Application of a transition metal chalcophosphate layered material in a neuromorphic synapse device based on ion migration mechanism

Abstract

The application provides application of a transition metal chalcophosphate layered material in a simulated neural synapse device based on an ion migration mechanism. The simulated neural synapse device comprises, from bottom to top, a supporting substrate, a back gate electrode, a dielectric layer, a source-drain electrode and a packaging layer; the source-drain electrode comprises a source electrode and a drain electrode, and a conductive channel is arranged between the source electrode and the drain electrode, and the conductive channel is the transition metal chalcophosphate layered material. The application realizes the simulated neural synapse electronic device with the transition metal chalcophosphate layered material as the conductive channel by regulating the transport process of intrinsic ions in the material lattice, and effectively solves the problems of existing metal conductive wire resistive memory, such as unstable metal atoms, random changes of high-low resistance state switching voltage and excessive on-state current.

Description

Technical Field

[0001] This invention belongs to the field of layered semiconductor materials and novel non-von Neumann computing technology, specifically relating to the application of a transition metal chalcogenide layered material based on ion migration mechanism in a neural synapse-like device. Background Technology

[0002] Due to the ever-increasing demands for processor performance in modern society and the inherent limitations of silicon chip manufacturing processes, traditional computers have consistently faced the challenge of balancing power consumption and efficiency when performing large-scale data computations. With the advent of the artificial intelligence era, emerging technologies such as big data, cloud computing, and machine learning are placing even higher demands on the performance of existing computing devices. Since John von Neumann first proposed the concept of "stored program" and the binary principle in 1945, systems designed based on this concept and principle are collectively referred to as "von Neumann architecture" electronic computers. However, in electronic computers using traditional complementary metal-insulator-semiconductor (CMOS) integrated circuits as their core components, the CPU's operating speed is far greater than the memory's read / write speed. However, due to the separation of the CPU and memory, the CPU remains idle during data input or output, making it difficult to maximize the utilization of computing resources, thus forming the so-called "von Neumann bottleneck." Consequently, despite the booming development and application of artificial intelligence algorithms such as neural networks and deep learning, they still face the von Neumann bottleneck problem because they rely on the operating environment of traditional computing devices and architectures with CMOS integrated circuits as their core.

[0003] Inspired by the human brain's biological nervous system, hardware neuromorphic computing with highly parallel computing and adaptive learning capabilities has been developed, effectively addressing the computationally intensive problems inherent in the limitations of the von Neumann architecture. More importantly, further integration with existing neural network software systems holds the promise of a true leap forward in current artificial intelligence and the ideal form of future computing devices. Artificial synapses, as a crucial component of neuromorphic computing systems, possess high energy efficiency, low power consumption, and parallel storage and processing capabilities, demonstrating immense research potential in storage systems and in-memory computing, and have become a hot topic and focus of hardware neuromorphic computing research.

[0004] Among the many promising candidates for hardware neuromorphic computing, two-dimensional materials with memristor properties or two-dimensional van der Waals heterojunctions, due to their unique atomic-level thickness, can effectively reduce energy consumption caused by short-channel effects, thus facilitating a better balance between power consumption and operating efficiency. Furthermore, the weaker charge shielding capabilities and rich, unique material properties of two-dimensional materials enable them to better simulate various synaptic plasticities representing the ability to change weights in biological synapses. Therefore, in recent years, work has been reported on using two-dimensional materials and their van der Waals heterojunctions to construct non-volatile memory devices such as resistive random access memory, phase-change memory, ferroelectric tunnel junctions, and floating-gate transistors to simulate dynamic synaptic plasticity and perform neuromorphic computing.

[0005] It is worth noting that among these two-dimensional material synaptic devices, memristors based on the formation and breakage of metal conductive filaments offer a favorable opportunity to mimic advanced neural functions such as associative learning and activity-dependent plasticity due to the similarity between the physical processes of metal ion migration and ion information transmission in real biological synapses. However, memristors forming metal conductive filaments often suffer from practical problems such as creeping current, device oxidation, and highly variable switching voltage and current, which greatly hinder their practical application. Therefore, in order to construct neuromorphic systems that simulate real organisms, it remains urgent to explore new materials and devices with reliable synaptic properties.

