A neuron circuit

Abstract

The application discloses a neuron circuit, comprising: a trigger unit and an oscillation unit; the oscillation unit comprises a first memristor and a second memristor; one end of the first memristor is connected with one end of the second memristor, serving as an output end of the neuron circuit; the other end of the first memristor is connected with an output end of the trigger unit, and the other end of the second memristor is grounded; wherein the first memristor is used for adjusting a threshold value of the neuron circuit; when an input signal input into the trigger unit is greater than or equal to a first threshold value, the neuron circuit starts oscillation; when the input signal input into the trigger unit is less than the first threshold value, the neuron circuit stops oscillation; after continuous oscillation, the resistance value of the first memristor is increased, so that the first threshold value is increased; after the oscillation stops, the resistance value of the first memristor is gradually reduced, and the first threshold value is gradually fallen back, so that a dynamic threshold neuron is realized.

Description

Technical Field

[0001] This invention relates to the field of semiconductor devices, and more particularly to a neuron circuit. Background Technology

[0002] Artificial neurons are components that convert external analog signals into system pulse signals, and they have important applications in building spiking neural network systems. Current neurons generally have a fixed threshold, and in unsupervised learning and other neural network applications or artificial bionic neural systems, problems such as 'overactivation' or 'dead neurons' exist, leading to decreased neural network performance or failure of the artificial neural system. Summary of the Invention

[0003] This application provides a neuron circuit that solves the technical problem of 'overactivation' or 'dead neurons' caused by the fixed threshold of existing neuron circuits, thereby improving the stability of the neuron circuit.

[0004] On the one hand, this application provides the following technical solution through one embodiment:

[0005] A neuron circuit, comprising:

[0006] Triggering unit and oscillation unit;

[0007] The oscillation unit includes a first memristor and a second memristor; one end of the first memristor is connected to one end of the second memristor to serve as the output terminal of the neuron circuit; the other end of the first memristor is connected to the output terminal of the trigger unit, and the other end of the second memristor is grounded.

[0008] The first memristor is used to adjust the threshold of the neuron circuit. When the input signal to the trigger unit is greater than or equal to the first threshold, the neuron circuit starts to oscillate. When the input signal to the trigger unit is less than the first threshold, the neuron circuit stops oscillating. After continuous oscillation, the resistance of the first memristor increases to raise the first threshold. When the oscillation stops, the resistance of the first memristor gradually decreases, and the first threshold gradually falls back to achieve a dynamic threshold neuron.

[0009] Optionally, the triggering unit includes:

[0010] First resistor and first capacitor;

[0011] One end of the first resistor is connected to one end of the first capacitor to serve as the output terminal of the trigger unit, the other end of the first capacitor is grounded, and the other end of the first resistor serves as the input terminal of the trigger unit.

[0012] Optional, including:

[0013] The first memristor is a non-volatile memristor;

[0014] The second memristor is a volatile threshold-switching memristor.

[0015] Optionally, the memristor includes:

[0016] Upper electrode, intermediate layer and lower electrode.

[0017] Optional, including:

[0018] The first memristor intermediate layer material is a non-volatile resistive material with decay characteristics;

[0019] The material of the second memristor intermediate layer is a material with volatile threshold switching characteristics.

[0020] Optional, including:

[0021] The non-volatile resistive material with degradation characteristics includes any one or more of the following: Ta, Ag, W, Ru, and Cu;

[0022] The material having resistive switching properties and degradation properties includes any one or more of the following: WO x HfO2, TiO2, Al2O3, TaO x PCMO.

[0023] Optionally, the material of the lower electrode of the memristor is an inert conductive material.

[0024] Optionally, the inert conductive material includes any one or more of the following: TiN, Poly-Si, Pd, Pt, W, and Au.

[0025] Optionally, the material of the second memristor intermediate layer is a material with volatile threshold switching characteristics.

[0026] Optionally, the material having volatile threshold switching characteristics includes any one or more of the following: NbO2, VO2, SiTe, SiO2:Ag, a-Si:Cu, a-Si:Ag.

