Memristor based on low thermal conductivity intercalation and multi-mode EGC detection system

Abstract

The invention discloses a memristor based on a low-thermal-conductivity insertion layer and a multi-mode EGC detection system, and belongs to the field of integrated circuit design. Comprising a threshold switch layer located in the middle, the upper end and the lower end of the threshold switch layer are each provided with an insertion layer, and the sides, back to the threshold switch layer, of the two insertion layers are each provided with an electrode; the material of the threshold switch layer is NbOx, TaOx or VOx; the material of the intercalation layer is CdTe2; the electrode is an inert electrode. The thermal characteristics of a Pole-Frenkel mechanism are utilized, and the low-thermal-conductivity and high-conductivity intercalation layer is used for limiting heat loss, so that the threshold voltage is reduced, and the stability of the threshold switch is improved; meanwhile, the oscillation frequency of the memristor can be regulated and controlled by changing the insertion layer, so that the memristor has relatively high oscillation frequency under relatively low working voltage; the temperature and thickness characteristics of the memristor are applied to biological signal processing, and multi-mode ECG arrhythmia detection is developed; and a new strategy is provided for developing low-power-consumption electrocardiogram detection equipment.

Description

Technical Field

[0001] This invention belongs to the field of integrated circuit design, specifically relating to memristors based on low thermal conductivity intercalation and a multi-mode EGC detection system. Background Technology

[0002] With the ever-increasing demand for artificial intelligence (AI) and large-scale data processing, such as images, audio, and video, requiring efficient real-time processing, the inherent von Neumann bottleneck of traditional computing architectures has spurred an urgent need for high efficiency and low power consumption. Neuromorphic computing has attracted significant attention due to its high efficiency, low power consumption, and high integration. At the hardware level, neural network systems require artificial synapses and neurons. Neurons are primarily constructed using traditionally complex complementary metal-oxide-semiconductor (CMOS) technology. In contrast, volatile threshold-switching memristors, by mimicking the processes of biological neurons, achieve threshold-switching performance with minimal circuitry.

[0003] Currently, research on memristor-based artificial neurons is still in its early stages. Although memristors have high integration and promising applications, challenges remain in building neural systems due to issues such as threshold performance and reliability. Research has revealed that various methods can mitigate these challenges, such as doping and rough electrodes, while system thermal design engineering methods are still not perfect. Among various thermal storage materials, NbO based on the Poole-Frenkel mechanism... x Threshold-switched memristors have attracted much attention due to their excellent switching speed and durability. While the methods mentioned above can reduce the threshold voltage, they do not take into account the heat dissipation problem during operation caused by the Poole-Frenkel mechanism, which is highly dependent on the Joule heating effect. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a memristor based on low thermal conductivity intercalation and a multimodal EGC detection system, thereby solving the problems in the prior art.

[0005] The objective of this invention can be achieved through the following technical solutions: The memristor based on low thermal conductivity intercalation includes a threshold switch layer in the middle, an intercalation layer at the top and bottom of the threshold switch layer, and an electrode on the side of the two intercalation layers facing away from the threshold switch layer. The threshold switching layer is made of NbO. x TaO x or VO x The intercalation material is CdTe2; the electrode is an inert electrode.

[0006] Furthermore, the top electrode has a thickness of 15 nm, the bottom electrode has a thickness of 25 nm, the threshold switching layer has a thickness of 20 nm, and the two intercalation layers have the same thickness of 0-18 nm.

[0007] Furthermore, both electrodes are made of Pt.

[0008] The above-mentioned application of memristors based on low thermal conductivity intercalation in the fabrication of a multimodal ECG detection system.

[0009] A multimodal ECG detection system includes: three parallel branches, a feature fusion module, and a neural network; Each branch uses the memristor and corresponding activation function described above to extract the signal features of the electrocardiogram signal; the feature fusion module fuses the extracted signal features into a feature vector, which is then classified by the neural network to detect abnormal heart rates.

[0010] Furthermore, the intercalation thicknesses of the memristors in the three branches are 0 nm, 5 nm, and 10 nm, respectively.

[0011] Furthermore, the activation function corresponding to a memristor with 0nm intercalation on both sides is the sigmoid activation function: Where x is the preprocessed electrocardiogram signal.

[0012] Furthermore, the activation function corresponding to the tanh activation function for each of the two 5nm intercalated memristors is: Where x is the preprocessed electrocardiogram signal.

[0013] Furthermore, the activation function corresponding to the memristor with 10nm intercalation on both sides is the exponential activation function: Where x is the preprocessed electrocardiogram signal.

[0014] Furthermore, the activation function of the neural network is the ReLU activation function.

[0015] The beneficial effects of this invention are: 1. This invention utilizes the thermal characteristics of the Poole-Frenkel mechanism, employing intercalation layers with low thermal conductivity and high electrical conductivity to limit heat dissipation. This not only reduces the threshold voltage but also improves the stability of the threshold switch. Simultaneously, changing the intercalation layers allows for adjustment of the memristor's oscillation frequency, enabling it to have a larger oscillation frequency at lower operating voltages.

