RRAM-based blood test system and method

Abstract

The present invention relates to the technical field of integrated circuit design. Disclosed are an RRAM-based blood test system and method, used for solving the problem of how to improve the test efficiency and accuracy of blood test. The system comprises: a calculation chip, an RRAM chip, a power supply module, and a sensor module; the calculation chip is connected to the RRAM chip; the RRAM chip is connected to the sensor module; the power supply module is separately connected to the calculation chip and the RRAM chip; the sensor module is used for testing substances in blood and converting test result data into a plurality of pieces of first voltage data; the RRAM chip is used for carrying out data preprocessing on the plurality of pieces of received first voltage data to obtain second current data; and the calculation chip is used for performing function calculation on the second current data according to a preset function calculation model and on the basis of the second current data to obtain a classification result for the blood. In this way, the blood is tested by means of an integrated circuit, and the efficiency and accuracy of disease screening in blood are improved.

Description

A blood detection system and method based on RRAM

[0001] This application claims priority of a Chinese patent application filed with the China Patent Office on March 29, 2024, with application number 202410384193.9 and invention name “A blood detection system and method based on RRAM”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present invention relates to the technical field of integrated circuit design, and in particular to a blood detection system and method based on RRAM. Background Art

[0003] Currently, disease screening based on blood or urine tests is a common medical testing method in hospitals. However, whether in actual medical scenarios or scientific research, this type of disease screening through blood tests has many defects. For example, when drawing blood for a physical examination of a user, multiple tubes must be drawn to screen the required proteins or enzymes, and then a comprehensive disease screening can be performed. The equipment used is a relatively bulky medical analysis instrument, which makes the testing scene very fixed and inconvenient to move. After collecting the data, the data needs to be uploaded to the cloud and analyzed using a PC to obtain the analysis results. Obviously, the entire testing process from blood sampling to bleeding sample reporting is relatively long, and the blood sample stays in the blood vessel for a long time. At the same time, the analysis results cannot reflect the type of disease. This leads to low efficiency and poor accuracy in blood disease screening, which cannot meet user needs.

[0004] Summary of the Invention

[0005] The purpose of the present invention is to provide a blood testing system and method based on RRAM, which is used to solve the problems of the existing technology in that the entire process of blood testing, from blood sampling to bleeding sample reporting, is relatively long, and the analysis results cannot reflect the type of disease, resulting in low efficiency and poor accuracy in blood disease screening.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, embodiments of this specification provide a blood testing system based on RRAM, which may include:

[0008] A computing chip, an RRAM chip, a power module, and a sensor module; the computing chip is connected to the RRAM chip, the RRAM chip is connected to the sensor module, and the power module is connected to the computing chip and the RRAM chip respectively;

[0009] The sensor module is used to detect substances in the blood and convert the detection result data into a plurality of first voltage data;

[0010] The RRAM chip is used to perform data preprocessing on the received plurality of first voltage data to obtain second current data;

[0011] The computing chip is used to perform function calculation on the second current data according to a preset function calculation model based on the second current data to obtain a classification result for the blood; the classification result is used to represent a disease screening result for the blood.

[0012] In a second aspect, embodiments of this specification further provide an RRAM-based blood testing method, which is applied to the RRAM-based blood testing system described in the first aspect and may include:

[0013] Acquire a plurality of first voltage data sent by the sensor module;

[0014] Based on the plurality of first voltage data, preprocessing the plurality of first voltage data to obtain second current data;

[0015] Based on the second current data, a function calculation is performed on the second current data according to a preset function calculation model to obtain a classification result for the blood; the classification result is used to represent a disease screening result for the blood.

[0016] Compared with the prior art, the present invention provides an RRAM-based blood testing system, which is provided with a computing chip, an RRAM chip, a power module and a sensor module; the computing chip is connected to the RRAM chip, the RRAM chip is connected to the sensor module, and the power module is connected to the computing chip and the RRAM chip respectively; the sensor module is used to detect substances in the blood and convert the detection result data into multiple first voltage data; the RRAM chip is used to perform data preprocessing on the received multiple first voltage data to obtain second current data; the computing chip is used to perform function calculation on the second current data according to a preset function calculation model based on the second current data to obtain a classification result for the blood; the classification result is used to represent the disease screening result for the blood; thereby, it is achieved that the detection of diseases present in the blood can be quickly completed using a single tube of blood, and the classification result can be output to reflect the degree of verification of the disease; and the efficiency and accuracy of screening for diseases in the blood are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0018] FIG1 is a system architecture diagram of an RRAM-based blood testing system provided by the present invention.

