application in devices has been left vacant.\u003c/p\u003e \u003cp\u003eHere, we present a VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure synapse that attempts the optoelectronic device application of VP. The device exhibits an ultra-high dark-to-light ratio due to the relatively large bandgap and strong light-matter interactions of VP. The optoelectronic coupling and charge transfer within VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure lead to a strong threshold shift effect when weak light is applied. Thanks to this phenomenon, a dynamic range of over 10\u003csup\u003e6\u003c/sup\u003e with distinguishable conductance multi-states of 128 can be achieved at a light intensity of ~\u0026thinsp;8 \u0026micro;W. Moreover, we demonstrated bionic synaptic behaviors with the heterostructure device, including short-term and long-term potentiation (LTP) using light stimulation and long-term depression (LTD) using electric stimulation. Compared with mainstream analog memories, our heterostructure device owns ultralow off-state current in synaptic behaviors (large dynamic range), which could minimize the deviation in the conductance-weight mapping (the lowest but not zero conductance state in the device is mapped into the absolute zero weight in algorithm, thus bringing deviation between device conductance and algorithm weight in every following states), resulting in improvement in learning accuracy. Using VP-MoS\u003csub\u003e2\u003c/sub\u003e device conductance mapping, we simulated MNIST handwriting digit and high-complexity Fashion-MNIST image classification tasks to evaluate and validate the impact of dynamic range on accuracy. The accuracy was able to reach 95.23% and 79.65%, respectively, which is comparable to the ideal device (95.47%, 79.95%). This work opens the way for the use of VP in optoelectronics and provides a feasible strategy for large-dynamic-range synapses, thus further contributing to the accuracy improvement in high-complexity neuromorphic computation.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea shows the lattice structure of VP, which appears to be monoclinic with a space group of P\u003csub\u003e2/n\u003c/sub\u003e, with crystallographic lattice constants of a\u0026thinsp;=\u0026thinsp;9.210 \u0026Aring;, b\u0026thinsp;=\u0026thinsp;9.128 \u0026Aring;, c\u0026thinsp;=\u0026thinsp;21.893 \u0026Aring;, and β\u0026thinsp;=\u0026thinsp;97.776\u0026deg;\u003csup\u003e12\u003c/sup\u003e. The VP structure generally shows a bi-tubular structure with one layer stacked on another vertically along the z-direction, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea lower panel. The Raman spectrum of VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure excited by a 532 nm laser is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb. The Raman shift peaks at 183, 211, and 278 cm\u003csup\u003e-\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e represent for the variation modes of VP atoms while the peaks at 361, 379 and 476 cm\u003csup\u003e-\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e represent for the stretching of the atom cages as a whole. The complex Raman spectrum indicates a large density of photon states in VP, which further contributes to the strong light-matter interaction and electron-phonon scatterring\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. In addition, few-layered VP possesses a direct bandgap of ~\u0026thinsp;2.54 eV\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e and usually behaves as an n-type semiconductor, which indicates its potential to build optoelectronic devices. On this basis, we built VP photo-transistors and measured their characteristics, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec,d. Due to its unique lattice and bandgap characteristics, VP-based photodevice shows an extremely low dark current (~\u0026thinsp;fA) and a high light-to-dark ratio (~\u0026thinsp;10\u003csup\u003e5\u003c/sup\u003e). In addition, the VP phototransistors were exposed to the ambient air without any encapsulation and could still exhibit an on/off ratio of ~\u0026thinsp;100 and a light-to-dark ratio of ~\u0026thinsp;10\u003csup\u003e3\u003c/sup\u003e after 14 days, while black phosphorous would completely lose on/off characteristics in 2 days\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e, indicating that VP is the most stable phosphorus, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed (original data see \u003cb\u003eSupplementary Figure S1\u003c/b\u003e). However, as the photo-response of VP phototransistors is generally instantaneous (see \u003cb\u003eFigure S2\u003c/b\u003e in Supplementary Information), which is not suitable for synaptic applications, heterostructure was then introduced. The device schematic and process flow of our heterostructure device is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ee and Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ef, respectively. MoS\u003csub\u003e2\u003c/sub\u003e and VP were successively transferred to a 300 nm-thick silicon oxide substrate to form the vertically stacked heterostructure (Device I). Electron beam lithography (EBL) was used to form the electrode pattern and electron beam evaporation (EBE) deposits the source and drain electrodes. Top gate dielectric (30 nm HfO\u003csub\u003e2\u003c/sub\u003e) was then deposited using atomic layered deposition (ALD) with 1 nm seed layer (SiO\u003csub\u003e2\u003c/sub\u003e) using EBE, followed by top gate electrode deposition. Another MoS\u003csub\u003e2\u003c/sub\u003e device (Device II) without VP was built using the same material as the control. The VP stacked here on the top of MoS\u003csub\u003e2\u003c/sub\u003e works as a photogate, and the thickness of the dielectric and top gate have been carefully determined to ensure the transmission of light. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eg shows the scanning electron microscope (SEM) image of the heterostructure synaptic device. High-resolution scanning transmission electron microscopy (STEM) was used to characterize the device microstructure. