Storage and calculation integrated circuit

By introducing operational amplifiers and proportional amplifier modules into the integrated memory circuit, the problems of poor circuit flexibility and high power consumption in the prior art are solved, and high-precision matrix multiplication and addition operations are realized, suitable for non-volatile memory devices and volumes Accumulate neural network.

CN118708152BActive Publication Date: 2024-11-12TRUTH MEMORY CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411178757.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-12
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

In the high-precision matrix multiplication and addition operation, the existing memory and computing integrated circuit has problems such as poor circuit flexibility, high power consumption, high complexity and small switches of non-volatile memory devices, resulting in increased calculation accuracy and circuit complexity.

Method used

A memory-based computing integrated operation circuit is designed, including a memory array module, a reference array module, an operational amplifier and a proportional amplifier module. The memory array output signal and reference signal array are calculated through the operational amplifier, and the proportional amplifier module is used to weight it according to the operational cycle. Computing eliminates the low impact of memory array switches and implements high-precision calculations without additional circuitry.

Benefits of technology

It improves calculation accuracy and flexibility, reduces circuit complexity and power consumption, and realizes high-precision matrix multiplication and addition operations, which are suitable for MAC operations of peripheral circuits of nonvolatile memory devices and convolutional neural networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118708152B_ABST
    Figure CN118708152B_ABST
Patent Text Reader

Abstract

The present application provides a storage-computation integrated circuit, which relates to the field of integrated circuit technology, and includes: a storage array, a reference array, an operational amplifier, a proportional amplification module, and a storage module; the first input end of the operational amplifier is connected to the output end of the storage array, the second input end is connected to the output end of the reference array and the first end of the proportional amplification module, the output end is connected to the second end of the proportional amplification module, and the third end of the proportional amplification module is connected to the storage module. The operational amplifier is used to compare the output signal array generated by the storage array according to the input signal and the reference signal array generated by the reference array according to the input signal to obtain a comparison signal array. The proportional amplification module is used to weight the comparison signal array according to the operation cycle of the input signal to obtain an operation signal array. The storage module is used to store operation signal arrays of different operation cycles. The circuit structure is simple and the operation accuracy is high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of integrated circuit technology, and in particular to a storage-computation-in-one circuit. Background Art

[0002] With the rapid iteration of artificial intelligence technology, storage-computing technology, which combines data storage and processing, has become an important part of future computing architecture. Non-volatile memory devices have attracted much attention due to the non-volatility of their stored data during power outages.

[0003] At present, storage and computing are widely used in matrix multiplication and addition operations, and the accuracy requirements are gradually increasing. High-precision storage and computing solutions are mostly implemented in the form of digital storage and computing, which leads to problems such as poor circuit flexibility, high area and power consumption. Analog storage and computing usually completes data addition operations based on current operations, which leads to higher power consumption. In addition, non-volatile memory devices usually face the problem of relatively small switches, so it is necessary to introduce reference devices or reference signal comparison circuits to ensure the accuracy of operations; and in order to achieve high-precision multi-bit input data and weight data, additional auxiliary circuits need to be introduced for implementation, which increases the complexity of the circuit. Summary of the invention

[0004] The present application provides a storage-computation integrated circuit to provide a technical solution that can ensure calculation accuracy without the need for complex circuits.

[0005] The present application provides a storage-computation integrated circuit, which includes: a storage array module, a reference array module, an operational amplifier, a proportional amplification module and a data storage module.

[0006] The first input end of the operational amplifier is connected to the output end of the storage array module, the second input end is connected to the output end of the reference array module and the first end of the proportional amplification module, the output end of the operational amplifier is connected to the second end of the proportional amplification module, and the third end of the proportional amplification module is connected to the data storage module.

[0007] The operational amplifier is used to generate an output signal array based on the input signal according to the storage array module, and the reference array module is used to generate a comparison signal array based on the reference signal array generated by the input signal, wherein the storage array stores a plurality of first resistors characterized by different resistance states, and the output signal array is a signal array obtained by operating the input signal and the first resistors; the reference array module stores a plurality of second resistors corresponding one-to-one to the first resistors, and the reference signal array is a signal array obtained by operating the input signal and the second resistors.

[0008] The proportional amplification module is used to weight the comparison signal array with a corresponding proportion to obtain an operation signal array.