[0019] FIG2 is a first structural diagram of main components of a blood testing system based on RRAM provided by the present invention.

[0020] FIG3 is a second structural diagram of main components of an RRAM-based blood testing system provided by the present invention.

[0021] FIG4 is a schematic diagram of the structure of an IGZO sensor array of an RRAM-based blood detection system provided by the present invention.

[0022] FIG5 is a schematic diagram of the structure of a 1T1R memristor unit of an RRAM-based blood testing system provided by the present invention.

[0023] FIG6 is a flow chart of a blood detection method based on RRAM provided by the present invention. DETAILED DESCRIPTION

[0024] To facilitate a clear description of the technical solutions of the embodiments of the present invention, the words "first" and "second" are used in the embodiments of the present invention to distinguish between identical or similar items with substantially the same functions and effects. For example, the first threshold and the second threshold are merely used to distinguish between different thresholds and do not limit their order. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0025] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0026] In the present invention, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or multiple.

[0027] In the existing technology, disease screening based on blood has many defects. For example, when drawing blood for a patient's physical examination, multiple tubes must be drawn to screen for the required proteins or enzymes for comprehensive disease screening; there are also problems such as medical analysis equipment being fixed in the testing location, being bulky and inconvenient to use, etc.; as a result, during the blood test process, disease screening results cannot be quickly issued and the type of disease cannot be analyzed; therefore, how to draw a small amount of blood from the patient, perform rapid disease screening on the blood, and obtain disease classification results is a technical problem that urgently needs to be solved.

[0028] In view of this, this specification provides a blood testing system based on RRAM. By drawing a small amount of blood from the patient, such as 10 ml of blood, it is possible to quickly screen the underlying diseases in the blood sample and obtain the classification results of the underlying diseases; thereby improving the efficiency and accuracy of blood disease screening. The technical solution of the present invention is described below with reference to the accompanying drawings:

[0029] Please refer to FIG1 , which is a system architecture diagram of a blood testing system based on RRAM provided by the present invention.

[0030] In Figure 1, the RRAM-based blood testing system can include a three-layer integrated circuit structure. The top layer is the sensor module layer, the middle layer is the RRAM chip analog storage and computing integrated circuit, and the bottom layer is the computing chip circuit.

[0031] Since the algorithm deployment weights required for blood testing application scenarios are relatively large, in order to save resources and reduce energy consumption, the memory-computing integrated circuit based on memristors proposed in the embodiments of this specification is a good solution for implementing analog memory-computing. Memristors implement multiple conductance state modulations on RRAM devices, corresponding to the multiple weights in the matrix multiplication and addition calculations of convolutional neural networks, and use the weights programmed on RRAM to process the input data of the sensor module. And due to the non-volatility of memristors, if the memristor loses power, its set weights will not disappear, and there is no need to reprogram and set the weights before the next logical reasoning, which greatly improves efficiency; and the memory-computing integrated circuit based on memristors also has great advantages in reducing power consumption in realizing edge computing. Therefore, we design a memory-computing integrated circuit based on memristors to realize the calculation and classification functions of convolutional neural networks, and combine it with IGZO sensor arrays and computing chips to complete the overall blood detection circuit based on RRAM convolutional neural networks, which can achieve the technical effect of efficiently screening for diseases in the blood and detecting the degree of disease.

[0032] Further, please refer to FIG2 , which is a first structural diagram of main components of a blood testing system based on RRAM provided by the present invention.

[0033] In Figure 2, the main component structure of a blood testing system based on RRAM provided by the present invention may include:

[0034] A computing chip 1300, an RRAM chip 1200, a power module 1400 and a sensor module 1100; the computing chip 1300 is connected to the RRAM chip 1200, the RRAM chip 1200 is connected to the sensor module 1100, and the power module 1400 is connected to the computing chip 1300 and the RRAM chip 1200 respectively.

