A field effect transistor hydrogen sensor, a preparation method and a hydrogen detection method

Abstract

The application provides a field effect transistor hydrogen sensor, a preparation method and a hydrogen detection method. The sensor comprises a semiconductor substrate, a source electrode and a drain electrode, a channel, a dielectric layer and a sensitive gate electrode. The source electrode and the drain electrode are arranged on the semiconductor substrate in the order of the source electrode, the drain electrode and the source electrode. Two channels with the same size are formed on the surface of the semiconductor substrate. The dielectric layer is arranged on the two channels in a four-array mode. During operation, a threshold voltage is loaded on the sensitive gate electrode and the source electrode, the source electrode is grounded, and a fixed voltage is loaded on the drain electrode. When the sensitive gate electrode is exposed to a hydrogen atmosphere, the resistance of the sensitive gate electrode changes, the threshold voltage of the sensitive gate electrode changes, the current of the channel between the drain electrode and the source electrode changes, and the sensing function is realized.

Description

Technical Field

[0001] This invention relates to the field of sensor fabrication, and in particular to a field-effect transistor hydrogen sensor, its fabrication method, and a hydrogen detection method. Background Technology

[0002] Currently, commercially available MOSFET gas sensors are mainly used to detect explosive gases such as hydrogen and toxic gases such as CO and NO2. MOSFET chemical sensors involve fabricating source and drain regions on a substrate, growing a layer of SiO2 on the substrate surface, forming gates of various shapes on the SiO2 layer above the channel region, and finally fabricating a sensitive layer on the gate. The working principle is that the target analyte (e.g., hydrogen) reacts with the sensor's sensitive layer (e.g., a catalytic metal platinum), and the reaction products diffuse to the MOSFET's gate, altering the device's performance. The target analyte is identified by analyzing the changes in device performance. Changing the type and thickness of the sensor's sensitive layer can optimize sensitivity and selectivity, and can also alter the device's optimal operating temperature.

[0003] Chinese invention patent specification CN112666229A discloses a field-effect transistor (FET) hydrogen sensor, comprising an FET and a hydrogen-sensitive thin film. The gate of the FET is connected to one end of the hydrogen-sensitive thin film, forming a series structure. The contact point at the connection serves as a floating gate (FG), and the other end of the hydrogen-sensitive thin film serves as a control gate (CG). During operation, a fixed voltage is applied to the control gate (CG), the source (S) of the FET is grounded, and a fixed voltage is applied to the drain (D). The floating gate (FG) is made of metal. However, the floating gate structure can lead to electron leakage, affecting the electron retention characteristics on the floating gate and consequently impacting the sensor's control voltage, resulting in reduced gas testing stability. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a field-effect transistor hydrogen sensor, its preparation method, and a hydrogen detection method.

[0005] The first aspect of the present invention provides a field-effect transistor hydrogen sensor, comprising:

[0006] Semiconductor substrate;

[0007] Source and drain electrodes formed on the surface of a semiconductor substrate;

[0008] A channel formed on the surface of a semiconductor substrate between the source and drain electrodes;

[0009] A medium layer formed on the channel;

[0010] The sensitive gate formed on the dielectric layer; and

[0011] A filter layer formed on the sensitive gate;

[0012] In this arrangement, the source and drain electrodes on the semiconductor substrate are arranged in the order of source, drain, and source.

[0013] Two channels of equal size are formed on the surface of the semiconductor substrate;

[0014] The dielectric layer is arranged in a four-array configuration on the two channels;

[0015] During operation, threshold voltages are applied to the sensitive gate and source, with the source grounded and a fixed voltage applied to the drain. When the sensitive gate is exposed to a hydrogen atmosphere, its resistance changes, causing the threshold voltage of the sensitive gate to change, thereby changing the current in the channel between the drain and source, thus achieving the sensing function.

[0016] A second aspect of the present invention provides a method for fabricating a field-effect transistor hydrogen sensor, comprising:

[0017] (1) The source, drain and dielectric layer of the field-effect transistor hydrogen sensor are fabricated on a semiconductor substrate;

[0018] (2) Clean the surface of the dielectric layer;

[0019] (3) A sensitive gate region is formed by setting a mask in the dielectric layer, a palladium film is sputtered in the sensitive gate region, and the palladium film is grown in a controllable manner by magnetron sputtering, thereby forming a sensitive gate.