[0009] The data storage module is used to store the operation signal arrays of different operation cycles.

[0010] In an optional implementation, the proportional amplification module includes a proportional amplification unit and a proportional adjustment unit.

[0011] The first end of the proportional amplification unit is connected to the second input end of the operational amplifier, and the second end is connected to the output end of the operational amplifier, and is used to amplify the comparison signal array by a first corresponding proportion according to the operation cycle of the input signal to obtain an amplified signal array.

[0012] The first end of the ratio adjustment unit is connected to the output end of the operational amplifier, and the second end is connected to the storage module, and is used to adjust the amplified signal array to a second corresponding ratio according to the operation cycle of the input signal to obtain the operation signal array.

[0013] In an optional implementation, the proportional amplification unit includes a plurality of first branches and a first control subunit connected in parallel, and each of the first branches includes a third resistor and a first switch.

[0014] One end of the third resistor is connected to the second input end of the operational amplifier, and the other end is connected to the output end of the operational amplifier through the first switch. The first control subunit is connected to the control end of the first switch in the multiple first branches, and is used to select the target first branch among the multiple first branches according to the operation cycle of the input signal; the third resistor in the target first branch is used to determine the first corresponding ratio.

[0015] In an optional implementation, the circuit further includes a matching resistor, one end of the matching resistor is connected to the first input terminal of the operational amplifier, and the other end of the matching resistor is grounded.

[0016] The resistance value of the matching resistor is determined by the resistance value of the total equivalent resistor of the storage array module, the resistance value of the total equivalent resistor of the reference array module, and the resistance value of the equivalent resistor after the proportional amplification unit is enabled.

[0017] In an optional implementation, the ratio adjustment unit includes a plurality of second branches connected in parallel and a second control subunit, and each of the second branches includes a charging transistor and a second switch.

[0018] The first end of the charging transistor is used to connect to an external power supply, the second end is connected to the storage module through the second switch, and the third end is connected to the output end of the operational amplifier; the second control subunit is connected to the control end of the second switch in the multiple second branches, and is used to select the target second branch in the multiple second branches according to the operation cycle of the input signal.

[0019] The charging transistor in the target second branch is used to determine the second corresponding ratio and output a corresponding operation signal to the storage module.

[0020] In an optional embodiment, the data storage module includes a capacitor, a first end of the capacitor is connected to an end of the second switch away from the charging transistor, and a second end is grounded; the capacitor is used to store the operation signal array.

[0021] In an optional implementation, the circuit further includes an analog-to-digital conversion module, which is connected to the first end of the capacitor and is used to convert the voltage of the capacitor into a digital signal.

[0022] In an optional embodiment, the circuit also includes a digital-to-analog conversion module, which is connected to the input end of the storage array module and the input end of the reference array module, and is used to modulate the digital input signal into an analog voltage signal and input it to the storage array and the reference array, wherein the analog voltage signal is the input signal.

[0023] In an optional embodiment, the current signal in the comparison signal array is determined by the input signal, the equivalent resistance of the proportional amplification unit after being selected according to the current operation cycle, the row resistance value of the storage array module and the input signal in the current operation cycle, and the row resistance value of the reference array module and the input signal in the current operation cycle.

[0024] In an optional implementation, the current charging current of the data storage module is determined by the current signal in the comparison signal array and the equivalent transconductance after the proportional adjustment unit is selected in the current operation cycle.

[0025] In the technical solution provided by the present application, an operational amplifier is used to operate the output signal output by the storage array and the reference signal array output by the reference array, which can eliminate the influence of the low switching ratio of the storage device in the storage array on the calculation accuracy. In addition, according to the operation cycle of the input signal, the present application uses a proportional amplification module to calculate the amplification factor after the operational amplifier performs addition and subtraction operations, and can perform weighted data accumulation of different proportions on the input signal in different operation cycles, thereby improving the operation accuracy.