[0035] The sensor module 1100 is used to detect substances in the blood and convert the detection result data into a plurality of first voltage data; the RRAM chip 1200 is used to perform data preprocessing on the received plurality of first voltage data to obtain second current data.

[0036] The computing chip 1300 is used to perform function calculation on the second current data according to a preset function calculation model based on the second current data to obtain a classification result for the blood; the classification result is used to represent a disease screening result for the blood.

[0037] It should be noted that the RRAM-based blood testing system provided by the present invention is designed to screen for early disease and determine the severity of a disease by detecting substances in the blood, such as proteins or enzymes. Its main components can be an IGZO TFT sensor array combined with an RRAM array and a computational control chip. The IGZO TFT sensor array is responsible for substance detection. Blood drips into the sensor array, converting the substance information in the blood into electrical signals, such as blood concentration. The RRAM array then stores neural network weights, and the FPGA logic control module controls the integrated multiplication and addition operations. The computational chip is responsible for implementing the activation functions, pooling operations, and softmax functions required by the inference algorithm. Regarding the detection channels, they can be set according to needs, that is, they can be achieved by configuring different numbers of computing chips 1300, RRAM chips 1200 and sensor modules 1100; the following settings are used as examples in the embodiments of this specification, such as setting 20 detection input channels (15 detection channels + 5 blank controls), and 8 classification output channels / 2 classification output channels: two classification modes can be achieved: one is to perform early disease screening, and the second classification output is to see whether there is a risk of several underlying diseases; the second is to accurately judge the degree of a certain disease, which is divided into eight degrees. The inference algorithm can be changed by changing the weight deployment of the RRAM array to switch the classification mode.

[0038] Preferably, please refer to FIG3 , which is a second structural diagram of main components of a blood testing system based on RRAM provided by the present invention.

[0039] In FIG. 3 , the computing chip 1300 may include: a pooling unit 1320 connected to a RELU activation unit 1310 , and a SOFTMAX function unit 1330 connected to the pooling unit 1320 .

[0040] The RELU activation unit 1310 is used to activate the mathematical function of the convolution layer in the comprehensive calculation module, so that the calculation network model of the comprehensive calculation module has nonlinear capabilities; the pooling unit 1320 is used to complete the pooling layer function and reduce the size of the feature map generated by the convolution layer in the comprehensive calculation module; the SOFTMAX function unit 1330 is used to perform classification comparison based on the output values ​​of a preset number of output channels, and take the channel corresponding to the maximum output value as the judgment result of the blood.

[0041] Furthermore, the RRAM-based blood testing system provided in the embodiments of this specification may also include: an FPGA logic control module 1600 , an ADC module 1800 , a DAC module 1700 and a second readout circuit module 1500 .

[0042] The FPGA logic control module 1600 is connected to the RRAM chip 1200; one end of the second readout circuit module 1500 is connected to the computing chip 1300, and the other end is connected to the RRAM chip 1200; one end of the ADC module 1800 is connected to the computing chip 1300, and the other end is connected to the FPGA logic control module 1600; one end of the DAC module 1700 is connected to the RRAM chip 1200, and the other end is connected to the FPGA logic control module 1600.

[0043] The FPGA logic control module 1600 is used to perform instruction control and data transmission on the modules connected to it; the ADC module 1800 is used to convert the input analog signal into a digital signal output; the DAC module 1700 is used to convert the input digital signal into an analog signal output; the second readout circuit module 1500 is used to read the second current data output by the RRAM chip 1200.

[0044] Specifically, please refer to FIG4 , which is a schematic diagram of the IGZO sensor array structure of an RRAM-based blood detection system provided by the present invention.

[0045] In Figure 4, an IGZO sensor array performs blood testing, converting substance concentration signals into electrical signals. The IGZO TFT (Indium Gallium Zinc Oxide Thin Film Transistor) has three ports: G, D, and S. IGZO TFTs react with proteins or enzymes to generate current. Therefore, the IGZO sensor array can convert protein or enzyme concentrations in the blood into current signals.