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eh\u0026thinsp;\u003cb\u003e~\u0026thinsp;j\u003c/b\u003e shows the cross-section HAADF STEM images of the VP-MoS\u003csub\u003e2\u003c/sub\u003e under different magnifications, which implies a layered van der Waals structure and is consistent with the lattice in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea. The STEM image in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ej shows a single-layer VP thickness of ~\u0026thinsp;2.2 nm, which is consistent with the theoretical value\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ek shows the STEM image of the VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure and the corresponding energy dispersive spectroscope (EDS) mapping, which exhibits a clean van der Waals interface between VP and MoS\u003csub\u003e2\u003c/sub\u003e layers. The heterostructure region was further investigated using atomic force microscope (AFM), which is provided in \u003cb\u003eFigure S3\u003c/b\u003e in the Supplementary Information. The thickness of VP and MoS\u003csub\u003e2\u003c/sub\u003e is measured by AFM to be ~\u0026thinsp;6.66 nm and 5.85 nm, corresponding to 3 and 9 layers, respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe operating principle of our VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic device is described in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea. When a voltage is applied to the top gate, the device exhibits an n-type on/off switch, as shown by the dashed line in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea left panel. Subsequently, when a weak light stimulus is applied, a negative threshold shift of several V can be observed, leading to a \u0026ldquo;threshold window\u0026rdquo;, which is shown by the solid line in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea left panel. Different from the ordinary photocurrent effect where the off-state current increase obviously under illumination, here the photostimulation mainly act as a trigger to switch the on/off state instead of increasing the off-state current. Similar phenomena can be found in other heterostructures as well. By carefully selecting the operation point, the former off-state can be transformed to the on-state by light induction, resulting in a record large optical dynamic range (over 10\u003csup\u003e6\u003c/sup\u003e), far better than most analog switching memories (10\u0026thinsp;~\u0026thinsp;100), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea right panel. It is worth noting that the operation point is typically negative, which means that our device could maintain a high response at a lower off-state current than most mainstream analog memories, leading to great improvement in dynamic range. To further investigate the principle of threshold shift, we construct an energy band model for the heterostructure device (Device I), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb. The theoretical bandgaps of MoS\u003csub\u003e2\u003c/sub\u003e and VP are about 1.9 eV\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e and 2.54 eV\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, respectively. When stacked together without external voltage (flat band), VP and MoS\u003csub\u003e2\u003c/sub\u003e form a type-II heterostructure, with the bottom of the conductance band of MoS\u003csub\u003e2\u003c/sub\u003e higher than VP and the top of the valence band lower than VP. When a negative voltage is applied to the top gate, the device changes from the flat band to the programmed state. Photostimulation is then applied to the device in the programmed state to induce photogenerated electron-hole pairs. According to the energy band structure, electrons and holes move separately: holes move toward VP and are trapped by the potential well at the interface of VP and HfO\u003csub\u003e2\u003c/sub\u003e (the properties of VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure device without the top gate dielectric are shown in \u003cb\u003eSupplementary Figure S4\u003c/b\u003e, where the separated photogenerated holes are not efficiently trapped, making the threshold shift rather weak and leading to a limited dynamic range); while electrons move toward MoS\u003csub\u003e2\u003c/sub\u003e, leading to an increase in transient channel current. Upon withdrawal of the light stimulus, the trapped holes contribute to the conductivity of the channel, thus resulting in a negative threshold shift. Through applying a positive voltage to the top gate, the band structure changes, leading to a release process of the trapped holes. As a result, the device is reset to the initial state. In this way, the heterostructure device shows the coexistence of optical potentiation and electrical inhibiting, which exactly mimic biological excitatory and inhibitory plasticity and can be mapped to artificial neural networks for neuromorphic computing. As a comparison, we construct another energy band model without VP (Device II) to further investigate the role played by VP in such a process, see \u003cb\u003eSupplementary Figure S5\u003c/b\u003e. Without the strong light-matter interaction with VP, MoS\u003csub\u003e2\u003c/sub\u003e produces limited electron-hole pairs. More importantly, in this case, the separation of electrons and holes cannot be captured efficiently, so only a transient photocurrent exists without a significant threshold change. To confirm the validity of our energy band model, we further use COMSOL Multiphysics to conduct finite element simulation based on the above theoretical analysis. More details about the simulation could be found in \u003cb\u003eSection 6\u003c/b\u003e in Supplementary Information. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec depicts the distribution of holes in the heterostructure device under different conditions. When a positive voltage is applied at the top gate without light stimulation, the heterostructure device exhibits on current state with low hole densities, due to the fact that both VP and MoS\u003csub\u003e2\u003c/sub\u003e are n-type semiconductors. When the light stimulation was applied to the device, electron-hole pairs are generated and separated, with holes moving towards VP and trapped, resulting in the distribution shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec upper panel. In this case, although the concentration of carriers increases overall, the current of the device does not change much because the device is still in on-state. However, when a negative voltage is applied, the heterostructure device is set to off-state. When light stimulation is applied, the accumulation of holes in VP leads to threshold voltage shift, which turn the device from a non-conductive state into a conductive state, leading to a large dynamic range through light stimulation, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec lower panel.