Artificial neuron based on WSe2 electric anisotropy

Abstract

The invention discloses an artificial neuron based on WSe2 electrical anisotropy, comprising: a substrate; the WSe2 functional layer is formed on the substrate; a plurality of pairs of electrodes; the substrate is arranged on the WSe2 functional layer; the plurality of pairs of electrodes are arranged at intervals at a preset angle to form a multi-terminal field effect transistor structure; and the WSe2 functional layer realizes signal transmission between axons and multiple synapses through anisotropic carrier mobility induced by the intrinsic shielding layer. WSe serves as a core functional layer, the defect that traditional anisotropic materials such as black phosphorus and triangular selenium are sensitive to oxygen is overcome, and long-term reliable work of the device is guaranteed; meanwhile, the anisotropic carrier mobility is induced by an intrinsic shielding layer formed by W and Se atom clusters in WSe and an adsorption layer, lattice defects or impurity atoms do not need to be introduced by relying on defect engineering, and the problem of performance fluctuation caused by the fact that the defect types and the number are difficult to accurately control is avoided. And the device can keep stable signal response characteristics in different crystal directions.

Description

Technical Field

[0001] This invention relates to the field of fire safety, and in particular to an artificial neuron based on the electrical anisotropy of WSe2. Background Technology

[0002] Neurons consist of dendrites, a cell body, and an axon. Dendrites receive signals from other neurons, the cell body processes incoming signals and generates responses, and the axon transmits information to the next neuron. Figure 1 A) A fundamental property of neurons is the ability to receive and process multiple signals in parallel. For artificial neurons, the photoelectric anisotropy of two-dimensional materials plays a crucial role in axon-multisynaptic structures. This structure allows neurons to transmit information to multiple targets simultaneously, promoting integration and coordination to improve processing efficiency, adaptability, and learning capacity.

[0003] Many categories of two-dimensional materials have been applied to neuromorphic systems. Previous research first explored the inherent anisotropy in the crystal structure and band structure of materials. For example, Tian et al. simulated the heterogeneity of biological synapses, realizing an axon-multisynaptic network based on anisotropic two-dimensional black phosphorus. Qin et al. utilized the inherent low symmetry of triangular selenium (t-Se), with an anisotropy ratio as high as 8.6, and realized an inisotropic axon-multisynaptic network based on different signal processing along its c-axis and a-axis. However, materials with anisotropy are highly sensitive to oxygen, which is detrimental to device stability. Therefore, defect engineering has been developed by introducing specific defects, such as lattice defects, impurity atoms, or inherent defects, to modulate photoelectric anisotropy by controlling conduction in various directions. These defects can generate localized electronic states and alter the carrier mobility and recombination dynamics, thereby affecting the photoelectronic behavior of two-dimensional materials. For example, Vinod K et al. demonstrated heterosynaptic functionality based on a six-terminal MoS2 transistor by bias-induced defect motion in polycrystalline MoS2. Subsequently, Liu et al. utilized the anisotropic synaptic properties in initially anisotropic MoS2 devices through localized electron beam irradiation (EBI). However, due to fluctuations in material properties, the introduction of defects makes achieving consistent performance challenging. Controlling and optimizing the type and number of defects remains a complex and unresolved challenge in defect engineering. Recently, anisotropy has been achieved through microstructure engineering. Heterogeneous structures composed of superimposed two-dimensional material confinement exhibit interlayer electron interaction characteristics and induce anisotropy in terms of band alignment, carrier mobility, and optical properties between different layers. Furthermore, device structure design plays a crucial role in tuning optoelectronic performance.

[0004] Existing two-dimensional materials and artificial neuron technologies both suffer from significant drawbacks: inherently anisotropic materials lack stability, defect-engineered devices exhibit poor performance consistency, and microstructure and device design schemes face challenges such as structural complexity and high integration difficulty. These technological bottlenecks prevent existing artificial neurons from simultaneously meeting the comprehensive requirements of the neural ecosystem for functional diversity, stability, integration, and performance controllability, thus hindering the further development of neuromorphic electronics in complex multimodal signal processing scenarios. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is: insufficient stability and poor performance consistency.

[0006] The above-mentioned technical problems are solved by the following technical solution: This invention proposes an artificial neuron based on the electrical anisotropy of WSe2, comprising, substrate; The WSe2 functional layer is formed on the substrate; Multiple pairs of electrodes are disposed on the WSe2 functional layer; the multiple pairs of electrodes are arranged at preset angular intervals to form a multi-terminal field-effect transistor structure. The WSe2 functional layer enables signal transmission between axons and multiple synapses through the anisotropic carrier mobility induced by the intrinsic shielding layer.