[0006] This paper proposes a neural synapse-inspired device based on the lattice ion migration mechanism of transition metal dichalcogenide layered materials. Utilizing the high ionic conductivity of these layered materials, the hopping motion of intrinsic metal cations within the sulfur framework is modulated by an electric field, achieving effective control of ionic conductivity and biomimeticly simulating the membrane potential signal transmission process in real biological synapses. This method of controlling the distribution of intrinsic metal cations in the lattice to regulate the device's electrical properties provides a platform for realistically simulating higher neural functions of biological synapses and exploring novel biological functions related to ion dynamics. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a neural synapse-like electronic device that utilizes the characteristics of interlattice ion hopping migration and controls the transport process of intrinsic ions in the material lattice to realize a conductive channel using a transition metal chalcogenide layered material. This effectively solves the problems of easy oxidation and instability of metal atoms, random variation of high and low resistance switching voltage, and excessive on-state current in existing metal conductive wire memristors.

[0008] To achieve the above objectives, the technical solution adopted by the present invention to solve its technical problem is: to provide an application of a transition metal chalcogenide layered material based on an ion migration mechanism in a neural synapse-like device.

[0009] Furthermore, the general chemical formula for transition metal chalcogenide layered materials is M 4+ [N2X6] 4- [M] 2+ ]2[N2X6] 4- Or M 1+ M 3+ [N2X6] 4- .

[0010] Furthermore, the transition metal chalcogenide layered materials are SnP2S6, SnP2Se6, Sn2P2S6, Sn2P2Se6, CuInP2S6, CuInP2Se6, CuBiP2S6, CuBiP2Se6, CuCrP2S6, CuCrP2Se6, AgBiP2Se6, or AgBiP2S6.

[0011] Furthermore, M can be Cu, Ag, In, Bi, Cr, or Sn.

[0012] Furthermore, X can be S, Se, or Te.

[0013] Furthermore, N can be P, As, or Sb.

[0014] Furthermore, the neural synapse-like device includes a support substrate, a back gate electrode, a dielectric layer, source and drain electrodes, and an encapsulation layer stacked sequentially from bottom to top; the source and drain electrodes include a source electrode and a drain electrode, and a conductive channel is provided between the source electrode and the drain electrode, the conductive channel being a transition metal thiophosphate layered material.

[0015] Furthermore, the source electrode and drain electrode are connected to both ends of the conductive channel to form a Schottky contact or an ohmic contact containing a clean van der Waals interface.

[0016] Furthermore, the conductive channel can be single-layered, multi-layered, or bulk.

[0017] Furthermore, the back gate electrode can be a metal electrode, a heavily doped semiconductor, a two-dimensional metal, or a half-metal material.

[0018] Furthermore, the dielectric layer is SiO2, Al2O3, or h-BN; the source and drain electrodes are metal electrodes, two-dimensional metal electrodes, or half-metal material electrodes; and the encapsulation layer is h-BN or Al2O3.

[0019] The present invention relates to the application of transition metal chalcogenide layered materials in neural synapse-like devices based on ion migration mechanisms. This refers to using an external electric field to regulate the jumping motion of metal cations between crystal lattices in transition metal chalcogenide layered materials, simulating the transmembrane behavior of ions in biological synapses, and realizing biomimetic simulation of various synaptic plasticities and higher neural functions.

[0020] This invention employs layered transition metal thiophosphates with room-temperature ferroelectricity as the conductive channel material. By combining this with the intralattice metal ion hopping migration process, resistive memory behavior related to the source-drain voltage scanning range is achieved, thus better simulating membrane potential changes in biological synapses and realizing the simulation of higher-order synaptic characteristics. Specifically, in the nervous system, a common type of synapse is the chemical synapse, through which preneurons connect to postneurons. When the action potential reaches the presynaptic terminal, the voltage-gated Caa... 2+ The opening of the channel leads to the release of neurotransmitters, which then integrate into receptors at the postsynaptic terminal, thereby achieving the conversion from electrical signals to biochemical signals, accompanied by the regulation of synaptic strength. In our proposed novel synaptic simulation device, the working mechanism can be described as follows: by applying an in-plane (out-of-plane) electric field, metal cations jump from their equilibrium positions in the crystal lattice and then migrate along the direction of the electric field within the lattice framework composed of anions. The directional migration of metal cations will result in high ionic conductivity and gradually accumulate near the cathode.