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eExperiments are implemented to verify these analyses. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea \u003cb\u003eupper panel\u003c/b\u003e shows the transfer curve of the VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure device under varying illuminations, where the source-drain voltage (V\u003csub\u003eds\u003c/sub\u003e) is fixed at 1 V and the laser wavelength is 473 nm. Obvious threshold shift can be observed for illumination as low as 6 \u0026micro;W and expand with increasing intensity. The actual laser intensity may be lower, taking into account the influence of the top gate dielectric and electrodes. As the intensity increases to 20 \u0026micro;W, the threshold voltage has changed from about \u0026minus;\u0026thinsp;4 V to below \u0026minus;\u0026thinsp;8 V, which is desired for the large dynamic range. According to the transfer curve, when the top gate voltage (V\u003csub\u003etg\u003c/sub\u003e) is fixed in this range (-4 V to -8 V), a 20 \u0026micro;W optical stimulation could lead to a large light on/off ratio up to over 10\u003csup\u003e6\u003c/sup\u003e. In addition, we test the transfer curve of Device II as a control, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea \u003cb\u003elower panel\u003c/b\u003e. In this device, almost no change in threshold voltage is observed even when 30 mW of strong light is applied, which is in line with our previous analysis. To visualize the threshold shift of the VP-MoS\u003csub\u003e2\u003c/sub\u003e device, we collected the source-drain currents for different top gate voltages as well as for different light intensities, as shown by color mapping in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb (see original curves in \u003cb\u003eFigure S8\u003c/b\u003e in Supplementary Information). Considering 10 nA as the boundary between on and off switching, the dashed line could reflect the shift in threshold voltage, that is, the threshold changes from \u0026minus;\u0026thinsp;12 V to -16 V, which leaves a \u0026ldquo;threshold window\u0026rdquo; of ~\u0026thinsp;4 V, sufficient for synaptic device operation. We chose an operation point of -4 V and test the output curve under varying light intensities, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec. At such an operation point, the output current increases to around 1 \u0026micro;A under 8 \u0026micro;W illumination, which is consistent with the transfer curve. Such a phenomenon has been repeatedly verified in different batches of devices with this heterostructure (\u003cb\u003eFigure S7\u003c/b\u003e in Supplementary Information). Although the threshold voltage varies slightly with the thickness of 2D materials, this phenomenon could be reproduced in each device, which means that there is a stable strategy to improve the optical dynamic range. Next, we stimulate the device with a single laser spike and test its response, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed. When it comes to neuromorphic computing, the laser spike here could simulate presynaptic input, and the channel current is monitored as post-synaptic current (PSC). The stimuli are applied at different operation points (different V\u003csub\u003etg\u003c/sub\u003e). Although the base current of the device decreases with increasing V\u003csub\u003etg\u003c/sub\u003e, the PSCs almost all reach the \u0026micro;A level, which is consistent with the on-state current of our heterostructure device. As a result, the excitement ratio (PSC/base current) increases rapidly as V\u003csub\u003etg\u003c/sub\u003e increases, and can exceed 10\u003csup\u003e6\u003c/sup\u003e thanks to the extremely low dark current of VP. Similar tests are carried out using Device II (see \u003cb\u003eFigure S9\u003c/b\u003e in Supplementary Information). A comparison of the two devices is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee, where the squares represent the performance of the VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure device (Device I) and the circles represent the MoS\u003csub\u003e2\u003c/sub\u003e transistor (Device II). To eliminate the effect of different original threshold voltages, the x-axis has been unified as base current rather than top gate voltage. Regarding the PSC amplitude (left axis), the MoS\u003csub\u003e2\u003c/sub\u003e transistor shows high PSC amplitude only when the base current itself is relatively high, whereas VP-MoS\u003csub\u003e2\u003c/sub\u003e exhibits high PSC amplitude irrespective of the base current. This discrepancy leads to at least a 3-magnitude improvement in excitement ratio (right axis), representing a stronger synaptic response to the stimulus.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe further explored the potential of VP-MoS\u003csub\u003e2\u003c/sub\u003e device for the simulation of synaptic plasticity and behavior. Dual laser pulses with different intervals were applied to the device to test the paired-pulse facilitation (PPF) characteristics, which is essential to simulate biologically short-term plasticity (STP). The inset in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea shows a output waveform, where the base current is around 100 fA, and A\u003csub\u003e1\u003c/sub\u003e and A\u003csub\u003e2\u003c/sub\u003e represent the amplitudes after the first and second laser pulses. The second pulse exhibits a stronger response than the first pulse, and quickly recovers to the base current, showing typical PPF characteristics. Waveforms based on different interval times are shown in \u003cb\u003eFigure S10\u003c/b\u003e in Supplementary Information. When the interval is as low as 150 ms, our device shows a fairly high PPF index (determined by A\u003csub\u003e2\u003c/sub\u003e/A\u003csub\u003e1\u003c/sub\u003e) of up to 853%, and gradually recovers to 100% as the interval increases above 3000 ms. The dashed line here indicates the fitting curve at the experimental points, which obeys an exponential decay, consistent with the theoretical result\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Moreover, the long-term plasticity of the heterostructure synaptic device was explored by increasing the number of laser pulses. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb, 30 laser spikes with different intensities were applied to achieve a progressive excitatory PSC modulation, which simulated the LTP plasticity. The PSC gain (determined as A\u003csub\u003e30\u003c/sub\u003e/A\u003csub\u003e1\u003c/sub\u003e) increases significantly as the base current A\u003csub\u003e0\u003c/sub\u003e decreases (increase in V\u003csub\u003etg\u003c/sub\u003e) as well as the increased laser power. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec shows the PSC gain (A\u003csub\u003en\u003c/sub\u003e/A\u003csub\u003e1\u003c/sub\u003e) as a function of top gate voltage and pulse number. When the pulse number is as low as 5, the synapse device generally exhibits STP, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec \u003cb\u003eleft panel\u003c/b\u003e. In this case, the PSC increases linearly with the accumulation of laser spikes and falls back to the base current a few seconds after the removal of laser. The device shows a clear transition from STP to LTP as the pulse number increases, with an increase in PSC gain and retention time (\u003cb\u003eFigure S11\u003c/b\u003e and \u003cb\u003eFigure S12\u003c/b\u003e in Supplementary Information). In addition, a higher top gate voltage will also significantly increase the PSC gain, leading to a large dynamic range of over 10\u003csup\u003e6\u003c/sup\u003e (\u003cb\u003eFigure S12\u003c/b\u003e in Supplementary Information). However, the extremely high dynamic range is achieved at the expense of linearity and retention time, which are also critical for neuromorphic computing. To balance all these key metrics, the operation point was carefully selected to achieve a large dynamic range (~\u0026thinsp;10\u003csup\u003e6\u003c/sup\u003e) with 30 conductance states, fair linearity (1.31) and retention time (~\u0026thinsp;40 s), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed \u003cb\u003ered curve\u003c/b\u003e (more details see \u003cb\u003eFigure S13\u003c/b\u003e in Supplementary Information), which could meet the requirements of high-precision neuromorphic computing\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. A large dynamic range means the potential to contain more conductance states. To further investigate the capability of the heterostructure device to obtain distinguishable conductance multi-states, we applied more pulses for both potentiation and depression. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed shows the normalized long-term synaptic plasticity using optical pulses for potentiation and electrical pulses for depression under different conditions. By applying 128 light pulses for stimulation, 128 stable, non-crossing, and distinguishable conductance states are generated. The original waveforms are shown in \u003cb\u003eFigure S14\u003c/b\u003e in Supplementary Information, with a base state of around 100 fA, and after 128 pulses of stimulation, the current rises to\u0026thinsp;~\u0026thinsp;\u0026micro;A, implying a dynamic range of over 10\u003csup\u003e6\u003c/sup\u003e, which indicates a strong capability to map synaptic device conductance to neural network weights. In addition, by applying over 200 pulses, we can obtain up to ~\u0026thinsp;180 distinguishable conductance states (original waveforms shown in \u003cb\u003eFigure S15\u003c/b\u003e in Supplementary Information). Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ee depicts the waveforms of each state extracted from Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed, which are distinguishable and stable. The device also exhibits an extremely high A\u003csub\u003en\u003c/sub\u003e/A\u003csub\u003e1\u003c/sub\u003e of ~\u0026thinsp;10\u003csup\u003e4\u003c/sup\u003e, showing the potential to obtain more conductance states. However, limited by the instability of light stimulation and strong light response of the device, state intervals smaller than ~\u0026thinsp;nA hardly exist stably, limiting the further increase in the number of states. The dynamic ranges and number of conductance state statistics for synaptic devices based on emerging analog memories are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ef. The devices are divided into four categories according to working mechanisms: charge-trapping electrical devices\u003csup\u003e\u003cspan additionalcitationids=\"CR23 CR24 CR25 CR26\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e, ferroelectric devices\u003csup\u003e\u003cspan additionalcitationids=\"CR29 CR30 CR31 CR32\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e, electrical memristors\u003csup\u003e\u003cspan additionalcitationids=\"CR35 CR36 CR37\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e, and optoelectronic devices\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan additionalcitationids=\"CR40 CR41 CR42\" citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e, which are based on a variety of promising materials, including 2D, organic, oxide (e.g. indium gallium zinc oxide) and perovskite semiconductors. Thanks to the extremely low off-state current, our VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure synapse exhibits a record high dynamic range of over 10\u003csup\u003e6\u003c/sup\u003e, as well as 128 distinguishable conductance multi-sates, far outperforming that of current mainstream analog memories, providing a new strategy for improving the dynamic range and multi-states in synaptic devices.