[0007] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: the substrate is a silicon substrate with an oxide layer, and the WSe2 functional layer is prepared by physical vapor deposition.

[0008] In a preferred embodiment of the WSe2-based anisotropic artificial neuron of the present invention: the thickness of the WSe2 functional layer is 1-3 nm, and the Raman spectroscopy detection peak of the 2H phase of the WSe2 functional layer is 250 cm⁻¹. -1 250cm -1 and 309cm -1 .

[0009] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: the multiple pairs of electrodes are three pairs of diagonal Ag electrodes, and the preset angle is a 60° interval between the electrodes; the three pairs of diagonal Ag electrodes form a six-terminal field-effect transistor structure.

[0010] In a preferred embodiment of the artificial neuron based on the electrical anisotropy of WSe2 described in this invention: the WSe2 functional layer is a P-channel, and the switching current ratio is greater than 10 when the drain-source voltage VDS is 0.5V. 6 When the gate voltage is below -60V, the carrier mobility anisotropy ranges from 16% to 95%.

[0011] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: the Ag electrode forms a Schottky contact with the WSe2 functional layer, and the Schottky barrier heights are 85.4±10.9meV, 74.9±5.37meV, and 99.2±1.67meV, respectively.

[0012] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: the WSe2 functional layer is capable of receiving electrical pulses and laser pulses, wherein the electrical pulses are used to induce excitatory postsynaptic currents and inhibitory postsynaptic currents; The decay time of the postsynaptic current triggered by the laser pulse is greater than 95 seconds; The regulation of the electrical pulses includes polarity regulation, width regulation, and amplitude regulation. The regulation of the electrical pulses is used to achieve the conversion between short-term memory and long-term memory, as well as long-term enhancement and long-term inhibition behaviors. The adjustment of the laser pulse includes intensity adjustment, width adjustment and period adjustment, and synaptic weighted gradient modulation is achieved through the adjustment of the laser pulse.

[0013] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: the parameter range of the electrical pulse adjustment is polarity ±2~±10V, width 0.4~1.6s, and amplitude 2~8V.

[0014] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: the parameters for adjusting the laser pulse are intensity 5.3~7.2mW / cm², width 1~3s, and period 1~5s.

[0015] In a preferred embodiment of the artificial neuron based on WSe2 electrical anisotropy described in this invention: applied to the control of a robotic arm, the anisotropic synaptic current is converted into a voltage signal by a transimpedance amplifier, and the voltage signal is digitized by a microcontroller to control the hand gestures of the robotic arm. The hand gestures of the robotic arm include stretching and bending of the thumb, index finger, and little finger.

[0016] The beneficial effects of this invention are as follows: This invention selects WSe2, which has excellent environmental stability, as the core functional layer, avoiding the defects of traditional anisotropic materials such as black phosphorus and triangular selenium that are sensitive to oxygen, and ensuring long-term reliable operation of the device; at the same time, relying on the intrinsic shielding layer composed of W and Se atom clusters and adsorption layers in WSe2 to induce anisotropic carrier mobility, there is no need to rely on defect engineering to introduce lattice defects or impurity atoms, avoiding the performance fluctuation problem caused by the difficulty in accurately controlling the type and number of defects, so that the device can maintain stable signal response characteristics in different crystal directions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Obviously, the drawings described below only relate to some embodiments of the present invention and are not intended to limit the present invention. Wherein: Figure 1 The diagram shows the functional regions of the human brain, the components of neurons, artificial neurons based on WSe2 electrical anisotropy, angle-resolved diagonal electrode transmission curves, and a schematic diagram of a controlled robotic finger. Figure 2 A simulated diagram of photoelectric dendrites is shown; Figure 3 A schematic diagram of the plasticity of an artificial axon modulated by an electrical signal is shown. Figure 4 A schematic diagram illustrating the plasticity of an artificial axon modulated by light signals is shown. Figure 5 The diagram illustrates the MNIST dataset and neural network structure, handwritten digit recognition accuracy, and light pulse-controlled robotic hand gestures. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0019] The terminology used in this invention is that which is currently widely used in the art in consideration of the function of the invention; however, these terms may vary according to the intent of those skilled in the art, precedent, or new technology in the art. Furthermore, specific terms may be chosen by the applicant, and in such cases, their detailed meanings will be described in the detailed description of the invention. Therefore, the terms used in this specification should not be construed as simple names, but rather based on their meanings and the overall description of the invention.