[0026] Furthermore, the present application utilizes an operational amplifier to implement the subtraction operation of the output signal array and the reference signal array. Compared with the prior art in which the analog storage and calculation integrated circuit uses an additional digital subtractor module or auxiliary circuit to implement the subtraction operation of the data column voltage value and the reference column voltage value, which introduces additional area overhead and power consumption overhead, the present application does not need to introduce additional modules or auxiliary circuits. Therefore, the present application can provide a technical solution that does not require complex circuits and can ensure calculation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0028] Figure 1 A schematic diagram of the structure of a storage-computation-in-one computing circuit provided in an embodiment of the present application;

[0029] Figure 2 A schematic diagram of the structure of another storage-computation-in-one computing circuit provided in an embodiment of the present application;

[0030] Figure 3 A schematic diagram of a specific structure of a storage-computation-in-one computing circuit provided in an embodiment of the present application;

[0031] Figure 4 A schematic diagram of the operation structure of a storage-computation-in-one operation circuit provided in an embodiment of the present application;

[0032] Figure 5 A schematic diagram of the operation structure of another storage-computation-in-one operation circuit provided in an embodiment of the present application.

[0033] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used herein in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. It is understood that the terms "first", "second", etc. used in the present application can be used to describe various information or data in this article, but these elements are not limited by these terms. These terms are only used to distinguish the first information from another information. For example, without departing from the scope of the present application, the first action information can be referred to as the second action information, and similarly, the second action information can be referred to as the first action information. Both the first action information and the second action information are action information, but they are not the same action information.

[0036] First, the terms involved in this application are explained:

[0037] Storage and computing integrated circuit: is an emerging computing architecture that aims to integrate data storage and computing functions into the same hardware unit, thereby reducing data transmission and improving computing efficiency and energy efficiency.

[0038] Storage array module: It is one of the core components in the Compute-in-Memory (CIM) architecture. It consists of a large number of storage units arranged in a matrix structure of rows and columns.

[0039] Reference array module: It consists of a set of predefined memory cells that store known reference data or states. Reference array modules are used to generate reference signals (such as reference voltages or currents) that are used for comparison and calibration during data operations.

[0040] Resistance state: refers to the properties of a material or device in different resistance states. These resistance states can be changed by applying different voltages or currents, thereby enabling data storage and processing.

[0041] With the rapid iteration of artificial intelligence technology, storage-computing technology, which combines data storage and processing, has become an important part of future computing architecture. Non-volatile memory devices have attracted much attention due to the non-volatility of their stored data during power outages.

[0042] At present, storage and computing are widely used in matrix multiplication and addition operations, and the accuracy requirements are gradually increasing. High-precision storage and computing solutions are mostly implemented in the form of digital storage and computing, which leads to problems such as poor circuit flexibility, high area and power consumption. Analog storage and computing usually completes data addition operations based on current operations, which leads to higher power consumption. In addition, non-volatile memory devices usually face the problem of relatively small switches, so it is necessary to introduce reference devices or reference signal comparison circuits to ensure the accuracy of operations; and in order to achieve high-precision multi-bit input data and weight data, additional auxiliary circuits need to be introduced for implementation, which increases the complexity of the circuit.

[0043] Based on this, an embodiment of the present application provides a storage and calculation integrated circuit, which uses operational amplifiers to implement multiplication and addition operations of input numbers and weight numbers, and simultaneously completes subtraction operations of the storage array and the reference array, and uses a proportional amplification module to perform weighted summation of data in different operation cycles of the input signal in different proportions, with low circuit overhead.

[0044] The storage-computation integrated circuit provided in the embodiment of the present application can be applied to the peripheral circuit of the non-volatile memory, and provides a high-precision computing solution without changing the storage array structure. It can also be based on this solution to form a complete non-volatile high-precision storage-computation integrated architecture, and serve as a dedicated computing unit to perform MAC (Multiply-Accumulate) operations in convolutional neural networks.

[0045] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0046] Reference Figure 1 The improved storage-computation integrated circuit of the embodiment of the present application includes: a storage array module 10, a reference array module 20, an operational amplifier 30, a proportional amplification module 40 and a data storage module 50.

[0047] The first input terminal of the operational amplifier 30 is connected to the output terminal of the storage array module 10, the second input terminal is connected to the output terminal of the reference array module 30 and the first terminal of the proportional amplification module 40, the output terminal is connected to the second terminal of the proportional amplification module 40, and the third terminal of the proportional amplification module 40 is connected to the data storage module.

[0048] The storage array module 10 is used to store data with different weights, and to generate a corresponding output signal array according to the input signal and the data with different weights.

[0049] Optionally, the storage array module 10 can implement the writing of weight data through the write bit lines and column source lines. After being written, different weight data can be stored in different storage cells of the storage array module 10.