[0046] Furthermore, the output current of the IGZO sensor array undergoes data processing through a first readout circuit module, which includes a transimpedance amplifier and a fully differential filter for noise reduction and amplification, converting the current signal into a voltage signal. The transimpedance amplifier converts the output current signal from the IGZO into a voltage signal that is stably input into the RRAM array. The fully differential filter prevents noise aliasing caused by sampling during data conversion. The bandwidth setting can be limited to 30kHz.

[0047] As an example, the sensor module 1100 described in the embodiment of this specification may include an IGZO indium gallium zinc oxide thin film transistor sensing array module 1110 and a first readout circuit module 1120; one end of the first readout circuit module 1120 is connected to the IGZO indium gallium zinc oxide thin film transistor sensing array module 1110, and the other end is connected to the RRAM chip 1200.

[0048] The IGZO indium gallium zinc oxide thin film transistor sensing array module 1110 is used to detect substances in the blood and obtain multiple current detection data; the first readout circuit module 1120 is used to read out multiple current detection data and convert the multiple current detection data into multiple first voltage data.

[0049] Further, please refer to FIG5 , which is a schematic diagram of the structure of a 1T1R memristor unit of an RRAM-based blood testing system provided by the present invention.

[0050] In FIG5 , the input of the RRAM array is the analog voltage output by the first readout circuit module. Matrix multiplication and addition calculations are performed based on the RRAM storage and calculation principle to obtain the calculation result of the output cumulative current.

[0051] For memristor RRAM, the classic operations are read operations (Read) and write operations (Write). Among them, the write operation is further divided into FORMING operation, SET operation and RESET operation. The FORMING operation forms conductive filaments from scratch in the resistive layer by applying a one-time high voltage, causing the memristor to change from the initial ultra-high resistance state to a low resistance state (Low Resistance State, LRS). The SET operation is to apply a voltage pulse with the same polarity as the FORMING voltage to the memristor, but the voltage amplitude is generally smaller, causing the memristor to change from HRS (High Resistance State, HRS) to LRS (Low Resistance State, LRS). The RESET operation is to apply a voltage pulse with the opposite polarity to the FORMING operation to the memristor, interrupting the formed conductive path, causing the device to rise from the low resistance state to the high resistance state again. The read operation applies a read voltage less than the threshold voltage to both ends of the memristor, which does not change the resistance of the memristor, but can read the information in a specific memristor unit.

[0052] Analog resistive switching devices typically possess bidirectional continuous resistive switching capabilities, enabling continuous adjustment of the memristor's conductance state during both SET and RESET operations. Using a constant-amplitude voltage pulse for SET or RESET operations can achieve a resistive switching process from the memristor's minimum conductance to its maximum conductance, or vice versa, modulating the memristor to a desired intermediate conductance state.

[0053] If the convolutional neural network to be built is large in scale, with a large number of neurons, the number of weights that need to be deployed is also enormous. If only a single memristor is used for weight encoding, since the memristors are interconnected, crosstalk will occur when the conductance state of each memristor is modulated, affecting the weight encoding of other memristors, resulting in a decrease in the accuracy of the function calculation of blood measurement data. However, the use of a 1-Transistor 1-Resistor (1T1R) memristor unit can avoid this problem and improve the accuracy of RRAM output data.

[0054] As shown in Figure 5 (left panel a), each memristor cell has three ports, connected to the bit line (BL), word line (WL), and source line (SL) of the memristor array. The programming voltage amplitude and programming pulse duration applied to a single memristor cell determine the corresponding memristor cell's operation. Figure 5 (center panel b) illustrates the operating conditions for a memristor cell write operation. During the FORMING operation, the SL terminal of the 1T1R structure is grounded, BL is connected to the FORMING voltage, and WL is connected to a high level to select the memristor cell. The SET operation is similar to the FORMING operation, except that the SET voltage amplitude is smaller. During the RESET operation, the SL terminal of the 1T1R structure is connected to the RESET voltage, BL is grounded, and WL is connected to a high level to select the memristor cell. The read operation is shown in Figure 5 (right panel c), where the read voltage Vread is connected to BL, SL is grounded, and WL is connected to a high level to select the memristor cell. It can be seen that through this 1T1R structure, it is very simple to control the on and off of a memristor in the memristor array. The current path between the memristors is isolated, and the conductivity state of each memristor can be programmed without affecting other devices, achieving more accurate output current data, thereby improving the accuracy of blood sample testing by the entire blood testing system.