[0020] This embodiment provides an artificial neuron device based on the electrical anisotropy of WSe2. The core objective is to utilize the excellent stability of WSe2 and the anisotropic carrier mobility induced by the intrinsic shielding layer to construct a multi-terminal neuromorphic device with a simple structure and controllable performance. At the same time, it realizes the dendritic function with photoelectric synergy and the axon and multi-synaptic information transmission capabilities, thereby improving the practicality of synapses.

[0021] Reference Figure 1 ,exist Figure 1Figure A shows a schematic diagram of the functional regions and neuronal components of the human brain; Figure B shows a schematic diagram of the structure of an artificial neuron device based on the electrical anisotropy of WSe2; Figure C shows the transfer curve of the angle-resolved diagonal electrode of WSe2-FETs; the inset shows an optical microscope image of the WSe2 sample and the AFM thickness; Figure D shows a schematic diagram of the controlled finger of a robot operator.

[0022] The artificial neuron device based on the electrical anisotropy of WSe2 includes a substrate; the substrate is the supporting structure of the entire neuron device, and its function is to provide a stable support base for the entire artificial neuron device.

[0023] The WSe2-based anisotropic artificial neuron device includes a WSe2 functional layer formed on the substrate. The WSe2 functional layer is the core structure for realizing electrical anisotropy and signal processing functions. It contains an intrinsic shielding layer composed of W and Se atomic clusters and an adsorption layer. The intrinsic shielding layer can induce anisotropic carrier mobility by modulating the migration characteristics of carriers.

[0024] The artificial neuron device based on the electrical anisotropy of WSe2 includes multiple pairs of electrodes disposed on the WSe2 functional layer; the multiple pairs of electrodes are arranged at preset angular intervals to form a multi-terminal field-effect transistor structure. The multiple pairs of electrodes are arranged according to a preset angular interval rule, and together they constitute the multi-terminal field-effect transistor structure. The multi-terminal field-effect transistor structure is used to simulate the connection morphology of axons and multiple synapses in biological neurons.

[0025] The anisotropic carrier mobility induced by the intrinsic shielding layer of the WSe2 functional layer results in inherent differences in the transmission efficiency of carriers in different directions of the WSe2 functional layer. This direction-dependent carrier transmission characteristic can simulate the core function of biological neuron axons transmitting information to multiple synapses, and realize signal transmission between axons and multiple synapses.

[0026] As an optional embodiment, the substrate is a silicon substrate with an oxide layer, and the WSe2 functional layer is prepared by physical vapor deposition.

[0027] The oxide layer, serving as the dielectric layer between the WSe2 functional layer and the silicon substrate, provides support for the growth of the WSe2 functional layer and enables electrical isolation between the WSe2 functional layer and the substrate, preventing irrelevant electrical signals from interfering with the device, thereby allowing anisotropic carriers to transport stably.

[0028] The thickness of the WSe2 functional layer is 1~3 nm, and the Raman spectroscopy detection of the 2H phase of the WSe2 functional layer shows a characteristic peak at 250 cm⁻¹. -1 250cm -1 and 309cm-1 In this embodiment, the thickness of the WSe2 functional layer is 2nm. Under this thickness condition, it can ensure that the WSe2 functional layer forms a complete layered crystal structure, and at the same time provide a stable environment for the intrinsic shielding layer composed of W and Se atomic clusters and adsorption layers.

[0029] Raman spectroscopy analysis revealed that the WSe2 functional layer exhibited characteristic peaks of the 2H phase, with a peak value at 250 cm⁻¹. -1 The characteristic peak at that location corresponds to the E of 2H-WSe2. 12 G mode, 260cm -1 The characteristic peak at 309 cm⁻¹ corresponds to the Alg mode of 2H-WSe₂. -1 The characteristic peak B2g at the location 1 The presence of the aforementioned characteristic peaks, stemming from interlayer interactions, confirms that the WSe2 functional layer possesses a regular 2H phase crystal structure.

[0030] As an optional embodiment, the multiple pairs of electrodes are three pairs of diagonal Ag electrodes, and the preset angle is a 60° interval between the electrodes; the three pairs of diagonal Ag electrodes form a six-terminal field-effect transistor structure, and the three pairs of diagonal Ag electrodes are named E1, E2 and E3 electrode pairs respectively. Ag is chosen as the electrode material because of its good conductivity, which can form a stable electrical connection with the WSe2 functional layer and ensure stable signal transmission.