[0027] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0028] This invention discloses a neuron circuit, comprising a trigger unit and an oscillation unit. The oscillation unit includes a first memristor and a second memristor. One end of the first memristor is connected to one end of the second memristor to serve as the output terminal of the neuron circuit. The other end of the first memristor is connected to the output terminal of the trigger unit, and the other end of the second memristor is grounded. The first memristor is used to adjust the threshold of the neuron circuit. When the input signal to the trigger unit is greater than or equal to the first threshold, the neuron circuit begins to oscillate; when the input signal to the trigger unit is less than the first threshold, the neuron circuit stops oscillating. After continuous oscillation, the resistance of the first memristor increases, thereby increasing the first threshold. When the oscillation stops, the resistance of the first memristor gradually decreases, and the first threshold gradually decreases, thus achieving a dynamic threshold neuron. Therefore, this invention solves the technical problem of fixed thresholds in existing neuron circuits, which leads to overactivation or dead neurons, and improves the stability of the neuron circuit. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a structural diagram of the neuron circuit in the embodiments of this application;

[0031] Figure 2 This is a pulse curve diagram of the neuron circuit under constant voltage in the embodiments of this application;

[0032] Figure 3 This is a threshold characteristic diagram of the neuron circuit in the embodiments of this application;

[0033] Figure 4 This is a dynamic threshold characteristic diagram of the neuron circuit in the embodiments of this application;

[0034] Figure 5 This is a diagram showing the response characteristics of a non-volatile memristor under pulses in an embodiment of this application.

[0035] Figure 6 This is a degradation characteristic diagram of a non-volatile memristor in an embodiment of this application;

[0036] Figure 7 This is an IV characteristic diagram of the volatile threshold switching memristor in the embodiments of this application;

[0037] Figure 8This is a schematic diagram of the fabrication method of the memristor in the embodiments of this application. Figure 1 ;

[0038] Figure 9 This is a schematic diagram of the fabrication method of the memristor in the embodiments of this application. Figure 2 ;

[0039] Figure 10 This is a schematic diagram of the fabrication method of the memristor in the embodiments of this application. Figure 3 ;

[0040] Figure 11 This is a schematic diagram of the fabrication method of the memristor in the embodiments of this application. Figure 4 ;

[0041] Figure 12 This is a schematic diagram of the fabrication method of the memristor in the embodiments of this application. Figure 5 . Detailed Implementation

[0042] This application provides a neuron circuit that solves the technical problem of 'overactivation' or 'dead neurons' caused by the fixed threshold of existing neuron circuits, thereby improving the stability of the neuron circuit.

[0043] The technical solution of this application embodiment is to solve the above-mentioned technical problems, and the general idea is as follows:

[0044] A neuron circuit, comprising:

[0045] Triggering unit and oscillation unit;

[0046] The oscillation unit includes a first memristor and a second memristor; one end of the first memristor is connected to one end of the second memristor to serve as the output terminal of the neuron circuit; the other end of the first memristor is connected to the output terminal of the trigger unit, and the other end of the second memristor is grounded.

[0047] The first memristor is used to adjust the threshold of the neuron circuit. When the input signal to the trigger unit is greater than or equal to the first threshold, the neuron circuit starts to oscillate. When the input signal to the trigger unit is less than the first threshold, the neuron circuit stops oscillating. After continuous oscillation, the resistance of the first memristor increases to raise the first threshold. When the oscillation stops, the resistance of the first memristor gradually decreases, and the first threshold gradually falls back to achieve a dynamic threshold neuron.

[0048] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0049] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0050] Example 1

[0051] like Figure 1 As shown, a neuron circuit is provided, including:

[0052] Trigger unit 100 and oscillation unit 200;

[0053] The oscillation unit 200 includes a first memristor 210 and a second memristor 220; one end of the first memristor 210 is connected to one end of the second memristor 220 to serve as the output terminal of the neuron circuit; the other end of the first memristor 210 is connected to the output terminal of the trigger unit 100, and the other end of the second memristor 220 is grounded.

[0054] In practical implementation, the neuron circuit of this application has the following characteristics, such as... Figure 2 , Figure 3 and Figure 4 As shown. Figure 2 This is the pulse curve of the neuron under a constant voltage when the resistance of the first memristor 210 remains unchanged. Figure 3 To determine the threshold characteristics of the neuron when the resistance of the first memristor 210 remains constant. Figure 4 This refers to the dynamic threshold characteristics of the neuron when the resistance of the first memristor 210 changes.

[0055] Furthermore, the first memristor 210 can be used to adjust the threshold of the neuron circuit; when the input signal of the input trigger unit 100 is greater than or equal to the first threshold, the neuron circuit starts to oscillate, and when the input signal of the input trigger unit 100 is less than the first threshold, the neuron circuit stops oscillating; after continuous oscillation, the resistance value of the first memristor 210 increases, so as to increase the first threshold; when the oscillation stops, the resistance value of the first memristor 210 gradually decreases, and the first threshold gradually falls back, so as to realize the dynamic threshold neuron.