[0050] Further, see Figure 3 The storage array module 10 includes a plurality of first resistors 101 characterized by different resistance states; the resistance states of the first resistors 101 can be used to represent different weight values. The output signal array can be a signal obtained by operating the input signal with the first resistors 101. Optionally, the storage array module 10 is a non-volatile memory device array.

[0051] In the embodiment of the present application, weight data can be pre-stored in the storage array module, and the resistance state of each first resistor represents a weight value. In the signal processing stage, the input signal interacts with the weight data to generate a corresponding output signal array.

[0052] In the calculation stage, the storage array module 10 generates an output signal array according to the operation of the input signal and the data with different weights.

[0053] The reference array module 20 is used to generate a reference signal according to an input signal.

[0054] Optionally, in the reference array module 20, the reference storage cells can be set to fixed resistance states. These fixed resistance states are used to generate stable reference signals. In the calculation phase, the input signal and the storage cells in the reference array module 20 jointly generate a reference signal array.

[0055] Further, see Figure 3 The reference array module 20 includes a plurality of second resistors 201 characterized by the same resistance state, and the plurality of second resistors 201 are arranged in one-to-one correspondence with the plurality of first resistors 101 . The reference signal array is a signal obtained by operating the input signal with the second resistors 201 .

[0056] The resistance states of the plurality of second resistors are fixed, and are used to generate a stable reference signal array. The reference signal array signal is used to calibrate the output signal array output by the storage array module to improve the accuracy of the calculation.

[0057] Based on this, the embodiment of the present application can realize the storage of data with different weights based on the storage array module, and can also obtain the output signal array according to the relationship between the input signal and the stored data, which can significantly improve the calculation efficiency. The reference array module is used to calibrate the output signal array, which can eliminate the influence of the relatively low switching ratio of the storage device in the storage array module on the calculation accuracy and improve the accuracy of the calculation.

[0058] The operational amplifier 30 has a first input terminal connected to the output terminal of the storage array module 10 and a second input terminal connected to the output terminal of the reference array module 20, and is used to compare the output signal array with the reference signal array to obtain a comparison signal array.

[0059] It is worth noting that within one operation cycle, the operational amplifier 30 determines the signal at the corresponding position in the comparison signal array according to the output signal array and the reference signal array, wherein the corresponding position in the comparison signal array can be determined by the operation cycle of the current input signal. It can also be understood that within one operation cycle, the output of the operational amplifier 30 is a signal, which can be a voltage signal. The output of the operational amplifier corresponding to multiple operation cycles can form the above-mentioned comparison signal array.

[0060] Optionally, the current signal in the comparison signal array is determined by the input signal, the equivalent resistance of the proportional amplification unit after being selected according to the current operation cycle, the row resistance value in the storage array module that operates with the input signal in the current operation cycle, and the row resistance value in the reference array module that operates with the input signal in the current operation cycle.

[0061] Based on this, the comparison signal array output by the operational amplifier 30 reflects the difference between the output signal array of the storage array module 10 and the reference signal of the reference array module 20. Through the reference signal array, the operational amplifier 30 can effectively calibrate the output signal array, and eliminate the influence of the relatively low switching ratio of the storage device in the storage array module 10 on the calculation accuracy, thereby improving the accuracy and reliability of the calculation.

[0062] The proportional amplification module 40 has a first end connected to the second input end of the operational amplifier 30 and a second end connected to the output end of the operational amplifier 30, and is used to weight the comparison signal array according to the operation cycle of the input signal to obtain an operation signal array. The data storage module 50 is connected to the third end of the proportional amplification module and is used to store signals of different operation cycles in the operation signal array.

[0063] It should be understood that within one operation cycle, the proportional amplification module 40 is used to weight the output result of the operational amplifier 30 corresponding to the current operation cycle in a corresponding proportion to determine the signal at the corresponding position in the operation signal array, wherein the corresponding position in the operation signal array can be determined by the operation cycle of the current input signal. It can also be understood that within one operation cycle, the output of the proportional amplification module 40 is a signal, which can be a current signal. The output of the proportional amplification module 40 corresponding to multiple operation cycles can form the above-mentioned operation signal array.

[0064] Optionally, the current input signal of the data storage module is determined by the current signal in the comparison signal array and the equivalent transconductance after the proportional adjustment unit is selected in the current operation cycle.