[0031] The preset angle is the interval angle between adjacent electrode pairs, specifically 60°. This angle interval design is adapted to the crystal structure of the WSe2 functional layer and the anisotropic carrier migration characteristics induced by the intrinsic shielding layer. It can make full use of the carrier transport differences of WSe2 in different crystal directions to complete the transmission of multi-directional signals. The three pairs of diagonal Ag electrodes, through the above arrangement, together with the WSe2 functional layer on the substrate, form a six-terminal field-effect transistor structure. This structure can simulate the axon and multi-synaptic network morphology of biological neurons, enabling artificial neurons to achieve parallel processing of multiple input and output signals on a single WSe2 functional layer.

[0032] Please refer to Figure 2 , Figure 2 Figure A shows a schematic diagram of the photoelectric dendritic structure of WSe2; Figure B shows EPSC excited by an electrical pulse; Figure C shows modulated EPSC triggered by different electrical pulse widths; Figure D shows EPSC induced by different amplitudes; Figure E shows the learning-forgetting-relearning process simulated by electrical pulses; Figure F shows EPSC generated by light pulse stimulation; Figure G shows the modulation of EPSC by changing the width of the light pulse; Figure H shows the modulation of EPSC by changing the intensity of the light pulse; Figure I shows the learning-forgetting-relearning process simulated by light pulses.

[0033] As an optional embodiment, the WSe2 functional layer is a P-type channel. When the gate voltage is below -60V, the electrical anisotropy of the carrier mobility ranges from 16% to 95%. The P-type channel characteristic is that the drain-source current increases significantly with the negative increase of the gate voltage. This characteristic is related to the crystal structure and carrier transport characteristics of WSe2, providing a basis for the device to achieve stable electrical control.

[0034] After testing, the drain-source voltage was set to 0.5V, which is the commonly used operating value for the device. Under this condition, the switching current ratio of the WSe2 functional layer is greater than 10. 6 The ultra-high switching current ratio can effectively ensure the clarity of signal transmission and anti-interference ability, and avoid invalid signals from interfering with the transmission of axon and multi-synaptic signals.

[0035] When the gate voltage is below -60V, the carrier mobility of the WSe2 functional layer has an electrical anisotropy range of 16% to 95%. This anisotropy range is due to the modulation effect of the intrinsic shielding layer composed of W and Se atom clusters and adsorption layers. The difference in carrier mobility in different directions can be directly converted into different signal transmission efficiencies between each electrode pair.

[0036] As an optional embodiment, the Ag electrode forms a Schottky contact with the WSe2 functional layer, with Schottky barrier heights of 85.4±10.9 meV, 74.9±5.37 meV, and 99.2±1.67 meV, respectively. The Schottky contact is clearly verified by nonlinear output curve testing of the device, conforming to the typical electrical characteristics of a Schottky contact. The Schottky barrier height is obtained through temperature-dependent output curve testing and calculation, and the testing process strictly follows the standard procedure for characterizing the electrical performance of the device to ensure the accuracy of the values. Specifically, the Schottky barrier heights formed by the three pairs of Ag diagonal electrodes E1, E2, and E3 with the WSe2 functional layer are 85.4±10.9meV, 74.9±5.37meV, and 99.2±1.67meV, respectively. These values ​​correspond to the differences in contact characteristics between the three pairs of electrodes and the WSe2 functional layer. According to the electrical transport law IDS∝exp[-qΦB / kBT], the current change induced by the above-mentioned Schottky barrier height differences is small and is not the main source of the device's electrical anisotropy. This further confirms that the anisotropic carrier mobility induced by the intrinsic shielding layer of the WSe2 functional layer is the core of realizing the differentiated transmission of axonal and multi-synaptic signals.

[0037] As an optional embodiment, the WSe2 functional layer is capable of receiving electrical pulses and laser pulses, the electrical pulses being used to induce excitatory postsynaptic currents (EPSC) and inhibitory postsynaptic currents (IPSC). The decay time of the postsynaptic current triggered by the laser pulse is greater than 95 seconds; The regulation of the electrical pulses includes polarity regulation, width regulation, and amplitude regulation. The regulation of the electrical pulses is used to achieve the conversion between short-term memory and long-term memory, as well as long-term enhancement and long-term inhibition behaviors. The adjustment of the laser pulse includes intensity adjustment, width adjustment and period adjustment, and synaptic weighted gradient modulation is achieved through the adjustment of the laser pulse.