[0020] (4) A loose and porous SiO2 layer is sputtered on the sensitive gate surface as a filter layer using magnetron sputtering.

[0021] A third aspect of the present invention provides a method for detecting hydrogen gas, wherein a field-effect transistor hydrogen gas sensor prepared by the aforementioned preparation method is used for detection to obtain a sensing signal;

[0022] When outputting hydrogen concentration, the sensing signal is input into the pre-trained sensing signal-hydrogen concentration output model to obtain the corresponding hydrogen concentration.

[0023] The training steps for the induction signal-hydrogen concentration output model are as follows:

[0024] (1) The current change rate between the source and drain of the field-effect transistor hydrogen sensor is used as the sensor sensitivity. A training dataset is obtained through a large number of tests to establish the correspondence between different hydrogen concentrations and sensor sensitivity. A four-element neural network structure is established, in which the sensitivity of each sensor represents a neuron to work, and the detected hydrogen concentration is used as the input value for the next calculation.

[0025] (2) Normalize the input values;

[0026] (3) Weight each input value, with the initial weights being randomly assigned;

[0027] (4) Calculate the residual of each layer: calculate the difference between the predicted output value and the expected value of the training dataset; if it meets the allowable error, it can be output; if the error is large, backpropagation is performed to calculate the residual of each layer.

[0028] Output layer → Hidden layer: Residual = -(Output value - Sample value) * Derivative of activation function;

[0029] Hidden layer → Hidden layer: Residual = (weighted sum of residuals of each node in the right layer) * derivative of activation function;

[0030] (5) Update weights: Update weights according to the error range using the error weighted derivative formula;

[0031] Input layer: Weight increase = Sigmoid function value of the current node * residual of the corresponding node in the right layer * learning rate;

[0032] Hidden layer: Weight increase = input value * residual of the corresponding node in the right layer * learning rate;

[0033] The weight increase of the offset value = the residual of the corresponding node in the right layer * the learning rate;

[0034] (6) Output results: Re-output the results and iterate the process multiple times. In each iteration, the entire training set is processed simultaneously, and finally the comprehensive hydrogen detection concentration is output.

[0035] This invention has significant practical effects and remarkable progress compared to the prior art. Specifically, this invention has the following advantages:

[0036] 1. This invention works on the principle of regulating the source-drain channel current by the gate voltage of a field-effect transistor. Since no large current flows through the sensitive gate, power consumption is greatly reduced, and there is no sensor poisoning.

[0037] 2. The palladium-plated sensitive gate of the present invention enables the field-effect transistor to operate in the subthreshold region, which can greatly improve the sensor sensitivity, and through the amplification effect of the field-effect transistor itself, the output signal is more stable.

[0038] 3. This invention eliminates the influence of different sensitive gates by designing a neural network algorithm, thereby achieving comprehensive detection of hydrogen concentration. Attached Figure Description

[0039] Figure 1 This is a front view schematic diagram of the field-effect transistor hydrogen sensor of the present invention.

[0040] Figure 2 This is a top view schematic diagram of the field-effect transistor hydrogen sensor of the present invention.

[0041] Figure 3 This is a schematic diagram of the four-element neural network structure of the present invention.

[0042] Figure 4 This is a graph showing the sensor's performance under different heating voltages when testing 2% hydrogen during the verification test of this invention. Detailed Implementation

[0043] To provide a detailed description of the technical solution, model, and effects of this invention, further explanation and illustration are provided below in conjunction with the accompanying drawings. It should be clarified that the examples provided are for illustrative purposes only, and the invention is not limited to these single examples.

[0044] like Figure 1 and Figure 2 As shown, this embodiment proposes a field-effect transistor hydrogen sensor, comprising:

[0045] Semiconductor substrate 101;

[0046] The source electrode 103 and drain electrode 102 are formed on the upper surface of the semiconductor substrate;

[0047] A channel is formed on the surface of the semiconductor substrate between the source electrode 103 and the drain electrode 102;

[0048] Medium layer 104 formed on the channel;

[0049] A sensitive gate 105 formed on dielectric layer 104; and

[0050] A filter layer 106 is formed on the sensitive gate 105;

[0051] In this arrangement, the source 103 and drain 102 on the semiconductor substrate 101 are arranged in the order of source 103, drain 102, and source 103.