[0016] 2. Considering the compatibility between physical devices and neuromorphic computing requirements in practical applications, the temperature and thickness characteristics of memristors were applied to biological signal processing, leading to the development of a multimodal, integrated memristor system for ECG arrhythmia detection. This provides a new strategy for developing low-power ECG detection devices and paves the way for highly reliable neuromorphic hardware systems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the memristor structure of the present invention; Figure 2 This is a graph showing the relationship between the threshold voltage of the memristor of the present invention and the intercalation thickness under different ambient temperatures; Figure 3 This is a graph showing the relationship between the frequency and operating voltage of the memristor of the present invention under different intercalation thicknesses and different ambient temperatures; Figure 4 This is a schematic diagram of the framework of the multimodal EGC detection system of the present invention; Figure 5 This is a graph showing the performance evaluation results of the multimodal model of this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 like Figure 1 As shown, a memristor based on low thermal conductivity intercalation includes a threshold switch layer in the middle, an intercalation layer at the top and bottom of the threshold switch layer, and an electrode on the side of the two intercalation layers facing away from the threshold switch layer, which are respectively denoted as the top electrode and the bottom electrode. In this embodiment, the threshold switching layer is made of NbO. x Both the top and bottom electrodes are made of Pt, and the intercalation material is CdTe2, which has a thermal conductivity of about 0.33 W / (m•K) at 300 K and decreases with increasing temperature, thus confining Joule heat to the threshold switching layer. At the same time, its excellent conductivity can reduce the influence of parasitic resistance on the threshold switching characteristics.

[0021] In this embodiment, the thickness of the top electrode is 15 nm, the thickness of the bottom electrode is 25 nm, the thickness of the two intercalation layers is the same, both being 5 nm; the thickness of the threshold switching layer is 20 nm.

[0022] In other embodiments, the thickness of the intercalation layer is 0-18 nm; the material of the threshold switching layer can also be TaO. x VO x The electrode material can be other inert electrodes besides Pt, such as Ti, W, TiN, etc.

[0023] Example 2 In this embodiment, the threshold switching performance and frequency characteristics of the memristor in Embodiment 1 are studied. (1) The effect of ambient temperature on the threshold voltage as a function of intercalation thickness; The experimental procedure was as follows: a triangular wave current with an amplitude of 0.23mA and a duration of 0.2ms was applied to the top electrode, the low electrode was grounded, and the memristor was intercalated on both sides at 3nm, 5nm, 7nm and 9nm respectively, so that the IV curves were measured at different ambient temperatures of 296K, 301K, 3036K and 311K.

[0024] The relationship between threshold voltage and intercalation thickness under different ambient temperatures is as follows: Figure 2 As shown, by adding intercalation layers on both sides of the threshold switching layer, the threshold voltage is reduced and the stability of the threshold switch is improved. Under ambient temperatures ranging from 296K to 311K, the threshold voltage is statistically analyzed by increasing the thickness of the intercalation layers (3nm-9nm). The change in threshold voltage is used to quantify thermal stability. V th1 V th2 These represent the threshold voltages at the lowest and highest temperatures, respectively. Using this method, the stability of the threshold switch is improved by 24.7%.

[0025] (2) The influence of intercalation thickness and ambient temperature on the relationship between memristor frequency and operating voltage; The experimental procedure was as follows: a 15kΩ resistor was connected in series with the memristor, and a 58pF capacitor was connected in parallel with the memristor to construct a LIF neuron circuit for frequency measurement. The relationship between the memristor frequency and operating voltage was measured at different thicknesses under a fixed ambient temperature of 301K; and the relationship between the memristor frequency and operating voltage was measured at different ambient temperatures with 5nm intercalation layers on each side.

[0026] The relationship between memristor frequency and operating voltage under different intercalation thicknesses and ambient temperatures is as follows: Figure 3 As shown; where, Figure 3(a) in the figure reflects the relationship between the memristor frequency and the operating voltage under different intercalation thicknesses; Figure 3 (b) reflects the relationship between the memristor frequency and the operating voltage under different ambient temperatures.

[0027] like Figure 3 As shown in (a), by changing the thickness of the intercalation structure, the memristor can achieve a larger oscillation frequency at a lower voltage; as Figure 3 As shown in (b), the oscillation frequency of the memristor is increased to a certain extent by changing the ambient temperature.

[0028] Example 3 In this embodiment, based on the memristor extracted in Example 1, and combined with the frequency response mechanism of temperature and thickness of intercalated memristors, a multimodal ECG detection system is proposed. like Figure 4 As shown, the multimodal ECG detection system includes three parallel branches, a feature fusion module, and a neural network. Each branch of the multimodal ECG detection system uses a memristor with a specific intercalation thickness and a corresponding activation function to extract signal features from the ECG signal. The feature fusion module fuses the extracted signal features into a 51-dimensional feature vector, which is then classified by a neural network with a ReLU activation function to detect abnormal heart rates. The system achieves an accuracy of 90.0%.