[0021] Furthermore, by utilizing the non-uniform distribution of metal ions and their vacancies to form a structure similar to a pn diode, various logic operation functions can be realized. By taking advantage of the poor conductivity of the device channel before the migration of metal cations in layered transition metal chalcogenide, and the significant increase in ionic conductivity in the channel after ion migration, the operating current of the source and drain of existing conductive wire-type synaptic simulation devices can be greatly reduced, thereby realizing a low-power synaptic electronic device based on the ion migration mechanism.

[0022] In summary, the present invention has the following advantages:

[0023] 1. The synaptic electronic device based on two-dimensional ferroelectric semiconductor of the present invention achieves input control of the ion conductivity in the channel, i.e. the synaptic weight of the conductive channel, by regulating the lattice jumping migration of metal cations in the layered transition metal chalcogenide conductive channel through source and drain electrodes.

[0024] 2. This invention uses layered transition metal chalcogenide phosphate material as the conductive channel, and utilizes the non-uniform distribution of metal cations and vacancies within its lattice to form a structure similar to a pn diode, thereby realizing a variety of logic operation functions;

[0025] 3. This invention utilizes the poor conductivity of the device channel before the migration of metal cations in layered transition metal thiophosphates, and the significant increase in ionic conductivity in the channel after ion migration, to greatly reduce the operating current of the source and drain electrodes of existing conductive wire-type synaptic simulation devices. It realizes a low-power synaptic electronic device while utilizing the ion migration mechanism.

[0026] 4. Based on the regulation of the jumping migration of metal cations in layered transition metal thiophosphates between in-plane (out-of-plane) lattices, this invention can be further extended to multi-port heterogeneous synaptic electronic devices with multiple electrode inputs and outputs. This provides a platform for exploring advanced synaptic characteristics such as competition, cooperation, and association in biological synapses, and also provides a basic circuit unit with diverse structural designs and easy integration for hardware neuromorphic computing. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of a neural synapse-inspired device.

[0028] Figure 2 This is a graph showing the memristor characteristics of a neural synapse-like device at source-drain input.

[0029] Figure 3 The pn junction characteristic curves of the simulated neural synapse device after excitation by different source-drain voltages;

[0030] Figure 4 To simulate the higher-level characteristic of biological synapses—associative ability—into a neural synapse-inspired device;

[0031] Figure 5 To simulate the superplasticity in biological synapses using neural synapse-inspired devices;

[0032] Figure 6 This is a biomimetic simulation of the sensory nerve response to different external environmental stimuli by a neural synapse device;

[0033] The components are: 1. Supporting substrate; 2. Back gate electrode; 3. Dielectric layer; 4. Conductive channel; 5. Source electrode; 6. Drain electrode; 7. Encapsulation layer. Detailed Implementation

[0034] Example 1

[0035] The application of transition metal chalcogenide layered materials in synaptic-inspired devices based on ion migration mechanisms, taking CuInP2S6 as an example, is illustrated by the following specific steps in its fabrication method:

[0036] 1) Fabrication of metal Au electrodes on SiO2 substrates: First, a strip-shaped mask pattern with a spacing of 5 μm and a width of 10 μm is photolithographically etched on a heavily doped Si / SiO2 substrate. Then, a thermal evaporation deposition device is used to... A 50 nm thick Au film was deposited on the substrate at a certain rate, and then the substrate was immersed in acetone solution for 30 min before lift-off to obtain a patterned Au electrode.

[0037] 2) Preparation of PVA dry transfer film: First, a 10wt% PVA aqueous solution was prepared. Then, 5mL of the solution was dropped onto a cleaned SiO2 substrate and then placed on a heating plate at 50℃ for 10min to obtain a PVA film. Then, a 3mm×3mm PVA film was cut with a blade and placed on a PDMS substrate of the same size at one end of a transparent glass slide to obtain a PVA dry transfer film.

[0038] 3) PVA dry transfer of Au electrode: With the aid of a microscope and a three-dimensional displacement platform, the sample stage is first heated to 50°C, and then the corner of the PVA dry transfer film is slowly and evenly attached to the Au electrode on the SiO2 substrate. After the heating stage cools to room temperature, the glass slide is slowly lifted to realize the dry transfer of the metal electrode to the PVA film.