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFinally, we used the NeuroSim multilayer perceptron (MLP) neural network simulator\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e to validate the ability of VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic device with a large dynamic range for high-complexity image classification tasks. The neural network used is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea, which consists of an input layer, a hidden layer and an output layer. Each neuron node in one layer is connected to each node in the following layer, forming a fully connected neural network. Neuron nodes are connected via synaptic devices, and device conductance represents network weights. W\u003csub\u003eIH\u003c/sub\u003e and W\u003csub\u003eHO\u003c/sub\u003e represents the weight matrix between the input and hidden layers and between the hidden and output layers, respectively. We have made the necessary modifications to the original network model in the NeuroSim simulator to make it suitable for our classification tasks. We performed the image classification based on two standard datasets: MNIST and Fashion-MNIST (an MNIST-like dataset with higher complexity)\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e. The input image data has been pre-processed into grayscale data of each pixel as the input layer (20\u0026times;20 for MNIST and 28\u0026times;28 for Fashion). The network contains 100 hidden neurons and 10 output neurons (referring to 10 kinds of labels, i.e. handwriting digits or objects), more details about the simulation could be found in \u003cb\u003eSection 16\u0026ndash;19\u003c/b\u003e in Supplementary Information. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb shows the distribution of weights before and after MNIST training, consisting of W\u003csub\u003eIH\u003c/sub\u003e and W\u003csub\u003eHO\u003c/sub\u003e. The initial weights for both matrices are set randomly from 7 states (0, \u0026plusmn;\u0026thinsp;1, \u0026plusmn;0.33, \u0026plusmn;\u0026thinsp;0.66), and the weights after training are updated to 128 states, corresponding to the 128 conductance states of VP-MoS\u003csub\u003e2\u003c/sub\u003e device, indicating the update of the network weights.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe classification accuracy as a function of training epochs is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec. According to the simulation, our device eventually reach an accuracy of 95.23% for MNIST and 79.65% for the Fashion dataset, which is close to those of ideal devices (95.47% and 79.95%). As the original algorithm of the simulator is designed especially for the 20\u0026times;20 MNIST dataset, the learning accuracy for the Fashion dataset is relatively lower but still could reflect the superiority of our device by comparing it with the ideal device. The detailed parameters and results of our simulation could be founded in \u003cb\u003eTable S2\u003c/b\u003e in Supplementary Information. It is worth mentioning that in the actual classification process, the conductance states of real devices are mapped into the synaptic weights in the algorithm, where the lowest conductance state is transferred into weight 0. However, due to the physical limitation of the device, the absolute 0 weight is unreachable, which would affect all the conductance-weight mapping and finally lead to accuracy deviation between the ideal condition and the physical device\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Compared with most mainstream analog memories, our heterostructure synaptic device owns a lower off-state current and could minimize such deviation, thus leading to negligible error from ideal cases. Therefore, we further investigated the dependence of classification accuracy on dynamic range. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed shows the final classification accuracy for different dynamic ranges and the number of conductance states while keeping other parameters constant (consistent with the previous simulation for VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure device). In these cases, the change in dynamic range occurs with a decrease in the off-state while keeping the on-state fixed. Simulation results show that for simple classification tasks like MNIST, accuracy is strongly suppressed when the dynamic range is below 10, and for dynamic range over 100, the increase of dynamic range could still lead to improvement in learning accuracy. In addition, the increase in states number will also improve learning accuracy significantly. In classification tasks with higher complexity like Fashion-MNIST, such dependence becomes more pronounced. In Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ee, we further investigated the learning performance of different synaptic devices in image classification tasks with different complexity. The x-axis shows the complexity of different datasets, which are defined as the average information entropy of the images (See \u003cb\u003eSection 19\u003c/b\u003e in Supplementary Information) and the y-axis shows the error between ideal cases (software implementation) and physical devices. In those cases, most synaptic devices show relatively high error (mostly over 2%) due to their relatively low dynamic range and fewer state numbers\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan additionalcitationids=\"CR47 CR48 CR49\" citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e. By combining large dynamic ranges and sufficient state numbers, our VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic device shows negligible error from ideal cases, even for the high-complexity Fashion-MNIST classification task.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn conclusion, we have achieved the first device application of the most stable phosphene VP and proposed the VP-MoS\u003csub\u003e2\u003c/sub\u003e heterostructure for optoelectronic synapse. VP exhibits extremely low dark currents and high dark-to-light ratios and restricts photogenerated holes by forming the type-II heterostructure with MoS\u003csub\u003e2\u003c/sub\u003e, leading to a strong shift in threshold voltage. This phenomenon was exploited to realize optically induced synaptic plasticities with large dynamic ranges. By carefully selecting the operation point, key metrics including dynamic range, linearity and retention time could be balanced. A record dynamic range (over 10\u003csup\u003e6\u003c/sup\u003e) with a high number of conductance states (128) was obtained, which is superior to current synaptic devices. Furthermore, there is still room for improvement in these indicators. For example, the light-matter interaction could be enhanced by optimizing the design of the top gate dielectric. And the retention time is limited by the charge-trapping capability of the interface, which can be further improved by introducing additional physical fields. The larger dynamic range allows for more levels of conductance states, making it suitable for high-complexity tasks and further improving accuracy. We explored the potential of VP-MoS\u003csub\u003e2\u003c/sub\u003e synaptic device for MNIST and high-complexity Fashion-MNIST image classification through NeuroSim simulator and achieved accuracies comparable to that of ideal devices. The results indicate the significance of a large dynamic range for accuracy optimization. This work demonstrates the potential of VP as a unique optoelectronic material for device application and provides a viable strategy for high-precision neuromorphic computing with large-dynamic-range synaptic devices.