[0029] In this embodiment, the three branches are as follows: 1) The memristor branches with 0nm intercalation on both sides adopt the sigmoid activation function to simulate fast switching and capture the rapidly changing QRS complex. 2) The two-sided intercalation has a 5nm branch and uses the tanh activation function, which corresponds to a medium switching speed and is beneficial for detecting bidirectional changes in P-waves and T-waves.

[0030] 3) Each of the two intercalation layers has a 10nm branch and uses an exponential activation function to reflect a slower response speed, which is used to extract the baseline trend. The expressions for each activation function are shown in Table 1.

[0031] Table 1. Activation function expressions for each branch in, This indicates the electrocardiogram signal after preprocessing.

[0032] Furthermore, the feature fusion module extracts the high-frequency, mid-frequency, and low-frequency characteristics of heart rate and fuses them into a multi-dimensional feature vector, including frequency characteristics, time change rate, temperature, etc. In this embodiment, the feature fusion module is essentially a concatenation of "multi-branch temperature-sensing convolution + soft-weighted gating fusion + temperature-sensing multi-head self-attention + fully connected classification head": three convolutional branches (0nm / 3-7-15 cores, 5nm / 5-11-21 cores, 10nm / 7-15-31 cores) first extract high / mid / low-frequency features and activate them through temperature modulation; then, the branch weights are calculated through an adaptive soft-weighted gating layer (GlobalAvgPool + Dense 64→32→softmax), the branch features are projected onto the same dimension and summed according to the weights, and refined by Dense(128,ReLU)+BN+Dropout; the fusion result enters an 8-head temperature-sensing self-attention (d_model=128) to modulate the attention distribution using temperature weights; finally, it is processed by a fully connected classification head (Dense... (256→128→64→softmax) Outputs the probability of arrhythmia category.

[0033] The neural network has a ReLU activation function with 100×50 neurons, which is used to classify feature vectors and distinguish the types of arrhythmias.

[0034] Example 4 In this embodiment, the detection accuracy of the multimodal ECG detection system proposed in Example 3 is experimentally tested. The experimental procedure was as follows: Key features of the electrocardiogram (ECG) signal were extracted using three parallel branches, fused into a 51-dimensional feature vector, and finally classified using a neural network with a ReLU activation function. During the experiment, the temperature range was 301K-316K, and the linear gain relationship was observed. The thickness is reflected in three branches with different frequency characteristics: 0nm, 5nm, and 10nm. System validation uses the MIT-BIH arrhythmia library with a sampling rate of 360MHz, injecting Gaussian noise of σ=0.05, ±10 sample offsets, and amplitude scaling of 0.8-1.2 times to enhance the model's generalization ability. Performance evaluation uses a hierarchical 5-fold cross-validation method.

[0035] Experimental results are as follows Figure 5 As shown, the multimodal fusion system has a superior accuracy of 90.0% compared to other deep learning models.

[0036] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A memristor based on low thermal conductivity intercalation, characterized in that, It includes a threshold switch layer in the middle, with an intercalation layer at each of the upper and lower ends of the threshold switch layer, and an electrode on each of the two intercalation layers facing away from the threshold switch layer. The material of the threshold switch layer is NbO x , TaO x or VO x ; the intercalated material is CdTe2; and the electrode is an inert electrode.

2. The memristor based on low thermal conductivity intercalation according to claim 1, characterized in that, The electrode at the top has a thickness of 15 nm, and the electrode at the bottom has a thickness of 25 nm; the threshold switching layer has a thickness of 20 nm, and the two intercalation layers have the same thickness of 0-18 nm.

3. The memristor based on low thermal conductivity intercalation according to claim 1, characterized in that, Both electrodes are made of Pt.

4. The application of the memristor based on low thermal conductivity intercalation as described in any one of claims 1-3 in the fabrication of a multimodal ECG detection system.

5. A multimodal ECG detection system, characterized in that, include: Three parallel branches, a feature fusion module, and a neural network; Each branch uses the memristor and corresponding activation function as described in any one of claims 1-3 to extract signal features of the electrocardiogram signal; feature The fusion module fuses the extracted signal features into a feature vector, which is then classified by a neural network to detect abnormal heart rates.

6. The multimodal ECG detection system according to claim 5, characterized in that, The intercalation thicknesses of the memristors in the three branches are 0 nm, 5 nm, and 10 nm, respectively.

7. The multimodal ECG detection system according to claim 6, characterized in that, The activation function for a memristor with 0nm intercalation on both sides is the sigmoid activation function: Where x is the preprocessed electrocardiogram signal.

8. A multimodal ECG detection system according to claim 6, characterized in that, The activation function for a memristor with 5nm intercalation on both sides is the tanh activation function: Where x is the preprocessed electrocardiogram signal.

9. A multimodal ECG detection system according to claim 6, characterized in that, The activation function for a memristor with 10nm intercalation on each side is the exponential activation function: Where x is the preprocessed electrocardiogram signal.

10. A multimodal ECG detection system according to claim 5, characterized in that, The activation function of the neural network is the ReLU activation function.

Images