Resistor array-based sensing and calculation system and its control method

Abstract

The present invention discloses a resistor array-based sensing and calculation system and its control method. The sensing and calculation system comprises: a resistor array comprising a plurality of sensing resistors, each of which has one end connected to a first voltage; an acquisition node connected to the other end of each of the sensing resistors; and a data acquisition module connected to the acquisition node for acquiring the output of the resistor array. The present invention utilizes a resistor array to sense non-electrical quantity information, and a data acquisition module to acquire the output of the resistor array, thereby completing the acquisition and calculation of non-electrical quantity information. The sensing and calculation are integrated, reducing computational complexity, hardware resources, and power consumption. The present invention can be widely applied in the field of integrated circuit design technology.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated circuit design, and in particular to a resistor array-based sensing and calculation system and a control method thereof. Background Art

[0002] Sensor devices are widely used in fields such as IoT terminals and embedded scenarios. Generally, sensor devices are used to generate response outputs to the measured physical quantities. After certain processing, the response outputs are converted into digital electrical signals and transmitted to the processor as the data sensed by the sensor.

[0003] With the development of artificial intelligence, neural network algorithms are increasingly being used in IoT terminals. When paired with sensors, they enable intelligent detection of images, touch, and other functions. A typical processing method is to convert sensor output into digital electrical signals as input for neural network calculations. The neural network algorithm is then processed in a processor to generate recognition results. This process involves processing and transmitting large amounts of sensor output data. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a sensing and computing system based on a resistor array and a control method thereof.

[0005] In one aspect, an embodiment of the present invention provides a resistor array-based sensing and computing system, comprising:

[0006] A resistor array, comprising a plurality of sensing resistors, wherein one end of each of the sensing resistors is connected to a first voltage;

[0007] a collection node connected to the other end of each of the sensing resistors;

[0008] A data acquisition module is connected to the acquisition node and is used to obtain the output result of the resistor array.

[0009] Furthermore, the resistor array-based sensing and computing system further includes a resistor array control module, and the resistor array control module includes a plurality of switches;

[0010] One end of the switch is used to access a first voltage, and the other end of the switch unit is connected to one end of the sensing resistor.

[0011] Furthermore, the resistor array-based sensing and computing system includes a control module;

[0012] The control module is used to control the resistor array control module.

[0013] Furthermore, the resistor array-based sensing and calculation system includes a voltage generating module;

[0014] The voltage generating module is used to generate the first voltage.

[0015] Furthermore, the data acquisition module includes a transimpedance amplifier, a first analog-to-digital converter and a feedback resistor;

[0016] The positive input terminal of the transimpedance amplifier is grounded, the negative input terminal of the transimpedance amplifier is connected to the acquisition node, one end of the feedback resistor is connected to the negative input terminal of the transimpedance amplifier, and the other end of the feedback resistor is connected to the output terminal of the transimpedance amplifier;

[0017] The output end of the transimpedance amplifier is also connected to the first analog-to-digital converter.

[0018] Furthermore, the data acquisition module includes an acquisition resistor and a second analog-to-digital converter;

[0019] One end of the acquisition resistor is connected to the acquisition node, and the other end of the acquisition resistor is connected to the second analog-to-digital converter.

[0020] On the other hand, the present application also provides another control method, which is applied to the aforementioned resistor array-based sensing and computing system, and the control method includes the following steps:

[0021] Determine the first convolution window;

[0022] Dividing the resistor array into regions according to the convolution window to obtain a plurality of first resistor array regions;

[0023] The output results of the first resistor array regions are obtained in sequence.

[0024] On the other hand, the present application also provides another control method, which is applied to the aforementioned resistor array-based sensing and computing system, and the control method includes the following steps:

[0025] Determining a convolution unit, wherein the convolution unit includes a plurality of sensing resistors; using the convolution unit to divide the resistor array-based sensing and calculation system into regions to obtain a first resistor array;

[0026] Determine the second convolution window;

[0027] Dividing the first resistor array into regions according to a second convolution window to obtain a plurality of second resistor array regions;

[0028] The output results of several second resistor array regions are obtained simultaneously.