[0038] Based on the photoelectric properties of the WSe2 functional layer, it can simultaneously receive two external stimuli: electrical pulses and laser pulses, enabling responses to multiple signals. When the electrical pulse acts on the WSe2 functional layer, the injection and capture of charge carriers are regulated by the voltage pulse applied to the gate: a positive voltage pulse injects electrons into the P-type WSe2 channel and they are captured by the interface defects between WSe2 and the dielectric layer. After the pulse is removed, the electrons are released, leading to an increase in hole concentration, which in turn induces an excitatory postsynaptic current. A negative voltage pulse injects holes into the channel, which recombine with the electrons captured in the defects, inducing an inhibitory postsynaptic current. The induction process of both currents is consistent with the synaptic response characteristics of the biological nervous system.

[0039] When the laser pulse irradiates the WSe2 functional layer, it will generate electron-hole pairs. The increase in carrier concentration expands the channel conductivity. At the same time, a large number of electrons are captured by intrinsic defects and interface defects, which causes the postsynaptic current to increase suddenly. After the laser irradiation is removed, the postsynaptic current decays exponentially due to the recombination of electrons and holes and the slow release of captured electrons. The decay time is greater than 95s, which shows excellent optical memory performance.

[0040] The regulation of the electrical pulse includes polarity regulation, width regulation, and amplitude regulation. Polarity regulation can selectively induce excitatory or inhibitory postsynaptic currents. Width regulation results in a larger amplitude of the excitatory postsynaptic current and a longer retention time after pulse termination, enabling the conversion from short-term memory to long-term memory. Amplitude regulation allows for the gradual enhancement or weakening of synaptic weights through repeated application of electrical pulses of different amplitudes, corresponding to long-term enhancement and long-term inhibition behaviors in biological nervous systems, respectively.

[0041] The adjustment of the laser pulse includes intensity adjustment, width adjustment, and period adjustment: by adjusting the width, the longer the artificial dendrites are exposed to the laser, the greater the amplitude of the excitatory postsynaptic current, the stronger the synaptic connection and the longer the relaxation time, thus realizing the transition from short-term memory to long-term memory; by adjusting the intensity, the increase in the intensity of the laser pulse will correspondingly increase the change in the excitatory postsynaptic current; by adjusting the period, the conversion of multi-level memory retention states can be realized.

[0042] The parameter range for the electrical pulse adjustment was determined through testing and is compatible with the P-type channel characteristics of the WSe2 functional layer and the carrier migration law induced by the intrinsic shielding layer. Specifically, the polarity adjustment range of the electrical pulse is ±2 to ±10V. This range corresponds to the effective control interval of the gate voltage on the hysteresis behavior of the device's transport characteristics. As the polarity voltage increases from ±2V to ±10V, the hysteresis window gradually expands, enabling wide-range adjustment of synaptic weights. The pulse width can be adjusted from 0.4 to 1.6 s, which covers the critical value for the transition from short-term memory to long-term memory. When the pulse width is gradually increased from 0.4 s to 1.6 s, the amplitude of the excitatory postsynaptic current continues to increase and the retention time after the pulse terminates is significantly prolonged. The amplitude of the electrical pulse can be adjusted from 2 to 8V, which matches the stimulation intensity required for synaptic reinforcement in biological nervous systems. By applying positive pulses of different amplitudes such as 2V, 5V, and 8V, the synaptic weight can be gradually increased with repeated stimulation, accurately reproducing long-term enhanced behavior, while negative pulses correspond to inducing long-term inhibition.

[0043] The parameter range for laser pulse modulation is determined based on the photoelectric properties of the WSe2 functional layer, ensuring that the laser pulse can effectively excite electron-hole pairs and modulate synaptic weights.

[0044] The intensity of the laser pulse can be adjusted in the range of 5.3 to 7.2 mW / cm², which covers the effective intensity range of synaptic weight gradient modulation. When the intensity increases from 5.3 mW / cm² to 7.2 mW / cm², the change in excitatory postsynaptic current increases accordingly, thereby enhancing the synaptic connection strength and stably reproducing the transition from short-term memory to long-term memory. The laser pulse width can be adjusted from 1 to 3 seconds, which is compatible with the dynamic characteristics of carrier capture and release. When the pulse width increases from 1 to 3 seconds, the amplitude of the excitatory postsynaptic current increases from 0.21 nA to 0.32 nA. The synaptic connection is strengthened with the extension of illumination time, and the relaxation time is extended simultaneously, thus allowing for a stable transition of the memory state. The laser pulse period can be adjusted from 1 to 5 seconds, which corresponds to the transition requirements of multi-level memory retention states.