[0056] Optionally, the trigger unit 100 includes a first resistor 110 and a first capacitor 120. One end of the first resistor 110 and one end of the first capacitor 120 are connected to serve as the output terminal of the trigger unit 100, the other end of the first capacitor 120 is grounded, and the other end of the first resistor 110 serves as the input terminal of the trigger unit 100.

[0057] In the specific implementation, the first memristor 210 is a non-volatile memristor, and the second memristor 220 is a volatile threshold-switching memristor. When there is an input signal at the Input terminal, the circuit charges the first capacitor 120 through the first resistor 110. When the voltage across the second memristor 220 exceeds its threshold voltage, the second memristor 220 switches to a low-resistance state, and the capacitor discharges through it, generating a voltage pulse across the first memristor 210, which adjusts the resistance value of the first memristor 210. When the voltage across the first capacitor 120 drops to the holding voltage of the second memristor 220, the second memristor 220 switches to a high-resistance state, and the capacitor is recharged through the first resistor 110. This process repeats, generating an oscillating output pulse signal at the Output terminal. Meanwhile, continuous voltage pulses are generated across the first memristor 210, gradually adjusting the resistance of the first memristor 210, thus increasing the threshold of the neuron system. When the input stops, or the input is insufficient to turn on the second memristor 220, there are no voltage pulses across the first memristor 210. Due to the decay characteristics of the first memristor 210, the resistance of the first memristor 210 gradually decreases, and the threshold of the neuron system gradually falls back.

[0058] As an optional embodiment, the basic structure of both the first memristor 210 and the second memristor 220 includes an upper electrode, an intermediate layer, and a lower electrode. Different types of memristors can be formed by using different materials to fabricate the upper electrode, intermediate layer, and lower electrode.

[0059] For example, if the first memristor 210 is a non-volatile memristor, then the material of the intermediate layer of the first memristor 210 is a non-volatile resistive switching material with degradation characteristics; the material of the intermediate layer of the second memristor 220 is a material with resistive switching characteristics and degradation characteristics, to form a resistive switching functional layer; the material of the lower electrode of the first memristor 210 is an inert conductive material. The non-volatile resistive switching material with degradation characteristics includes any one or more of the following: Ta, Ag, W, Ru, and Cu; the material with resistive switching characteristics and degradation characteristics includes any one or more of the following: WO4. x HfO2, TiO2, Al2O3, TaO x PCMO, inert conductive materials include any one or more of the following: TiN, Poly-Si, Pd, Pt, W and Au.

[0060] In practical implementation, when voltage pulses (positive or negative pulses) are continuously applied to the first memristor 210, the device's resistance gradually increases, while the current under the read voltage gradually decreases. Furthermore, after a relatively long hold time, the device's resistance exhibits degradation characteristics. The pulse response characteristics and long-term degradation characteristics of the first memristor 210 are as follows: Figure 5 and Figure 6 As shown.

[0061] For example, if the second memristor 220 is a volatile threshold-switching memristor, then the material of the upper electrode of the second memristor 220 is not limited and can be any conductive material; the material of the middle layer of the second memristor 220 is a material with volatile threshold-switching characteristics; and the material of the lower electrode of the second memristor 220 is an inert conductive material. The materials with volatile threshold-switching characteristics include any one or more of the following: NbO2, VO2, SiTe, SiO2:Ag, a-Si:Cu, a-Si:Ag; and the inert conductive materials include any one or more of the following: TiN, Poly-Si, Pd, Pt, W, and Au.

[0062] In the specific implementation process, such as Figure 7 As shown, when the voltage applied to the second memristor 220 exceeds a certain voltage value (V) TH+ or V TH- When the voltage is less than a certain value (V), the second memristor 220 changes from a high-resistance state to a low-resistance state. During voltage flyback, when the voltage is less than a certain value (V)... HOLD+ or V HOLD- The low impedance state is insufficient to maintain the low impedance state, and the second memristor 220 returns to the high impedance state from the low impedance state.

[0063] In addition, this application also provides methods for fabricating the above two memristors, including:

[0064] First, such as Figure 8 As shown, a substrate 1 is provided, which is a silicon substrate. Of course, substrate 1 can be any suitable semiconductor substrate material, and this embodiment is not limited thereto. For example, substrate 1 can be germanium, silicon germanide, SOI (semiconductor-on-insulator), silicon carbide, gallium arsenide, or any group III / V compound semiconductor, etc.