[0029] Furthermore, the step of determining the convolution unit includes the following steps:

[0030] Determine the side length of the convolution kernel and the moving step size of the convolution kernel;

[0031] The number of the sensing resistors in the convolution unit is determined according to the side length of the convolution kernel and the moving step length of the convolution kernel.

[0032] The beneficial effects of the present invention are: by utilizing a resistor array to sense non-electrical quantity information and a data acquisition module to acquire the output results of the resistor array, the acquisition and calculation of non-electrical quantity information are completed, sensing and calculation are integrated, the calculation complexity is reduced, and hardware resources and power consumption are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a module block diagram of a resistor array-based sensing and computing system according to an embodiment of the present invention;

[0034] Figure 2 It is a structural schematic diagram of the resistor array of the present invention;

[0035] Figure 3 is a schematic structural diagram of another resistor array of the present invention;

[0036] Figure 4 It is a schematic diagram of the serial convolution principle of the present invention;

[0037] Figure 5 is a distribution diagram of the second resistor array region of the present invention;

[0038] Figure 6 is a flow chart of steps of a control method of the present invention;

[0039] Figure 7 It is a flow chart of the steps of another control method of the present invention. DETAILED DESCRIPTION

[0040] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it should not be understood as a limitation on the scope of protection of the present invention.

[0041] In the present invention, if directions (up, down, left, right, front and back) are described, it is only for the convenience of describing the technical solution of the present invention, and does not indicate or imply that the technical features referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention.

[0042] In the present invention, "several" means one or more, "multiple" means more than two, "greater than," "less than," "exceeds," etc. are understood to exclude the number itself; "above," "below," "within," etc. are understood to include the number itself. In the description of the present invention, the use of "first" or "second" is solely for the purpose of distinguishing technical features and is not to be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0043] In the present invention, unless otherwise expressly defined, terms such as "disposed," "installed," and "connected" should be interpreted broadly. For example, they may refer to direct connection or indirect connection through an intermediate medium; fixed connection or detachable connection or integral molding; mechanical connection or electrical connection or mutual communication; and internal connection between two components or interaction between two components. Those skilled in the art can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0044] The present invention will be further explained and illustrated below with reference to the accompanying drawings and specific embodiments.

[0045] First, refer to Figure 1 and Figure 2 , an embodiment of the present invention provides a sensing and computing system based on a resistor array, comprising:

[0046] A resistor array includes a plurality of sensing resistors, wherein one end of each sensing resistor is connected to a first voltage;

[0047] A collection node M, where the collection node M is connected to the other end of each sensing resistor;

[0048] The data acquisition module is connected to the acquisition node M and is used to obtain the output result of the resistor array.

[0049] Specifically, the sensing and computing system of the present application has both sensing and computing functions.

[0050] Among them, the resistor array is used to sense non-electrical information. The resistor array includes a number of sensing resistors, which are used to sense or capture non-electrical information. The resistance of the sensing resistors in the resistor array changes with the change of non-electrical information in the environment. The types of sensing resistors include but are not limited to piezoresistors, photoresistors, thermistors, magnetoresistors, etc. When collecting the magnetic field strength information of the banknote, when the resistor array is close to the banknote (at this time, the resistor array includes a number of magnetoresistors), the magnetic field strength on the banknote will cause the resistance (conductance) of the sensing resistors in the resistor array to change. At this time, the resistance (conductance value) of the sensing resistors at each position in the resistor array can reflect the magnetic field strength information of the current sensing resistor location. By the same token, a resistor array composed of a number of piezoresistors can be used to measure the pressure information at the location of the piezoresistor, and a resistor array composed of a number of photoresistors can be used to measure the light intensity information at the current location of the photoresistor.