[0055] Furthermore, the computing chip 1300 may include three major modules: a RELU activation unit 1310, a pooling unit 1320, and a softmax function unit 1330. The functions of each unit are described as follows:

[0056] The RELU activation unit 1310 primarily utilizes a comparator after the convolutional layer to implement the activation function, giving the network nonlinear capabilities. The RELU function was chosen here because it solves the problem of vanishing gradients in the positive range; it has fast computational speed, simple discrimination, and converges much faster than Sigmoid and Tanh, further improving the efficiency of data analysis for blood sample testing data.

[0057] The main function of pooling unit 1320 is to complete the pooling layer function, reducing the size of the feature map generated by the convolutional layer and reducing the amount of computation. Here, maximum pooling is performed, achieving a quadruple pooling effect, reducing the amount of computation by a factor of four. The pooling module is connected after the RELU activation module and performs maximum pooling on the activation result. The pooled output is then connected to the next layer of network calculation.

[0058] The main function of the SOFTMAX function unit 1330 is to complete the softmax function and obtain the final classification output result. It should be noted that the number of channels for the softmax function classification can be set according to actual needs and is not specifically limited in this specification. In the embodiment of this specification, the probability distribution output of the softmax function is changed to a specific number of channels 0-7 for example only. Because the softmax function outputs the value with the maximum classification probability, it can be directly compared to determine which channel has the largest value and output the corresponding number of channels to achieve the classification of eight diseases.

[0059] Based on this, an embodiment of the present invention provides a blood testing circuit system capable of screening for basic diseases. This circuit system integrates sensing and computing, and can calculate and process information on the concentration of substances in the blood to derive the risk and severity of the basic disease. By adopting an edge intelligent computing chip, the system power consumption can be reduced. At the same time, the integrated design of sensing, information processing and computing also improves the speed and safety of the screening system, avoiding the need for bulky medical analysis equipment that makes the detection scene very fixed and inconvenient; and avoiding the need for a user to draw blood from multiple tubes to screen for the required proteins or enzymes during a physical examination. A small amount of blood can be used for comprehensive disease screening, thereby improving the efficiency and accuracy of disease screening in the blood.

[0060] In a second aspect, based on the same technical concepts as the first aspect, embodiments of this specification further provide an RRAM-based blood testing method, which is applied to the RRAM-based blood testing system disclosed in the first aspect. Please refer to Figure 6, which is a schematic flow chart of an RRAM-based blood testing method provided by the present invention.

[0061] In FIG6 , the method may include:

[0062] Step 610: Acquire a plurality of first voltage data sent by the sensor module.

[0063] Step 620: Based on the plurality of first voltage data, pre-process the plurality of first voltage data to obtain second current data.

[0064] Step 630: Based on the second current data, perform function calculation on the second current data according to a preset function calculation model to obtain a classification result for the blood; the classification result is used to represent a disease screening result for the blood.

[0065] In steps 610 to 630, the RRAM chip obtains multiple first voltage data sent by the upper-layer sensor module, performs positive and negative weighting processing on the multiple first voltage data according to the neural network model preset in the RRAM chip, and performs data bias to obtain second current data. The second current data is input into the lower-layer computing chip. The computing chip uses a pre-set function to perform function calculation on the second current data to obtain a screening and classification result for the blood. Two classification modes can be achieved: one is early disease screening; the other is determining the risk of several underlying diseases in patients and accurately judging the severity of a certain underlying disease.

[0066] Preferably, before step 610, that is, before obtaining the multiple first voltage data sent by the sensor module, it can include: using the sensor module to detect substances in the blood to obtain multiple current data; preprocessing the multiple current data to convert the multiple current data into multiple voltage data; determining the multiple voltage data as multiple first piezoelectric data; the preprocessing process includes at least one of using a transimpedance amplifier to amplify the multiple current data and using a fully differential filter to denoise the multiple current data.