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e \u003cb\u003eFabrication of the heterostructure synaptic device.\u003c/b\u003e Few-layer MoS\u003csub\u003e2\u003c/sub\u003e and VP were exfoliated and transferred to a silicon substrate with 300nm-thick SiO\u003csub\u003e2\u003c/sub\u003e, forming the heterostructure. The electrode pattern was defined by EBL and Cr/Au metal was deposited by EBE to form the source and drain. For the top gate dielectric, 2 nm SiO\u003csub\u003e2\u003c/sub\u003e was deposited as a seed layer using EBE, followed by ALD to form 30 nm HfO\u003csub\u003e2\u003c/sub\u003e as the dielectric layer. Then the top gate electrodes were formed again through EBL and EBE.\u003c/p\u003e \u003cp\u003e \u003cb\u003eFinite element simulation of heterostructure device.\u003c/b\u003e The simulation process is realized based on COMSOL Multiphysics. We combined the semiconductor module and electromagnetic waves module to simulate the generation, migration, and trapped process of photo-generate carriers. The material characteristics of VP and MoS\u003csub\u003e2\u003c/sub\u003e are set according to former research. The vertical dimension of the device is set to 10 nm, which is negligible compared with a channel length of 3 \u0026micro;m. The top gate dielectric (HfO\u003csub\u003e2\u003c/sub\u003e) is placed at the top of VP, where the Shockley-Read-Hall model is used to simulate the trap-assisted surface recombination process. Through the Newton iteration method, the carrier concentration and corresponding port current are calculated under different voltage and illumination conditions.\u003c/p\u003e \u003cp\u003e \u003cb\u003eCharacterization and measurement of the heterostructure synaptic device.\u003c/b\u003e The surface morphology of the heterostructure device was characterized by AFM, showing the thickness of VP and MoS\u003csub\u003e2\u003c/sub\u003e are about 6.66, and 5.85 nm, respectively. To examine the constructed van der Waals heterojunction interface, a cross-sectional analysis was performed using HAADF STEM technology with EDS elements mapping analysis. In addition, the 2D layered materials were characterized by Raman spectroscopy. VP showed strong peaks near 361, 379, 476 cm\u003csup\u003e-\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e, and MoS\u003csub\u003e2\u003c/sub\u003e showed peaks at 383, 410 cm\u003csup\u003e-\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. The Cascade probe station equipped with Keysight B1500A semiconductor analyzer was used to characterize the electrical properties of the heterostructure device and the simulation of synaptic plasticity under the ambient environment. The optical stimulation is applied through a 473 nm blue light laser and the corresponding light intensity is measured by a light intensity meter.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSimulation of image classification based on the fully connected neural network.\u003c/b\u003e The simulation process is realized based on the NeuroSim MLP simulator. A complete epoch includes loading the data set, defining a fully connected network model, training the network, testing the network and calculating the classification accuracy. The training process includes feed-forward and back-propagation processes. To execute the simulation using the Fashion-MNIST dataset, the topology of the network has been modified, and the learning rate has been carefully changed to optimize the accuracy.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eData availability\u003c/h2\u003e \u003cp\u003eThe data that support the findings of this study are available from the corresponding authors upon reasonable request.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eCode availability\u003c/h2\u003e \u003cp\u003eThe codes used for neural network simulation are available from the corresponding authors upon reasonable request.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the National Key Research and Development Program of China (2021YFA1200500), National Natural Science Foundation of China (61925402\u0026nbsp;and\u0026nbsp;62090032), Science and Technology Commission of Shanghai Municipality (19JC1416600),\u0026nbsp;and China\u0026nbsp;\u003cspan style=\"text-align: inherit;\"\u003ePostdoctoral Science Foundation (2022M720032).\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eX. Liu and S. Wang co-wrote the manuscript.\u0026nbsp;Z. Di, H. Wu and C. Liu helped in material characterization. P. Zhou conceived the idea and supervised the work. All authors provided suggestions for revisions and improvements to the work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdditional information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrespondence and requests for materials should be addressed to P Zhou or S. Wang.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWang S. Y., Chen X. Z., Huang X. H., Zhang D. W.\u0026amp;Zhou P. Neuromorphic Engineering for Hardware Computational Acceleration and Biomimetic Perception Motion Integration. Adv Intell Syst-Ger \u003cb\u003e2\u003c/b\u003e, (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoo H. J. Intelligence on Silicon: From Deep-Neural-Network Accelerators to Brain Mimicking AI-SoCs. Isscc Dig Tech Pap I \u003cb\u003e62\u003c/b\u003e, 20\u0026ndash;28 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoy K., Jaiswal A.\u0026amp;Panda P. Towards spike-based machine intelligence with neuromorphic computing. Nature \u003cb\u003e575\u003c/b\u003e, 607\u0026ndash;617 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang W. Q., et al. Neuro-inspired computing chips. Nature Electronics \u003cb\u003e3\u003c/b\u003e, 371\u0026ndash;382 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi S., Yang J.