[0051] In one embodiment, referring to Figure 2 , taking the resistor array including 3*3 magnetoresistive resistors and using the sensing and computing system to sense and calculate the magnetic field strength in the environment as an example, the sensing and computing system based on the resistor array of the present application is explained.

[0052] The sensing and computing system is placed in an environment where sensing and computing are currently required, and the following formula is obtained:

[0053] G i =F(H i )

[0054] Where Hi is the magnetic field intensity at the magnetoresistor Ri at the i-th position in the resistor array, Gi is the conductance at the magnetoresistor at the i-th position in the resistor array due to the influence of the magnetic field intensity Hi, and F(.) is the functional relationship between magnetic field intensity and conductance, which is a monotonic function. Of course, the functional relationship between magnetic field and resistance can also be obtained. In this embodiment, i ranges from 1 to 9.

[0055] One end of each magnetoresistor is also connected to a first voltage Vi. The first voltage acts on the magnetoresistor Ri. Therefore, an induced current flows through each magnetoresistor. At this time, the current Ii passing through the magnetoresistor Ri at the i-th position is:

[0056] I i =G i ·V i

[0057] The data acquisition module is used to obtain the output result of the resistor array, that is, to obtain the total current at the acquisition node M.

[0058] The induced currents through the magnetoresistive resistors on each branch converge to the acquisition node M. According to Kirchhoff's law, the total current flowing through the node M collected by the data acquisition module is:

[0059]

[0060] Since one of the parameters in the product calculation in this application is the conductance value of the magnetoresistor, and the conductance value has a fitted functional relationship with the spatial magnetic field intensity induced by the magnetoresistor and the functional relationship is monotonic, the actual variable in the circuit is the spatial magnetic field value Hi, which obtains the corresponding conductance value Gi through the functional relationship. In the calculation of the convolution layer of the neural network, Gi can be used as the input parameter of the neural network dot product calculation.

[0061] To sum up, the present application utilizes a resistor array to sense non-electrical information and a data acquisition module to collect the output results of the resistor array, thereby completing the collection and calculation of non-electrical information, integrating sensing and calculation. Compared with sensors in the prior art, there is no need to convert the sensor output into a digital signal for identification and calculation, which reduces the computational complexity, hardware resources and power consumption.

[0062] As an optional embodiment, refer to Figure 2 The data acquisition module includes a transimpedance amplifier TIA, a first analog-to-digital converter (not shown in the figure) and a feedback resistor R F ;

[0063] The positive input terminal of the transimpedance amplifier TIA is grounded, the negative input terminal of the transimpedance amplifier TIA is connected to the acquisition node M, and the feedback resistor R F One end is connected to the negative input of the transimpedance amplifier TIA, and the feedback resistor R F The other end is connected to the output end of the transimpedance amplifier TIA;

[0064] The output terminal of the transimpedance amplifier TIA is also connected to the first analog-to-digital converter.

[0065] In one embodiment, the data acquisition module includes a transimpedance amplifier TIA, such as Figure 2 As shown, the positive input terminal of the transimpedance amplifier TIA is grounded, and the negative input terminal is connected to the acquisition node M, where R F is the feedback resistor, and the output voltage of the transimpedance amplifier TIA is Vout:

[0066] Vout=I·R F

[0067] By measuring the output voltage Vout of the transimpedance amplifier TIA and converting it into a digital value using the first analog-to-digital converter, the total current passing through the 9 magnetic sensitive resistors multiplied by the feedback resistor R can be obtained. FThe voltage result of the resistance value is divided by the feedback resistance R F The output result of the resistor array can be obtained by measuring the resistance value, which is the total current output by the resistor array.

[0068] As an optional embodiment, refer to Figure 3 The data acquisition module includes an acquisition resistor Rref and a second analog-to-digital converter ( Figure 3 not shown);

[0069] One end of the acquisition resistor Rref is connected to the acquisition node M, and the other end of the acquisition resistor Rref is connected to the second analog-to-digital converter.