[0067] Specifically, a transimpedance amplifier and a fully differential filter perform denoising and amplification, converting the current signal into a voltage signal. The transimpedance amplifier converts the output current signal from the IGZO into a voltage signal that is stably input into the RRAM array. The fully differential filter denoises and amplifies the electrical signal, preventing noise aliasing caused by sampling during data conversion, thereby improving the accuracy of blood sample testing. The bandwidth during data processing can be limited to 30kHz.

[0068] Preferably, in step 620, preprocessing the plurality of first voltage data may include: performing positive weighting processing on the plurality of first voltage data based on the memristor value of the RRAM to obtain a first column of cumulative data; and performing negative weighting processing on the plurality of first voltage data to obtain a second column of cumulative data; and obtaining the second current data based on the first column of cumulative data and the second column of cumulative data.

[0069] Furthermore, first differential data can be obtained based on the first column of accumulated data and the second column of accumulated data; and based on the first differential data, data biasing is performed on the first differential data according to a preset biasing strategy to obtain second current data.

[0070] As an example, the voltage input signal of a memristor array can be encoded as the input vector of a convolutional neural network. The resistance value (conductance state) of the memristors can be encoded as the weight value of the convolution kernel. The output current of the memristor array can represent the result of the convolution. Inputting the output current into the neurons completes the forward inference process of the neural network. According to Ohm's law, applying a certain voltage to a conductor and obtaining a current value is equivalent to performing a multiplication operation. According to Kirchhoff's current law, multiple rows of output current values ​​aggregate the current flowing through each device, equivalent to performing a cumulative addition operation. Finally, the current values ​​output by each row are combined in a multiplication-accumulation operation, resulting in the result of matrix-vector multiplication. The input voltage signal matrix (V1, V2, V3, V4) is multiplied by the memristor values ​​G set in the array to obtain the output current signals (I1, I2, I3, I4). This completes a matrix multiplication operation and yields the result of the matrix-vector multiplication. The memristor values ​​G include both positive and negative memristor values.

[0071] The calculation result of matrix-vector multiplication can be calculated using the following formula:

[0072] Among them, the values ​​of i and j represent the corresponding channels in the matrix, V i Represents the voltage signal matrix, G ij Represents the corresponding memetic derivative in the matrix.

[0073] The computational complexity of traditional multiplication of an n-dimensional vector and an n×n matrix is ​​O(n2), while the complexity of matrix-vector multiplication using a memristor array based on storage and computing is reduced to O(1), which can greatly reduce the resources and time consumed by the operation and accelerate the convolutional neural network.

[0074] Because the values ​​in the convolution kernel's weight matrix can include positive, negative, and zero values, while the conductance values ​​of memristors are non-negative, the common practice is to use two memristors to represent each convolution kernel weight. First, the 3×3 convolution kernel is converted into a one-dimensional column vector of size 9, and the values ​​of its elements are mapped one by one to the memristor array, where every two memristors represent the weight value of a convolution kernel. Then, the elements of the input image array are converted into voltage signals, that is, [X1, X2, ..., X9] represents the pulse voltage value transmitted to the memristor array, and the input signal is transmitted to the array along the horizontal line. Based on Ohm's law and Kirchhoff's law, the total current obtained along the vertical line is proportional to the weighted input signal value flowing through the memristor array. Finally, the output data of the two columns of memristors are aggregated through a differential output circuit to obtain the final matrix-vector multiplication result.

[0075] The output data of two columns of memristors are aggregated through the differential output circuit, and the following formula can be used:

[0076] Among them, g i + and g i - represents the memristor value corresponding to the i-th pair of memristors, X i Represents the input voltage signal, R b represents the input resistance, and Y1 represents the output data of the two columns of memristors aggregated through the differential output circuit.

[0077] The accumulated current is fed to the computing chip for activation. After the REUL function and pooling operations are implemented, the result can be passed to the RRAM array for the same multiplication-addition calculation as in the above embodiment to obtain the fully connected layer output of the RRAM, and the output result is input to the computing chip.