[0045] Please refer to Figure 3 , Figure 3Figure A shows the transfer hysteresis characteristic curve of a six-terminal artificial axon; Figure B shows the EPSC of a six-terminal artificial axon induced by the same positive pulse (V=1.5V, W=0.5s); Figure C shows the multi-level conductance state under continuous electrical pulses (V=2V, W=0.5s, Δt=0.5s, N=10); Figure D shows the EPSC excited by a pair of electrical pulses (V=2V, W=0.5s, Δt=1.3s); Figure E shows the dependence of the PPF exponent on the pulse interval Δt; Figure F shows the LTP and LTD behavior of a six-terminal artificial axon (V=±1.5V, W=0.5s, Δt=0.5s) and VDS at 0.5V.

[0046] Where V refers to the amplitude of the voltage pulse, W refers to the width of the voltage pulse, Δt is the time interval between two adjacent voltage pulses, N refers to the number of pulses in the voltage pulse sequence, LTP is long-term enhancement, and LTD is long-term suppression.

[0047] Please refer to Figure 3 A. First, the transfer hysteresis curve of the six-terminal axon structure was measured at a VDS of 0.5V. Figure 3 A). The hysteresis and significant electrical anisotropy of the transmission curves from E1 to E3 imply potential applications in different synaptic behaviors elicited by the same stimulus. Then, the same positive voltage pulse (V=1.5V, W=0.5s) was applied to the postgate to induce EPSC in the artificial six-terminal axon structure. Figure 3 B). Although all channels exhibited excitatory synaptic responses, the variation in synaptic weight (ΔW) differed significantly between E1 and E3, with ΔW values ​​of 1.2 × 10⁻¹⁰ A, 3.0 × 10⁻¹⁰ A, and 1.0 × 10⁻¹⁰ A for E1, E2, and E3, respectively. The different synaptic weights from E1 to E3 stemmed from varying carrier mobility from E1 to E3 (Fig. S11). These different synaptic weights confirmed the fundamental ability to simultaneously transmit information to several subsequent neurons. Furthermore, by applying a positive voltage pulse sequence (V = 2V, W = 0.5s, Δt = 0.5s, N = 10) to achieve conduction modulation, the six-terminal axons exhibited various plasticities from E1 to E3. Figure 3 C) further confirms the different synaptic behaviors.

[0048] The different memory retention capacities from E1 to E3 were evaluated by comparing the PPF exponents. When a pair of consecutive electrical pulses (V=2V, W=0.5s) were applied at 1.3s intervals, the amplitude of the second postsynaptic current was higher than that of the first. Figure 3 D), indicating a successful simulation of memory retention capacity and PPF behavior. A PPF index is introduced to quantify its dependence on time intervals and to assess different memory retention capacities from E1 to E3. Figure 3E). The PPF exponents for E1, E2, and E3 are 77, 84, and 61, respectively, indicating differences in memory retention and information processing capabilities of the six-terminal axonal structure. The different PPF exponents from E1 to E3 also stem from the different carrier mobilities and release of trapped electrons from E1 to E3. This performance is consistent with biological axons with multiple synapses. Furthermore, the PPF exponent decays exponentially with increasing pulse interval. This decay can be fitted by a double exponential function:

[0049] Where A1 and A2 are the fast and slow fading amplitudes, and τ1 and τ2 are the fast and slow decay times in the PPF. We calculated that τ1 and τ2 are 26 ms and 2077 ms, 31 ms and 1109 ms, and 71 ms and 5200 ms at E1, E2, and E3, respectively. The order-of-magnitude difference between τ1 and τ2 from E1 to E3 is compatible with the synaptic timescale.

[0050] Furthermore, WSe2-based axon-multisynaptic neural structures are capable of both LTP and LTD behaviors. Figure 3 F). We applied a sequence of 20 positive voltage pulses (V=1.5V, W=0.5s, Δt=0.5s) to the device, followed by a sequence of 20 negative voltage pulses (V=-1.5V, W=0.5s, Δt=0.5s). We observed that the postsynaptic current gradually increased during the LTP period and gradually decreased during the LTD period, effectively replicating the excitation and inhibition behaviors observed in biological synaptic systems. Notably, the modulation of synaptic weights from E1 to E3 also exhibited a strong dependence on channel orientation, highlighting the ability to simulate the anisotropy of biological axonal networks through electrical signals.