[0070] Specifically, the present application also provides another embodiment of the data acquisition module. In this embodiment, the data acquisition module includes an acquisition resistor Rref and a second analog-to-digital converter, wherein the acquisition resistor Rref is a resistor with a small resistance value, and its resistance value is much smaller than the total resistance value of the resistor array. By obtaining the voltage across the acquisition resistor Rref and converting it into a digital quantity using the second analog-to-digital converter, the voltage result of the digital quantity is divided by the resistance value of the acquisition resistor Rref to obtain the output result of the resistor array, that is, the total current output by the resistor array.

[0071] As a further optional implementation, the resistor array-based sensing and computing system further includes a resistor array control module, and the resistor array control module includes a plurality of switches;

[0072] One end of the switch is used to access the first voltage, and the other end of the switch unit is connected to one end of the sensing resistor.

[0073] Specifically, refer to Figure 2 The present application also provides a resistor array control module, using the switch in the resistor array control module ( Figure 2 K1-K9 in FIG are switches) are used to control the on / off status of the power supply of each sensing resistor.

[0074] As a further optional embodiment, the resistor array-based sensing and computing system includes a control module;

[0075] The control module is used to control the resistor array control module.

[0076] Specifically, the control module is used to trigger a control signal to control the on and off status of the switches in the resistor array control module. The control module includes but is not limited to a processor, a single-chip microcomputer, or other circuit structures and devices that can generate a series of digital levels with controllable timing.

[0077] As an optional embodiment, refer to Figure 1 ,The sensing and computing system based on the resistor array includes a voltage generating module;

[0078] The voltage generating module is used to generate a first voltage.

[0079] Specifically, the voltage generating module is used to generate the input voltage of the sensing resistor, that is, the first voltage. The DAC circuit used by the voltage generating module has the characteristics of bipolar output and single-ended output, can generate positive voltage and negative voltage, and has high conversion accuracy (greater than or equal to 16 bits) and high frequency response bandwidth.

[0080] In the second aspect, the present application also provides a control method for the sensing and computing system based on the resistor array applied to the first aspect, using the control method to sense and compute non-electrical information on a single convolutional layer in a neural network, referring to Figure 6 , the control method includes the following steps S101-S103:

[0081] S101, determining a first convolution window;

[0082] S102, dividing the resistor array into regions according to the convolution window to obtain a plurality of first resistor array regions;

[0083] S103 , sequentially obtaining output results of the first resistor array region.

[0084] Specifically, the control method of this scheme can be applied to a single-layer convolution layer of a neural network. An example is given in which the input feature map size of the single-layer convolution layer is 5×5×1 (that is, the specification size of the resistor array is 5*5), there is no padding 0 on the periphery, the convolution kernel size is 3×3×1 (the convolution kernel determines the size of the first convolution window to be 3*3), and the convolution window moving step size is 2.

[0085] The first convolution window is used to divide the resistor array into regions, dividing the sensing resistors in the resistor array into a plurality of first resistor array regions. Figure 4 , including four first resistor array areas (dashed rounded boxes), namely Window1, window2, window3 and window4. The resistor array is divided into areas according to the first convolution window. Each of the obtained first resistor array areas includes 9 sensing resistors, namely, window1 includes D11, D12, D13, D21, D22, D23, D31, D32 and D33; window2 includes D13, D14, D15, D23, D24, D25, D33, D34 and D35; window3 includes D31, D32, D33, D41, D42, D43, D51, D52 and D53; window4 includes D33, D34, D35, D43, D33, D45, D53, D54 and D55.

[0086] Reference Figure 4 After the first resistor array area is divided, the first voltage (that is, the weight voltage) V11, V12, V13, V21, V22, V23, V31, V32 and V33 is added to each sensing resistor in window 1 to obtain the convolution result Vout1 of the first resistor array area of ​​window 1. In the same way, the convolution result Vout2 of the first resistor array area of ​​window 2, the convolution result Vout3 of the first resistor array of window 3, and the convolution result Vout4 of the first resistor array of window 4 can be obtained in sequence.