\u0026amp;Wang G. Emerging Memristive Artificial Synapses and Neurons for Energy-Efficient Neuromorphic Computing. Advanced Materials \u003cb\u003e32\u003c/b\u003e, (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi S., et al. SiGe epitaxial memory for neuromorphic computing with reproducible high performance based on engineered dislocations. Nature Materials \u003cb\u003e17\u003c/b\u003e, 335-+ (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S. Y., Liu X. X.\u0026amp;Zhou P. The Road for 2D Semiconductors in the Silicon Age. Advanced Materials (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S. Y., Zhang D. W.\u0026amp;Zhou P. Two-dimensional materials for synaptic electronics and neuromorphic systems. Sci Bull \u003cb\u003e64\u003c/b\u003e, 1056\u0026ndash;1066 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAn J. R., et al. Perspectives of 2D Materials for Optoelectronic Integration. Adv Funct Mater \u003cb\u003e32\u003c/b\u003e, (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTian H., et al. Anisotropic Black Phosphorus Synaptic Device for Neuromorphic Applications. Advanced Materials \u003cb\u003e28\u003c/b\u003e, 4991\u0026ndash;4997 (2016)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhmed T., et al. Optically Stimulated Artificial Synapse Based on Layered Black Phosphorus. Small \u003cb\u003e15\u003c/b\u003e, (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang L. H., et al. Structure and Properties of Violet Phosphorus and Its Phosphorene Exfoliation. Angew Chem Int Edit \u003cb\u003e59\u003c/b\u003e, 1074\u0026ndash;1080 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang B., et al. Cross structured two-dimensional violet phosphorene with extremely high deformation resistance. J Mater Chem A \u003cb\u003e9\u003c/b\u003e, 13855\u0026ndash;13860 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao R. Z., et al. Violet phosphorus quantum dots. J Mater Chem A \u003cb\u003e10\u003c/b\u003e, 245\u0026ndash;250 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang L. H., et al. Fast Identification of the Crystallographic Orientation of Violet Phosphorus Nanoflakes with Preferred In-Plane Cleavage Edge Orientation. Adv Funct Mater \u003cb\u003e32\u003c/b\u003e, (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRicciardulli A. G., Wang Y., Yang S.\u0026amp;Samori P. Two-Dimensional Violet Phosphorus: A p-Type Semiconductor for (Opto)electronics. J Am Chem Soc \u003cb\u003e144\u003c/b\u003e, 3660\u0026ndash;3666 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaumer F., et al. Synthesis, Characterization, and Device Application of Antimony-Substituted Violet Phosphorus: A Layered Material. Acs Nano \u003cb\u003e11\u003c/b\u003e, 4105\u0026ndash;4113 (2017)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y. Y., et al. Impurity-Induced Robust Trionic Effect in Layered Violet Phosphorus. Advanced Optical Materials \u003cb\u003e10\u003c/b\u003e, (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo Z. N., et al. Metal-Ion-Modified Black Phosphorus with Enhanced Stability and Transistor Performance. Advanced Materials \u003cb\u003e29\u003c/b\u003e, (2017)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMak K. F., Lee C., Hone J., Shan J.\u0026amp;Heinz T. F. Atomically Thin MoS2: A New Direct-Gap Semiconductor. Physical Review Letters \u003cb\u003e105\u003c/b\u003e, (2010)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZucker R. S.\u0026amp;Regehr W. G. Short-term synaptic plasticity. Annu Rev Physiol \u003cb\u003e64\u003c/b\u003e, 355\u0026ndash;405 (2002)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim S., et al. Pattern Recognition Using Carbon Nanotube Synaptic Transistors with an Adjustable Weight Update Protocol. Acs Nano \u003cb\u003e11\u003c/b\u003e, 2814\u0026ndash;2822 (2017)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRen Y., et al. Gate-Tunable Synaptic Plasticity through Controlled Polarity of Charge Trapping in Fullerene Composites. Adv Funct Mater \u003cb\u003e28\u003c/b\u003e, (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar M., Ban D. K., Kim S. M., Kim J.\u0026amp;Wong C. P. Vertically Aligned WS2 Layers for High-Performing Memristors and Artificial Synapses. Advanced Electronic Materials \u003cb\u003e5\u003c/b\u003e, (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S. Y., et al. A MoS2/PTCDA Hybrid Heterojunction Synapse with Efficient Photoelectric Dual Modulation and Versatility. Advanced Materials \u003cb\u003e31\u003c/b\u003e, (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNoh G., et al. Large Memory Window of van der Waals Heterostructure Devices Based on MOCVD-Grown 2D Layered Ge4Se9. Advanced Materials (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShim H., et al. An elastic and reconfigurable synaptic transistor based on a stretchable bilayer semiconductor. Nature Electronics (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeo M., et al. First Demonstration of a Logic-Process Compatible Junctionless Ferroelectric FinFET Synapse for Neuromorphic Applications. Ieee Electr Device L \u003cb\u003e39\u003c/b\u003e, 1445\u0026ndash;1448 (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJang S., et al. Ultrathin Conformable Organic Artificial Synapse for Wearable Intelligent Device Applications. Acs Appl Mater Inter \u003cb\u003e11\u003c/b\u003e, 1071\u0026ndash;1080 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim M. K.\u0026amp;Lee J. S. Ferroelectric Analog Synaptic Transistors. Nano Letters \u003cb\u003e19\u003c/b\u003e, 2044\u0026ndash;2050 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi B. C., et al. An Electronic Synapse Based on 2D Ferroelectric CuInP2S6. Advanced Electronic Materials \u003cb\u003e6\u003c/b\u003e, (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S. Y., et al. Two-dimensional ferroelectric channel transistors integrating ultra-fast memory and neural computing. Nature Communications \u003cb\u003e12\u003c/b\u003e, (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung W., Si M. W.