[0039] 4) Fabrication of metal source and drain electrodes: First, a few layers of sample material were obtained from CuInP2S6 crystal material using mechanical exfoliation and placed on a SiO2 target substrate containing heavily doped Si. Then, the SiO2 substrate was heated to 55°C using a sample stage. Under the assistance of a microscope, the metal electrode on the PVA transfer film and the CuInP2S6 sample were aligned and then slowly and uniformly pressed down. After the PVA film was in complete contact with the SiO2 substrate, it was waited for 2 minutes and then the glass slide was slowly and uniformly lifted. Since the metal electrode fell onto the SiO2 substrate along with the PVA film, a device sample with a PVA film on its surface was finally obtained.

[0040] 5) Preparation of Al2O3 encapsulation layer: First, place the sample in deionized water and heat it to 50°C for half an hour, then remove it and blow it dry with nitrogen gas; then, using an atomic layer deposition device, using Al(CH3)3 and H2O as precursors, deposit a 10nm thick Al2O3 encapsulation protective film on the CuInP2S6 surface at a deposition temperature of 175°C; finally, use gold wires to lead out the source / drain electrodes and back gate electrodes to complete the preparation.

[0041] Experimental Example

[0042] The relationship between voltage and current of the CuInP2S6 synaptic device prepared in Example 1 at the source and drain electrodes was measured to obtain the memristor characteristic curve, as shown in the figure. Figure 2 As shown.

[0043] Depend on Figure 2 It is known that by using CuInP2S6, a two-dimensional ferroelectric semiconductor material of layered transition metal chalcogenide as a conductive channel, and combining it with the in-plane Cu ion migration process, the resistive memory behavior related to the source-drain voltage scanning range is realized, namely the memristor characteristic.

[0044] The voltage and current relationship of the CuInP2S6 synaptic device prepared in Example 1 under different source-drain voltage excitations within a small voltage range was measured, as follows: Figure 3 As shown. Figure 3 From top to bottom on the far right are After 15Vincentive, After 10V incentive, After 8V incentive, After 6V incentive, and Pristine.

[0045] Depend on Figure 3 It is known that the directional migration of Cu ions under the influence of an electric field causes them to accumulate near the cathode, resulting in more vacancies in the region near the anode. Ultimately, the uneven distribution of Cu ions and their vacancies leads to the formation of a pn diode-like structure within the channel. By drawing an analogy with traditional diode logic circuits, it is hoped that a rich variety of logic operation forms can be developed.

[0046] The CuInP2S6 synaptic device prepared in Example 1 was tested through a series of processes such as... Figure 4 The excitation pulses shown simulate the associative learning function of synapses. Classical conditioning, one of the simplest forms of associative learning, also known as Pavlov's dog experiment, plays a crucial role in individual adaptation and brain learning. In Pavlov's dog experiment, food and a bell are referred to as the unconditioned stimulus (US) and conditioned stimulus (CS), respectively. Salivation caused by the US is called the unconditioned response (UR). Before training, the dog might salivate at the sight of food but would not exhibit a physiological response to the bell (CS). During training, the dog is stimulated by repeatedly ringing the bell and consuming food. Subsequently, an association is established between the US and CS, and the dog's salivation can be triggered solely by the CS. Using time-dependent LTP, Pavlovian conditioning can be simulated based on our CIPS device. Figure 4 As shown, pulses with amplitudes of 10V and 0V are defined as US and CS signals, respectively, and the corresponding response currents are recorded as unconditional response (UR) and conditional response (CR). Simultaneously, the 1pA output current threshold is set as the standard for valid response.

[0047] It is easy to see that the current generated by the device's response to a 0V pulse (CS) does not exceed the 1pA current threshold. However, when a 10V pulse (US) is applied to the device, it induces an effective postsynaptic current (PSC). Repeatedly applying 10V and 0V pulses to the presynaptic terminal as a "training mode" causes the PSC to reach above the 1pA threshold before decaying below it. Applying four combinations of 10V and 0V pulses to the presynaptic terminal as "retraining" induces an effective PSC more quickly, indicating that the "memory characteristic" has been successfully simulated. Furthermore, applying a 0V pulse (CS) again subsequently still induces a high PSC above 1pA, indicating that the "associative characteristic" has also been successfully simulated.