[0087] The control method of the second aspect is essentially a serial convolution scheme. In the first period at the beginning of the convolution, the first resistor array area determined by the determined convolution window is convolved to obtain an output result. In the second period, the convolution window is moved, and then the new first resistor array area within the convolution window is convolved to obtain the next output result. In this way, by moving the convolution window in each period, an output result of the first resistor array area is finally obtained in each period until the convolution stops.

[0088] In the third aspect, the present application also provides another control method for the sensing and computing system based on the resistor array applied to the first aspect, using the control method to sense and calculate the non-electrical information of the single convolution layer in the neural network, referring to Figure 7 , the control method includes the following steps S201-S204:

[0089] S201, determining a convolution unit, wherein the convolution unit includes a plurality of sensing resistors; using the convolution unit to divide the sensing and computing system based on the resistor array into regions to obtain a first resistor array;

[0090] S202, determining a second convolution window;

[0091] S203, dividing the first resistor array into regions according to the second convolution window to obtain a plurality of second resistor array regions;

[0092] S204 , simultaneously obtaining output results of several second resistor array regions.

[0093] Specifically, the control method of this scheme can also be applied to a single-layer convolution layer of a neural network. An example is given in which the input feature map size of the single-layer convolution layer is 5×5×1 (that is, the specification size of the resistor array is 5*5), there is no padding 0 on the periphery, the convolution kernel size is 3×3×1 (the convolution kernel determines the size of the second convolution window to be 3*3), and the convolution window moving step size is 2.

[0094] In the above serial convolution embodiment, the units in the first resistor array are sensing resistors, referring to Figure 5 In the parallel convolution scheme, the unit in the second resistor array area is a convolution unit, which includes several sensing resistors. For example, several adjacent sensing resistors can be divided into one convolution unit to obtain a first resistor array, and the unit in the first resistor array is a convolution unit.

[0095] The second convolution window is used to divide the first resistor array into regions, dividing the convolution units in the first resistor array into multiple convolution regions to obtain multiple second resistor array regions. Figure 5 , including four second resistor array areas (dashed rounded boxes), namely Window5, window6, window7 and window8. The first resistor array is divided into areas according to the second convolution window to obtain several first resistor array areas. Each first resistor array area includes 9 convolution units, namely window5 includes V11, V12, V13, V21, V22, V23, V31, V32 and V33, window6 includes V13, V14, V15, V23, V24, V25, V33, V34 and V35, window7 includes V31, V32, V33, V41, V42, V43, V51, V52 and V53, and window8 includes V33, V34, V35, V43, V33, V45, V53, V54 and V55.

[0096] After the second resistor array area is divided, the first voltage (that is, the weight voltage) is added to each sensing resistor of the first convolution unit in window5, the first voltage (that is, the weight voltage) is added to each sensing resistor of the second convolution unit in window6, the first voltage (that is, the weight voltage) is added to each sensing resistor of the third convolution unit in window7, and the first voltage (that is, the weight voltage) is added to each sensing resistor of the fourth convolution unit in window8. At the same time, the convolution result Vout5 of the second resistor array area of ​​window5, the convolution result Vout6 of the second resistor array area of ​​window6, the convolution result Vout7 of the second resistor array area of ​​window7, and the convolution result Vout8 of the second resistor array area of ​​window8 are obtained to complete the calculation of the entire convolution layer.

[0097] As an optional implementation, step S201 includes the following steps S2011-S2012:

[0098] S2011, determining the side length of the convolution kernel and the moving step length of the convolution kernel;

[0099] S2012. Determine the number of sensing resistors in the convolution unit content according to the side length of the convolution kernel and the moving step length of the convolution kernel.

[0100] Specifically, the present application also provides an embodiment for determining a convolution unit.