[0065] Next, as Figure 9 As shown, a SiO2 oxide layer 2 is formed on substrate 1 by thermal oxidation. The thickness of oxide layer 2 can be 100nm to 300nm, and of course, the thickness of oxide layer can be reduced or increased according to actual process conditions.

[0066] Then, as Figure 10 As shown, a lower electrode 3 is deposited on silicon oxide, and the thickness of the lower electrode 3 is 10 nm to 100 nm. The material of the lower electrode 3 can be an inert conductive material such as TiN, Poly-Si, Pd, Pt, W, or Au.

[0067] Next, as Figure 11As shown, a functional layer 4 is deposited on the lower electrode 3, and the thickness of the functional layer 4 is 5 nm to 50 nm. When the memristor is a non-volatile memristor, the material of the functional layer 4 is a material with resistive switching characteristics and degradation characteristics, including any one or more of the following: WO x HfO2, TiO2, Al2O3, TaO x PCMO; When the memristor is a volatile threshold-switching memristor, the material of functional layer 4 is a material with volatile threshold-switching characteristics, including any one or more of the following: NbO2, VO2, SiTe, SiO2:Ag, a-Si:Cu, a-Si:Ag.

[0068] Finally, as Figure 12 As shown, an upper electrode 5 is deposited on the functional layer 4. When the memristor is a non-volatile memristor, the material of the upper electrode of the memristor is a non-volatile resistive material with decay characteristics, including any one or more of the following: Ta, Ag, W, Ru and Cu.

[0069] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0070] This invention discloses a neuron circuit, comprising a trigger unit and an oscillation unit. The oscillation unit includes a first memristor and a second memristor. One end of the first memristor is connected to one end of the second memristor to serve as the output terminal of the neuron circuit. The other end of the first memristor is connected to the output terminal of the trigger unit, and the other end of the second memristor is grounded. The first memristor is used to adjust the threshold of the neuron circuit. When the input signal to the trigger unit is greater than or equal to the first threshold, the neuron circuit begins to oscillate; when the input signal to the trigger unit is less than the first threshold, the neuron circuit stops oscillating. After continuous oscillation, the resistance of the first memristor increases, thereby increasing the first threshold. When the oscillation stops, the resistance of the first memristor gradually decreases, and the first threshold gradually decreases, thus achieving a dynamic threshold neuron. Therefore, this invention solves the technical problem of fixed thresholds in existing neuron circuits, which leads to overactivation or dead neurons, and improves the stability of the neuron circuit.

[0071] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0072] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A neuron circuit, characterized in that, include: Triggering unit and oscillation unit; The oscillation unit includes a first memristor and a second memristor; one end of the first memristor is connected to one end of the second memristor to serve as the output terminal of the neuron circuit; the other end of the first memristor is connected to the output terminal of the trigger unit, and the other end of the second memristor is grounded. The first memristor is used to adjust the threshold of the neuron circuit. When the input signal to the trigger unit is greater than or equal to the first threshold, the neuron circuit starts to oscillate. When the input signal to the trigger unit is less than the first threshold, the neuron circuit stops oscillating. After continuous oscillation, the resistance of the first memristor increases to raise the first threshold. When the oscillation stops, the resistance of the first memristor gradually decreases, and the first threshold gradually falls back to achieve a dynamic threshold neuron. The memristor includes: an upper electrode, an intermediate layer, and a lower electrode; The first memristor is a non-volatile memristor, and the material of the intermediate layer of the first memristor is a non-volatile resistive switching material with decay characteristics; the second memristor is a volatile threshold switching memristor, and the material of the intermediate layer of the second memristor is a material with volatile threshold switching characteristics.

2. The neuron circuit as described in claim 1, characterized in that, The triggering unit includes: First resistor and first capacitor; One end of the first resistor is connected to one end of the first capacitor to serve as the output terminal of the trigger unit, the other end of the first capacitor is grounded, and the other end of the first resistor serves as the input terminal of the trigger unit.

3. The neuron circuit as described in claim 1, characterized in that, include: The non-volatile resistive material with degradation characteristics includes one or more of the following: Ta, Ag, W, Ru, and Cu.

4. The neuron circuit as described in claim 1, characterized in that, The material of the lower electrode of the memristor is an inert conductive material.

5. The neuron circuit as described in claim 4, characterized in that, The inert conductive material includes any one or more of the following: TiN, Poly-Si, Pd, Pt, W, and Au.

6. The neuron circuit as described in claim 1, characterized in that, The material having volatile threshold transition characteristics includes any one or more of the following: NbO2, VO2, SiTe, SiO2:Ag, a-Si:Cu, a-Si:Ag.

Images