\u0026amp;Ye P. D. D. First Demonstration of Ge Ferroelectric Nanowire FET as Synaptic Device for Online Learning in Neural Network with High Number of Conductance State and G(max)/G(min). Int El Devices Meet (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu R. J., et al. Vertical MoS2 Double-Layer Memristor with Electrochemical Metallization as an Atomic-Scale Synapse with Switching Thresholds Approaching 100 mV. Nano Letters \u003cb\u003e19\u003c/b\u003e, 2411\u0026ndash;2417 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRen Y., et al. Phosphorene nano-heterostructure based memristors with broadband response synaptic plasticity. Journal of Materials Chemistry C \u003cb\u003e6\u003c/b\u003e, (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu Z. W., et al. Ultrathin electronic synapse having high temporal/spatial uniformity and an Al2O3/graphene quantum dots/Al2O3 sandwich structure for neuromorphic computing. Npg Asia Mater \u003cb\u003e11\u003c/b\u003e, (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSangwan V. K., et al. Multi-terminal memtransistors from polycrystalline monolayer molybdenum disulfide. Nature \u003cb\u003e554\u003c/b\u003e, 500-+ (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePereira M. E., et al. Flexible Active Crossbar Arrays Using Amorphous Oxide Semiconductor Technology toward Artificial Neural Networks Hardware. Advanced Electronic Materials (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang J. X., et al. Deep-ultraviolet-triggered neuromorphic functions in In-Zn-O phototransistors. Applied Physics Letters \u003cb\u003e113\u003c/b\u003e, (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Q. T., et al. Photoelectric Plasticity in Oxide Thin Film Transistors with Tunable Synaptic Functions. Advanced Electronic Materials \u003cb\u003e4\u003c/b\u003e, (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Y., et al. Photonic Synapses Based on Inorganic Perovskite Quantum Dots for Neuromorphic Computing. Advanced Materials \u003cb\u003e30\u003c/b\u003e, (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Z. H., et al. All-in-one two-dimensional retinomorphic hardware device for motion detection and recognition. Nature Nanotechnology \u003cb\u003e17\u003c/b\u003e, 27-+ (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu K. Q., et al. An optoelectronic synapse based on alpha-In2Se3 with controllable temporal dynamics for multimode and multiscale reservoir computing. Nature Electronics (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen P. Y., Peng X. C.\u0026amp;Yu S. M. NeuroSim plus: An Integrated Device-to-Algorithm Framework for Benchmarking Synaptic Devices and Array Architectures. 2017 Ieee International Electron Devices Meeting (Iedm) (2017)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiao H, Rasul K, Vollgraf R. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShim H., et al. An elastic and reconfigurable synaptic transistor based on a stretchable bilayer semiconductor. Nature Electronics \u003cb\u003e5\u003c/b\u003e, 660\u0026ndash;671 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoshi V., et al. Accurate deep neural network inference using computational phase-change memory. Nature Communications \u003cb\u003e11\u003c/b\u003e, (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuo Z., et al. High-precision and linear weight updates by subnanosecond pulses in ferroelectric tunnel junction for neuro-inspired computing. Nature Communications \u003cb\u003e13\u003c/b\u003e, (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeo S., et al. An Optogenetics-Inspired Flexible van der Waals Optoelectronic Synapse and its Application to a Convolutional Neural Network. Advanced Materials \u003cb\u003e33\u003c/b\u003e, (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao R. R., et al. Compact artificial neuron based on anti-ferroelectric transistor. Nature Communications \u003cb\u003e13\u003c/b\u003e, (2022)\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-2441360/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2441360/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eNeuromorphic computing can efficiently handle data-intensive tasks and address the redundant data interaction required by traditional von Neumann architectures. Synaptic devices are essential components for neuromorphic computation. For high computational accuracy, synaptic devices need to retain good conductance linearity, but this leads to a limited dynamic range (10\u0026thinsp;~\u0026thinsp;100) and weight states, which impedes their processing of high-complexity tasks and restricts further advances in accuracy. Two-dimensional materials, such as transition metal disulfides and phosphorene, hold promise for the construction of synaptic devices with large dynamic ranges due to their strong light-matter interactions, while the stability of phosphorene remains an issue. Here, for the first time, we use the most stable violet phosphorene for device applications. The combination of violet phosphorene and molybdenum disulfide demonstrates an optoelectronic synapse with a record dynamic range of over 10\u003csup\u003e6\u003c/sup\u003e, benefiting from a significant threshold shift due to charge transfer and trapping in the heterostructure. Remarkable synaptic properties are demonstrated, including 128 distinguishable conductance states, electro-optical dependent plasticity, short-term paired-pulse facilitation, and long-term potentiation/depression. High-precision image classification with accuracies of 95.23% and 79.65% is achieved for MNIST and high-complexity Fashion-MNIST datasets, which is close to the ideal device (95.47%, 79.95%), indicating the potential of dynamic range and multi-states for optimizing accuracy. This work fills the device application gap of violet phosphorene and provides a strategy for building synaptic devices with large dynamic range to facilitate neuromorphic computing.\u003c/p\u003e","manuscriptTitle":"A Large-Dynamic-Range Violet Phosphorus Heterostructure Optoelectronic Synapse for High-Complexity Neuromorphic Computing","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-01-27 15:28:40","doi":"10.21203/rs.3.rs-2441360/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"2451119d-f63d-4521-b00b-23a9d65a32c2","owner":[],"postedDate":"January 27th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":18656977,"name":"Physical sciences/Materials science/Materials for devices/Electronic devices"},{"id":18656978,"name":"Physical sciences/Materials science/Nanoscale materials/Two-dimensional materials"}],"tags":[],"updatedAt":"2023-01-27T15:28:40+00:00","versionOfRecord":[],"versionCreatedAt":"2023-01-27 15:28:40","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-2441360","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2441360","identity":"rs-2441360","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}