[0048] The CuInP2S6 synaptic device prepared in Example 1, when excited by a small voltage pulse (4V, 1s) that does not cause a direct change in device current, will generate an excitatory postsynaptic current (EPSC) much higher than that induced by conventional plasticity when excited by an upcoming large voltage pulse (10V, 1s). Figure 5 As shown. This is very similar to chemigenesis in biology. In biological synapses, the strength is not a constant value, but depends on the past activity of the synapse at each moment; this higher-order form of synaptic plasticity is called chemigenesis. In our device, we simulate activity-dependent synaptic plasticity by modulating Cu ion migration. Specifically, when a low-voltage pulse (4V) is applied, although it has no effect on the overall conductivity of the device, some Cu ions have already migrated toward the cathode or away from their equilibrium position, and these propelled Cu ions fail to return to their initial state within the next 4 seconds. Since more copper ions have already migrated (or are more likely to jump out of the lattice), a high-voltage pulse (10V) applied immediately afterward can cause a higher EPSC.

[0049] The CuInP2S6 synaptic device prepared in Example 1, after being subjected to positive spike pulses (10V, 1s) and negative spike pulses (-2V, 1s) designated as heat (enhancing) and cold (depressing) signals, respectively, can achieve a sensory simulation similar to that of a biological neural nociceptor. Figure 6 As shown. Specifically, the concentration gradient after applying a negative spike (-2V) tends to cause ion back-diffusion, making it easier for more Cu ions in a non-equilibrium state to migrate induced by a positive spike pulse (10V), resulting in a higher PSC. This is like a person going from a cold environment to a hot environment and feeling "very hot"; on the other hand, if a person goes from a hot environment to a cold environment, they will feel "very cold".

[0050] Although specific embodiments of the present invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. Various modifications and variations that can be made by those skilled in the art without inventive effort within the scope described in the claims are still within the scope of protection of this patent.

Claims

1. The application of a transition metal chalcogenide layered material based on ion migration mechanism in neural synapse-inspired devices; characterized in that, The general chemical formula of the transition metal chalcogenide layered material is M. 4+ [N2X6] 4- 、 [M 2+ ]2[N2X6] 4- Or M 1+ M 3+ [N2X6] 4- M is Cu, Ag, In, Bi, Cr, or Sn; X is S, Se, or Te; N is P, As, or Sb; the device utilizes an external electric field to regulate the jumping motion of metal cations between crystal lattices, biomimeticly simulating the transmembrane behavior of ions in biological synapses, thus achieving biomimetic simulation of synaptic plasticity; the biomimetic neural synapse device includes a supporting substrate, a back gate electrode, a dielectric layer, source / drain electrodes, and an encapsulation layer stacked sequentially from bottom to top; the source / drain electrodes include a source electrode and a drain electrode, and a conductive channel is provided between the source electrode and the drain electrode, the conductive channel being a transition metal chalcogenide layered material.

2. The application of the transition metal chalcogenide layered material as described in claim 1 in neural synapse-inspired devices based on ion migration mechanisms, characterized in that... The source electrode and drain electrode are respectively connected to both ends of the conductive channel to form a Schottky contact or an ohmic contact containing a clean van der Waals interface.

3. The application of the transition metal chalcogenide layered material as described in claim 1 in neural synapse-inspired devices based on ion migration mechanisms, characterized in that... The conductive channel can be single-layered, multi-layered, or block-shaped.

4. The application of the transition metal chalcogenide layered material as described in claim 1 in neural synapse-inspired devices based on ion migration mechanisms, characterized in that... The back gate electrode is a metal electrode, a heavily doped semiconductor, a two-dimensional metal, or a half-metal material.

5. The application of the transition metal chalcogenide layered material as described in claim 1 in neural synapse-inspired devices based on ion migration mechanisms, characterized in that... The dielectric layer is SiO2, Al2O3, or h-BN; the source and drain electrodes are metal electrodes, two-dimensional metal electrodes, or half-metal material electrodes; the encapsulation layer is h-BN or Al2O3.

Images