[0101] Similarly, take the input feature map of a single convolution layer as an example, the size of the feature map is 5×5×1, the periphery is not padded with 0, the convolution kernel size is 3×3×1, and the convolution window moving step is 2. The side length C of the convolution kernel is determined to be 3, and the moving step S of the convolution kernel is determined to be 2. According to the formula (C-S+1) 2 It can be calculated that the convolution unit includes at least 4 sensing resistors.

[0102] The control method provided in the third aspect is essentially a parallel convolution scheme. When the convolution starts, the convolution units in each second resistor array area are convolved at the same time, thereby obtaining the same number of output results as the second convolution windows.

[0103] This application applies the above-mentioned serial convolution scheme and parallel convolution scheme to the above-mentioned resistor array-based sensing and computing system. When performing non-electrical quantity perception and calculation, for example, collecting magnetic field strength information on a banknote, the magnetic field strength information on the banknote can be converted into the resistance (conductance) of the sensing resistors at different positions on the resistor array. This will not lose magnetic field information, but can also realize multiplication and addition calculations based on current neural networks, thereby realizing sensor perception-computing integration. There is no need to convert the sensor output into a digital signal before performing recognition calculations, which reduces computational complexity, hardware resources and power consumption.

[0104] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A sensor-computing system based on a resistor array, characterized in that: include: A resistor array, comprising a plurality of sensing resistors, wherein one end of each of the sensing resistors is connected to a first voltage; a collection node connected to the other end of each of the sensing resistors; A convolution unit, configured to divide the resistor array into regions; a data acquisition module connected to the acquisition node and configured to acquire an output result of the resistor array, the data acquisition module comprising a transimpedance amplifier, a first analog-to-digital converter, and a feedback resistor; The positive input terminal of the transimpedance amplifier is grounded, the negative input terminal of the transimpedance amplifier is connected to the acquisition node, one end of the feedback resistor is connected to the negative input terminal of the transimpedance amplifier, and the other end of the feedback resistor is connected to the output terminal of the transimpedance amplifier; The output end of the transimpedance amplifier is also connected to the first analog-to-digital converter.

2. The resistor array-based sensing and computing system according to claim 1, wherein: The resistor array-based sensing and calculation system further includes a resistor array control module, and the resistor array control module includes a plurality of switches; One end of the switch is used to access a first voltage, and the other end of the switch is connected to one end of the sensing resistor.

3. The resistor array-based sensing and computing system according to claim 2, wherein: The resistor array-based sensing and computing system includes a control module; The control module is used to control the resistor array control module.

4. The resistor array-based sensing and computing system according to claim 1, wherein: The resistor array-based sensing and calculation system includes a voltage generating module; The voltage generating module is used to generate the first voltage.

5. The resistor array-based sensing and computing system according to any one of claims 1 to 4, characterized in that: The data acquisition module includes an acquisition resistor and a second analog-to-digital converter; One end of the acquisition resistor is connected to the acquisition node, and the other end of the acquisition resistor is connected to the second analog-to-digital converter.

6. A control method, characterized in that: The resistor array-based sensing and computing system according to any one of claims 1 to 5 comprises the following steps: Determine the first convolution window; Dividing the resistor array into regions according to the convolution window to obtain a plurality of first resistor array regions; The output results of the first resistor array regions are obtained in sequence.

7. A control method, characterized in that: The resistor array-based sensing and computing system according to any one of claims 1 to 5 comprises the following steps: Determining a convolution unit, wherein the convolution unit includes a plurality of sensing resistors; using the convolution unit to divide the resistor array-based sensing and calculation system into regions to obtain a first resistor array; Determine the second convolution window; Dividing the first resistor array into regions according to a second convolution window to obtain a plurality of second resistor array regions; The output results of several second resistor array regions are obtained simultaneously.

8. A control method according to claim 7, characterized in that: The step of determining the convolution unit includes the following steps: Determine the side length of the convolution kernel and the moving step size of the convolution kernel; The number of the sensing resistors in the convolution unit is determined according to the side length of the convolution kernel and the moving step length of the convolution kernel.

Images