[0051] Please refer to Figure 4 ,exist Figure 4 In the figure, Figure A shows the EPSC of a six-terminal artificial axon induced by the same light pulse (P=6.7mW / cm2, W=1s); Figure B shows the EPSC excited by a pair of light pulses (P=6.7mW / cm2, W=1s, Δt=0.1s); Figure C shows the dependence of the PPF exponent on Δt; Figure D shows the behavior of six-terminal artificial axon EPSC triggered by different light pulse periods; Figure E shows the EPSC triggered by light pulses of different powers (W=1s, Δt=1s), with the inset showing the dependence of the extracted six-terminal artificial axon EPSC on light intensity; Figure F shows the EPSC triggered by light pulses of different pulse widths (P=5.3mW / cm2, Δt=5s), with the inset showing the dependence of the EPSC on the pulse width. All tests were performed at VDS=0.5V.

[0052] Since artificial dendrites of neurons can be triggered by light signals, anisotropic transmission performs quite well in terminal artificial axons stimulated by light pulses. When an initial single light pulse is applied (P=6.7mW / cm2, W=1s), Figure 4 A), the synaptic weights are different from each other, with ΔW of 4.5×10⁻¹⁰ A, 1.8×10⁻⁹ A, and 3.8×10⁻¹⁰ A for E1, E2, and E3, respectively. These different synaptic weights suggest a promising ability to communicate differentially with other neurons. Subsequently, two consecutive light pulses (P=6.7 mW / cm², W=1 s, Δt=0.1 s) and various pulse intervals ( Figure 4 (BC) Verification and calculation of PPF behavior of six-terminal artificial axons. Various PPF indices of the light signal from E1 to E3 are consistent with the information transmission of biological axons in various directions. Furthermore, the period of the light pulse ( Figure 4 D), strength ( Figure 4 E) and width ( Figure 4 F) further demonstrates the multi-level retention and storage states of the six-terminal artificial axon. Increasing the pulse period leads to a plastic transition from LTM to STM, resulting in decreased memory performance, while increasing pulse intensity and width produces an increase in retention and storage states. Therefore, the six-terminal device exhibits anisotropic response to optical signals, which is highly consistent with the differential information transmission observed in biological axon-multisynaptic structures.

[0053] Please refer to Figure 5 , Figure 5 Figure A shows the MNIST dataset and neural network, consisting of 784 input neurons, 300 hidden neurons, and 10 output neurons. Figure B shows the recognition accuracy M of the MNIST simulation for each training iteration. Figure C is a schematic diagram of the hand gesture control of the robotic arm using light pulses.

[0054] The artificial neuron core is used for precise gesture control of the robot's manipulator. It utilizes the anisotropic carrier mobility induced by the intrinsic shielding layer of the WSe2 functional layer, as well as the different synaptic responses of the three pairs of Ag diagonal electrodes E1, E2, and E3, to directly convert the anisotropic electrical characteristics of the device into mechanical control signals.

[0055] The specific control process is as follows: When a light pulse is applied, under a drain-source voltage of 0.5V, the three pairs of Ag diagonal electrodes generate anisotropic synaptic currents with direction dependence. These currents are related to the stability of the Schottky contact and the synaptic plasticity regulated by the photoelectric pulse. Subsequently, the synaptic current is input to a transimpedance amplifier for signal conversion, and the gain of the transimpedance amplifier is set to 10. 10With a voltage of V / A and a bandwidth of 10kHz, the system ensures that weak synaptic currents are accurately amplified into a processable voltage signal. The converted voltage signal is transmitted to a microcontroller (STM32F103C8T6) for digital processing. The digitized signal is then transmitted to a computer via a serial port for monitoring, and simultaneously establishes a control connection with the steering mechanisms corresponding to the thumb, index finger, and little finger of the robot's manipulator.

[0056] The mapping relationship between the three pairs of Ag diagonal electrodes and the fingers of the robotic hand is as follows: electrode E1 corresponds to the thumb, electrode E2 corresponds to the index finger, and electrode E3 corresponds to the little finger. The computer determines the threshold voltage of each steering mechanism by analyzing the waveform characteristics of the output voltage of the three pairs of electrodes. This threshold voltage directly governs the movement state of the fingers: when the light pulse is continuously applied, the synaptic current corresponding to each electrode is stable, and the converted voltage signal reaches the steering mechanism threshold, driving the corresponding finger to perform a stretching action; when the light pulse is removed, the synaptic current gradually decays, the voltage signal is lower than the threshold, and the finger performs a bending action, thereby realizing the combined control of multiple gestures.