[0078] As an example, if the two columns of current values ​​are 20 (positive weight column) and 30 (negative weight column), the bias is set to 2, the result of subtracting and adding the bias is -8, and the first output data after the ReLU function is 0; if the two columns of current values ​​are 30 (positive weight column) and 20 (negative weight column), the bias is set to 2, the result of subtracting and adding the bias is 12, and the first output data after the ReLU function is 12.

[0079] Furthermore, data processing is performed through the function calculation model preset by the computing chip to obtain the classification result.

[0080] Preferably, in step 630, performing function calculation on the second current data according to a preset function calculation model may include: based on the second current data, using a RELU activation unit and adopting the formula:

[0081] Among them, x represents the input value of the ReLU function, the specific data is the first bias data, and ReLU(x) is the first output data of the RELU activation unit.

[0082] Furthermore, performing function calculation on the second current data according to a preset function calculation model may include: based on the first output data of the RELU activation unit, using the pooling unit to perform maximum pooling on the first output data of the RELU activation unit to obtain first pooled output data.

[0083] Specifically, performing function calculation on the second current data according to a preset function calculation model may further include: based on the first pooled output data, using the RRAM chip to perform data preprocessing on the first pooled output data to obtain third output data; wherein, the method of performing data preprocessing on the first pooled output data using the RRAM chip is the same as the method of performing data preprocessing using the RRAM chip disclosed in the above step 620, and will not be repeated here.

[0084] Based on the third output data, the SOFTMAX function unit adopts the formula:

[0085] Among them, e ηi is the input value of the i-th channel, e ηj is the input value of the jth channel, k is the number of channels, and softmax(i) is the output data of the corresponding i-th channel output by the SOFTMAX function unit.

[0086] Furthermore, based on the output data of the corresponding i-th channel output by the SOFTMAX function unit, the values ​​corresponding to the output data of each channel are compared, and the channel corresponding to the maximum value is used as the classification result of the blood.

[0087] As an example, the probability distribution output of the softmax function can be modified to output specific channel numbers 0-7. Since the softmax function outputs the value with the highest classification probability, the softmax function unit directly compares the channel with the highest value and outputs the corresponding channel number. The softmax function unit can implement eight-channel data classification. After obtaining the multiplication and addition results of the fully connected layer, a comparative calculation is performed to obtain the value of the channel with the highest classification probability, which is the final classification result. For example, if the first current is the largest, the output is 0, indicating the lowest disease severity. Similarly, if the eighth current is the largest, the output is 7, indicating the highest disease severity.

[0088] Based on this, an RRAM-based blood detection method proposed in the second aspect of an embodiment of the present invention is applied to an RRAM-based blood detection system proposed in the first aspect, by acquiring multiple first voltage data sent by the sensor module; based on the multiple first voltage data, the multiple first voltage data are preprocessed to obtain second current data; and further based on the second current data, a function calculation is performed on the second current data according to a preset function calculation model to obtain a classification result for the sampled blood; the classification result is used to represent the disease screening result for the blood; the use of integrated circuits to detect blood is realized, thereby improving the efficiency and accuracy of disease screening in the blood.

[0089] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit can implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0090] Although the present invention has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the invention. It will be apparent that various modifications and variations may be made to the present invention by those skilled in the art without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such modifications and variations as fall within the scope of the claims of the present invention and their equivalents.

Claims

1. A blood detection system based on RRAM, characterized in that: At least: A computing chip, an RRAM chip, a power module, and a sensor module; the computing chip is connected to the RRAM chip, the RRAM chip is connected to the sensor module, and the power module is connected to the computing chip and the RRAM chip respectively; The sensor module is used to detect substances in the blood and convert the detection result data into a plurality of first voltage data; The RRAM chip is used to perform data preprocessing on the received plurality of first voltage data to obtain second current data; The computing chip is configured to perform a function calculation on the second current data according to a preset function calculation model based on the second current data to obtain a classification result for the blood; The classification result is used to represent the disease screening result of the blood.