[0052] Two channels of equal size are formed on the surface of the semiconductor substrate 101;

[0053] The dielectric layer 104 is arranged in a quad array on the two channels;

[0054] A heating resistor 107 is disposed on the lower surface of the semiconductor substrate 101;

[0055] During operation, a threshold voltage is applied to the sensitive gate 105 and the source 103. The source 103 is grounded, and a fixed voltage is applied to the drain 102. A fixed voltage is also applied across the heating resistor 107. When the sensitive gate 105 is exposed to a hydrogen atmosphere, its resistance changes, which changes the threshold voltage of the sensitive gate 105, thereby changing the current in the channel between the source 103 and the drain 102, thus achieving the sensing function.

[0056] The field-effect transistor hydrogen sensor is prepared by the following method:

[0057] (1) The source, drain and dielectric layer of the field-effect transistor hydrogen sensor are fabricated on a semiconductor substrate;

[0058] (2) Clean the surface of the dielectric layer;

[0059] (3) A sensitive gate region is formed by setting a mask in the dielectric layer, a palladium film is sputtered in the sensitive gate region, and the palladium film is grown in a controllable manner by magnetron sputtering, thereby forming a sensitive gate.

[0060] (4) A loose and porous SiO2 layer is sputtered on the sensitive gate surface using magnetron sputtering as a filter layer;

[0061] (5) Take a sheet of platinum metal as a heating resistor and fix it on the lower surface of the semiconductor substrate.

[0062] Furthermore, the equivalent resistance values ​​of the sensitive gate and the dielectric layer of the field-effect transistor hydrogen sensor are on the same order of magnitude. Voltage division between the sensitive gate and the dielectric layer is achieved by controlling the on-resistance of the dielectric layer and the resistance of the sensitive gate. The resistance of the sensitive gate and the threshold voltage are controlled by adjusting the thickness, pressure, and power of the magnetron sputtered palladium film, thereby controlling the resistance of the palladium film.

[0063] In other embodiments, different palladium alloy metal materials, such as PdAg, PdNi, and PdPt alloys, are used to prepare different sensitive gates. The concentration of hydrogen is detected by comprehensively calculating and learning the effects of different gates.

[0064] This embodiment also provides a method for hydrogen detection:

[0065] During detection, a field-effect transistor hydrogen sensor prepared by the aforementioned method is used to obtain an inductive signal;

[0066] When outputting hydrogen concentration, the sensing signal is input into the pre-trained sensing signal-hydrogen concentration output model to obtain the corresponding hydrogen concentration.

[0067] The training steps for the induction signal-hydrogen concentration output model are as follows:

[0068] (1) The current change rate between the source and drain of the field-effect transistor hydrogen sensor is used as the sensitivity of the sensor. A training dataset is obtained through a large number of tests to establish the correspondence between different concentrations of hydrogen and the sensor sensitivity. A four-element neural network structure is established, in which the sensitivity of each sensor represents a neuron to work, and the detected hydrogen concentration is used as the input value for the next step of calculation.

[0069] (2) Normalize the input values;

[0070] (3) Weight each input value, with the initial weights being randomly assigned;

[0071] (4) Calculate the residual of each layer: calculate the difference between the predicted output value and the expected value of the training dataset; if it meets the allowable error, it can be output; if the error is large, backpropagation is performed to calculate the residual of each layer.

[0072] Output layer → Hidden layer: Residual = -(Output value - Sample value) * Derivative of activation function;

[0073] Hidden layer → Hidden layer: Residual = (weighted sum of residuals of each node in the right layer) * derivative of activation function;

[0074] (5) Update weights: Update weights according to the error range using the error weighted derivative formula;

[0075] Input layer: Weight increase = Sigmoid function value of the current node * residual of the corresponding node in the right layer * learning rate;

[0076] Hidden layer: Weight increase = input value * residual of the corresponding node in the right layer * learning rate;

[0077] The weight increase of the offset value = the residual of the corresponding node in the right layer * the learning rate;

[0078] (6) Output results: Re-output the results and iterate the process multiple times. In each iteration, the entire training set is processed simultaneously, and finally the comprehensive hydrogen detection concentration is output.