[0057] This control method achieves signal differentiation through the inherent hardware anisotropy of the device, eliminating the need for complex software algorithms to analyze multi-channel signals and effectively reducing computational overhead. Compared with traditional multi-channel control schemes, it achieves synchronous feedback between finger movement state and the real-time waveform of anisotropic synaptic current of the WSe2 device, ensuring the accuracy and response speed of gesture control.

[0058] Finally, it should be noted that the methods and devices described in detail above are merely embodiments, and those skilled in the art can modify these embodiments in different ways as long as they do not depart from the scope of the present invention.

Claims

1. An artificial neuron based on the electrical anisotropy of WSe2, characterized in that: include, substrate; The WSe2 functional layer is formed on the substrate; Multiple pairs of electrodes are disposed on the WSe2 functional layer, and the multiple pairs of electrodes are arranged at preset angular intervals to form a multi-terminal field-effect transistor structure. The WSe2 functional layer enables signal transmission between axons and multiple synapses through the anisotropic carrier mobility induced by the intrinsic shielding layer.

2. The artificial neuron based on WSe2 electrical anisotropy according to claim 1, characterized in that: The substrate is a silicon substrate with an oxide layer, and the WSe2 functional layer is prepared by physical vapor deposition.

3. The artificial neuron based on WSe2 electrical anisotropy according to claim 1, characterized in that: The thickness of the WSe2 functional layer is 1~3 nm, and the Raman spectroscopy detection of the 2H phase of the WSe2 functional layer shows a characteristic peak at 250 cm⁻¹. -1 250cm -1 and 309cm -1 .

4. The artificial neuron based on WSe2 electrical anisotropy according to any one of claims 1 to 3, characterized in that: The multiple pairs of electrodes are three pairs of diagonal Ag electrodes, and the preset angle is a 60° interval between the electrodes; the three pairs of diagonal Ag electrodes form a six-terminal field-effect transistor structure.

5. The artificial neuron based on WSe2 electrical anisotropy according to claim 4, characterized in that: The WSe2 functional layer is a P-channel type, and the switching current ratio is greater than 10 when the drain-source voltage VDS is 0.5V. 6 When the gate voltage is below -60V, the carrier mobility anisotropy ranges from 16% to 95%.

6. The artificial neuron based on WSe2 electrical anisotropy according to claim 5, characterized in that: The Ag electrode forms a Schottky contact with the WSe2 functional layer, and the Schottky barrier heights are 85.4±10.9meV, 74.9±5.37meV, and 99.2±1.67meV, respectively.

7. The artificial neuron based on WSe2 electrical anisotropy according to claim 1, characterized in that: The WSe2 functional layer can receive electrical pulses and laser pulses, the electrical pulses being used to induce excitatory and inhibitory postsynaptic currents. The decay time of the postsynaptic current triggered by the laser pulse is greater than 95 seconds; The regulation of the electrical pulses includes polarity regulation, width regulation, and amplitude regulation. The regulation of the electrical pulses enables the conversion between short-term and long-term memory, as well as long-term enhancement and long-term inhibition behaviors. The adjustment of the laser pulse includes intensity adjustment, width adjustment and period adjustment, and synaptic weighted gradient modulation is achieved through the adjustment of the laser pulse.

8. The artificial neuron based on WSe2 electrical anisotropy according to claim 7, characterized in that: The parameters for the electrical pulse adjustment are: polarity ±2 to ±10V, width 0.4 to 1.6s, and amplitude 2 to 8V.

9. The artificial neuron based on WSe2 electrical anisotropy according to claim 7, characterized in that: The parameters for adjusting the laser pulse are: intensity 5.3~7.2mW / cm², pulse width 1~3s, and pulse period 1~5s.

10. The artificial neuron based on WSe2 electrical anisotropy according to any one of claims 1, 6 to 9, characterized in that: It is applied to the control of robotic arms. Anisotropic synaptic currents are converted into voltage signals through transimpedance amplifiers. The voltage signals are then digitized by a microcontroller and used to control the hand gestures of the robotic arm. The hand gestures of the robotic arm include stretching and bending of the thumb, index finger, and little finger.

Images