2. The system according to claim 1, wherein The computing chip includes: A pooling unit connected to the RELU activation unit, and a softmax function unit connected to the pooling unit; The RELU activation unit is used to activate the mathematical function of the convolution layer in the comprehensive calculation module, so that the calculation network model of the comprehensive calculation module has nonlinear capabilities; The pooling unit is used to complete the pooling layer function and reduce the size of the feature map generated by the convolution layer in the comprehensive calculation module; The SOFTMAX function unit is used to perform classification comparison based on the output values ​​of a preset number of output channels, and take the channel corresponding to the maximum output value as the judgment result of the blood.

3. The system according to claim 1, wherein: The sensor module includes an IGZO indium gallium zinc oxide thin film transistor sensing array module and a first readout circuit module; one end of the first readout circuit module is connected to the IGZO indium gallium zinc oxide thin film transistor sensing array module, and the other end is connected to the RRAM chip; The IGZO indium gallium zinc oxide thin film transistor sensor array module is used to detect substances in the blood and obtain multiple current detection data; The first readout circuit module is used to read out a plurality of current detection data and convert the plurality of current detection data into a plurality of first voltage data.

4. The system according to claim 1, wherein: Also includes: FPGA logic control module, ADC module, DAC module and second readout circuit module; The FPGA logic control module is connected to the RRAM chip; One end of the second readout circuit module is connected to the computing chip, and the other end is connected to the RRAM chip; one end of the ADC module is connected to the computing chip, and the other end is connected to the FPGA logic control module; one end of the DAC module is connected to the RRAM chip, and the other end is connected to the FPGA logic control module; The FPGA logic control module is used to perform instruction control and data transmission on the modules connected thereto; The ADC module is used to convert the input analog signal into a digital signal for output; The DAC module is used to convert the input digital signal into an analog signal for output; The second readout circuit module is used to read out second current data output by the RRAM chip.

5. A blood detection method based on RRAM, applied to the blood detection system based on RRAM according to any one of claims 1 to 4, characterized in that: Methods include: Acquire a plurality of first voltage data sent by the sensor module; Based on the plurality of first voltage data, preprocessing the plurality of first voltage data to obtain second current data; Based on the second current data, performing function calculation on the second current data according to a preset function calculation model to obtain a classification result for the blood; The classification result is used to represent the disease screening result of the blood.

6. The method according to claim 5, wherein The step of obtaining the plurality of first voltage data sent by the sensor module includes: The sensor module is used to detect substances in the blood to obtain a plurality of current data; Preprocessing the plurality of current data to convert the plurality of current data into a plurality of voltage data; determining the plurality of voltage data as a plurality of first voltage data; the preprocessing process includes at least one of amplifying the plurality of current data using a transimpedance amplifier and denoising the plurality of current data using a fully differential filter.

7. The method according to claim 5, wherein The preprocessing of the plurality of first voltage data includes: Based on the memristor value of the RRAM, performing positive weighting processing on the plurality of first voltage data to obtain a first column of cumulative data; and performing negative weighting processing on the plurality of the first voltage data to obtain a second column of accumulated data; The second current data is obtained based on the first column of accumulated data and the second column of accumulated data.

8. The method according to claim 7, wherein The obtaining of the second current data based on the first column of accumulated data and the second column of accumulated data includes: Obtaining first differential data based on the first column of accumulated data and the second column of accumulated data; Based on the first differential data, performing data bias on the first differential data according to a preset bias strategy to obtain second current data; The performing function calculation on the second current data according to a preset function calculation model includes: Based on the second current data, the RELU activation unit is used, using the formula: Among them, x represents the input value of the ReLU function, the specific data is the first bias data, and ReLU(x) is the first output data of the RELU activation unit.

9. The method according to claim 8, wherein The performing function calculation on the second current data according to a preset function calculation model includes: Based on the first output data of the RELU activation unit, the pooling unit is used to perform maximum pooling on the first output data of the RELU activation unit to obtain first pooled output data.

10. The method according to claim 9, wherein The performing function calculation on the second current data according to a preset function calculation model includes: Based on the first pooled output data, using the RRAM chip to perform data preprocessing on the first pooled output data to obtain third output data; Based on the third output data, the SOFTMAX function unit adopts the formula: in, is the input value of the i-th channel, is the input value of the jth channel, k is the number of channels, and softmax(i) is the output data of the corresponding i-th channel output by the SOFTMAX function unit.

Images