[0079] The specific four-element neural network structure in this embodiment is as follows: Figure 3 As shown:

[0080] x0, x1, x2, x3 represent the input layer, a0, a1, a2, a3 represent the hidden layers, and the third layer represents the output layer; a i (j) =The i-th neuron in the j-th layer, θ j =Weight matrix from layer j to layer j+1;

[0081] (1)

[0082] (2)

[0083] (3)

[0084] (4)

[0085] (5)

[0086] The activation function is:

[0087] The second layer a can be calculated using the first layer and the parameter matrix θ. (2) The value is then passed through the second layer a. (2) and parameter vector θ (2) The final result is calculated through an iterative process that progresses step by step. Figure 3 The direction indicated by the middle arrow represents the forward propagation process. All input parameters are weighted and summed, and the resulting value is passed down layer by layer until the final output layer. The more layers and neurons in each layer, the more weight parameters are generated. After propagating forward to the nth layer, a prediction result is output. This prediction result is compared with the actual result. If they do not match, forward propagation is performed to adjust the weights. Specifically, the partial derivative of the error is calculated using the final output result, and this partial derivative is then weighted and summed with the previous hidden layers. This process continues layer by layer until the input layer (not included in the calculation). Finally, the weights are updated using the partial derivatives calculated at each node. By comprehensively analyzing the different input results, the hydrogen concentration is detected.

[0088] Verification Test

[0089] With fixed gate-source and drain-source voltages, the performance of the hydrogen sensor in this embodiment was tested under different heating voltages when 2% hydrogen was tested. The sensitivity was measured by the rate of change of the source-drain current. The results showed that the current change rate of the sensor was good when the heating voltage was 5V. When the heating voltage was further increased, the current change rate remained basically unchanged, indicating that the sensitivity of the sensor could be maintained at a high level.

[0090] The response time and recovery time (T) of the hydrogen sensor were tested when the heating voltage was 5V. 90 The results showed that as the heating voltage increased from 1V to 6V, the response time and recovery time gradually shortened, with the response time reaching 20s and the recovery time reaching 85s. Figure 4 As shown.

[0091] Based on the correlation between different hydrogen concentrations and sensor sensitivity, a training dataset of hydrogen concentrations was obtained through extensive testing. Taking 2% hydrogen as an example, the computational training of the neural network algorithm is illustrated as follows:

[0092] 1. After the gas enters, the changes in the four sensitive grids result in different sensitivity values, corresponding to different test concentrations. Taking 1.7%, 1.6%, 2.2%, and 2.5% as examples, the average value is calculated to be 2%. Using this as the standard value, normalization is performed to obtain the input values ​​x0, x1, x2, and x3 as -0.3, -0.4, 0.2, and 0.5, respectively.

[0093] 2. Randomly weight each input value. The weighting value is randomly selected and ranges from -1 to 1. Substitute it into formulas (1), (2), (3), and (4) to obtain the hidden layer node values ​​as 0.3, 0.1, -0.2, and 0.4, respectively. Then, map the obtained values ​​through the logsig function to obtain the values ​​of a0, a1, a2, and a3 as 0.574, 0.525, 0.450, and 0.599, respectively. Use the same method to calculate the output layer value as 0.108 using formula (5).

[0094] 3. Error Calculation Output: Since the sample value is the hydrogen concentration of 2%, and the difference between this and the average value is 0, the error is 0.108. 2 The result was 0.0017. Based on the acceptable error, the detected hydrogen concentration was 2.108%. If the error was too large, iteration could continue. The residuals from the output layer to the hidden layer were -0.01, and the residuals from the hidden layer were -0.0005, -0.0007, 0.0002, and 0.0005, respectively. The process was then repeated from the previous layer to revise the weights and adjust the learning rate. If the learning rate was too small, the training speed would be too slow; if the learning rate was too large, some data would be lost during training. A learning rate of 0.6 was set, and the process was repeated multiple times. The final output value was 0.0012, indicating a hydrogen concentration of 2.0012% and an error of 0.0012. 2 The result was 0.00000144, and the hydrogen concentration result was in high agreement with the actual value.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for detecting hydrogen, characterized in that: During detection, a field-effect transistor hydrogen sensor is used to obtain the induced signal; When outputting hydrogen concentration, the sensing signal is input into the pre-trained sensing signal-hydrogen concentration output model to obtain the corresponding hydrogen concentration. The field-effect transistor hydrogen sensor includes: Semiconductor substrate; source and drain electrodes formed on the upper surface of the semiconductor substrate; channel formed on the surface of the semiconductor substrate between the source and drain electrodes; dielectric layer formed on the channel; sensitive gate electrode formed on the dielectric layer; and filter layer formed on the sensitive gate electrode. In this process, the source and drain electrodes on the semiconductor substrate are arranged in the order of source, drain, and source; two channels of equal size are formed on the surface of the semiconductor substrate; and the dielectric layer is arranged in a four-array configuration on the two channels. During operation, a threshold voltage is applied to the sensitive gate and the source, the source is grounded, and a fixed voltage is applied to the drain. When the sensitive gate is exposed to a hydrogen atmosphere, its resistance changes, which changes the threshold voltage of the sensitive gate, thereby changing the current in the channel between the drain and the source, thus realizing the sensing function. The training steps for the induction signal-hydrogen concentration output model are as follows: (1) The current change rate between the source and drain of the field-effect transistor hydrogen sensor is used as the sensor sensitivity. A training dataset is obtained through a large number of tests to establish the correspondence between different hydrogen concentrations and sensor sensitivity. A four-element neural network structure is established, in which the sensitivity of each sensor represents a neuron to work, and the detected hydrogen concentration is used as the input value for the next calculation. (2) Normalize the input values; (3) Weight each input value, with the initial weights being randomly assigned; (4) Calculate the residual of each layer: calculate the difference between the predicted output value and the expected value of the training dataset; if it meets the allowable error, it can be output; if the error is large, backpropagation is performed to calculate the residual of each layer. Output layer → Hidden layer: Residual = -(Output value - Sample value) * Derivative of activation function; Hidden layer → Hidden layer: Residual = (weighted sum of residuals of each node in the right layer) * derivative of activation function; (5) Update weights: Update weights according to the error range using the error weighted derivative formula; Input layer: Weight increase = Sigmoid function value of the current node * residual of the corresponding node in the right layer * learning rate; Hidden layer: Weight increase = input value * residual of the corresponding node in the right layer * learning rate; The weight increase of the offset value = the residual of the corresponding node in the right layer * the learning rate; (6) Output results: Re-output the results and iterate the process multiple times. In each iteration, the entire training set is processed simultaneously, and finally the comprehensive hydrogen detection concentration is output.

2. The hydrogen detection method according to claim 1, characterized in that, A heating resistor is provided on the lower surface of the semiconductor substrate.

3. The hydrogen detection method according to claim 1 or 2, characterized in that, The fabrication method of the field-effect transistor hydrogen sensor includes: (1) The source, drain and dielectric layer of the field-effect transistor hydrogen sensor of claim 1 are fabricated on a semiconductor substrate; (2) Clean the surface of the dielectric layer; (3) A sensitive gate region is formed by setting a mask in the dielectric layer, a palladium film is sputtered in the sensitive gate region, and the palladium film is grown in a controllable manner by magnetron sputtering, thereby forming a sensitive gate. (4) A loose and porous SiO2 layer is sputtered on the sensitive gate surface as a filter layer using magnetron sputtering.

4. The hydrogen detection method according to claim 3, characterized in that: A sheet of platinum metal is used as a heating resistor and fixed to the lower surface of a semiconductor substrate.

5. The hydrogen detection method according to claim 4, characterized in that: The equivalent resistance values ​​of the sensitive gate and the dielectric layer are on the same order of magnitude. Voltage division between the sensitive gate and the dielectric layer is achieved by controlling the on-resistance of the dielectric layer and the resistance of the sensitive gate. The resistance of the sensitive gate and the threshold voltage are controlled by adjusting the thickness, pressure and power of the magnetron sputtered palladium film, thereby controlling the resistance of the palladium film.

6. The hydrogen detection method according to claim 5, characterized in that: Each sensitive gate is made of a different palladium alloy metal material to form a different sensitive gate.

Images