[0065] Exemplarily, when the data storage module is a capacitor, and the output signal of the comparison and amplification module is a current signal, the current charging current of the capacitor is the current output current of the comparison and amplification module.

[0066] The proportional amplification module 40 in the embodiment of the present application can weight the comparison signal array generated by the operational amplifier 30 in a corresponding proportion according to the operation cycle of the input signal, which can significantly improve the flexibility and accuracy of the calculation.

[0067] Based on the above description, the storage-computation integrated operation circuit provided in the embodiment of the present application uses an operational amplifier to operate on the output signal array output by the storage array module and the reference signal output by the reference array module, which can eliminate the influence of the relatively low switching ratio of the storage device in the storage array module on the calculation accuracy. And according to the operation cycle of the input signal, the present application uses a proportional amplification module to calculate the amplification factor after the operational amplifier performs addition and subtraction operations, and can perform weighted data accumulation of different proportions on the input signal in different operation cycles, thereby improving the calculation accuracy.

[0068] Furthermore, the present application utilizes operational amplifiers to implement subtraction operations between the output signal array and the reference signal array. Compared with the prior art analog storage and computing circuits that usually use additional digital subtractor modules or auxiliary circuits to implement subtraction operations between data column voltage values ​​and reference column voltage values, which introduces additional area overhead and power consumption overhead, the present application does not require the introduction of additional modules or auxiliary circuits.

[0069] The above part describes the overall structure and function of the storage and computing integrated circuit. The following is a detailed description of the structure and function of each part of the circuit.

[0070] First, the structure and function of the proportional amplification module are introduced in detail:

[0071] Reference Figure 1 , Figure 2 and Figure 3 The proportional amplification module 40 includes a proportional amplification unit 401 and a proportional adjustment unit 402 .

[0072] The first end of the proportional amplification unit 401 is connected to the second input end of the operational amplifier 30, and the second end is connected to the output end of the operational amplifier 30, and is used to amplify the comparison signal array by a first corresponding proportion according to the operation cycle of the input signal to obtain an amplified signal array.

[0073] The first end of the ratio adjustment unit 402 is connected to the output end of the operational amplifier 30, and the second end is connected to the data storage module 50, and is used to adjust the amplified signal array by a second corresponding ratio according to the operation cycle of the input signal to obtain the operation signal array.

[0074] Among them, the first end of the proportional amplification unit 401 is the first end of the proportional amplification module 40 , the first ends of the proportional amplification unit 401 and the proportional adjustment unit 402 are the second ends of the proportional amplification module 40 , and the second end of the proportional adjustment unit 402 is the third end of the proportional amplification module 40 .

[0075] Based on this, the embodiment of the present application amplifies and adjusts the comparison signal array in corresponding proportions according to the different operation cycles of the input signal. Therefore, the present application can dynamically adjust the comparison signal array according to the different operation cycles of the input signal to adapt to different application scenarios, while also achieving high-precision processing of the comparison signal array.

[0076] Reference Figure 3 As an example, the above-mentioned proportional amplification unit 401 includes a plurality of first branches 4011 connected in parallel and a first control subunit (not shown in the figure), each of the first branches 4011 includes a third resistor and a first switch; one end of the third resistor is connected to the second input end of the operational amplifier, and the other end is connected to the output end of the operational amplifier through the first switch. The first control subunit is connected to the control end of the first switch in the plurality of first branches 4011, and is used to select the target first branch in the plurality of first branches 4011 according to the operation cycle of the input signal.

[0077] In this embodiment, different first branches 4011 can be selected by the first control subunit to change the equivalent resistance of the proportional amplification unit 401, thereby achieving different proportional amplification of the comparison signal array to adapt to different input signals and different operation cycles of the input signals.

[0078] Optionally, the current signal of the comparison signal array satisfies:

[0079] ;

[0080] in R f is the equivalent resistance of the proportional amplification unit after being selected according to the current operation cycle; V in is the input signal; R data’ is the row resistance value in the storage array module that operates with the input signal in the current operation cycle; Rref’ is the row resistance value in the reference array of the nonvolatile device that operates with the input signal in the current operation cycle.

[0081] The above-mentioned first control subunit is used to control the on / off state of each first switch according to the operation cycle of the input signal. It can be a microcontroller or a digital signal processor, and the embodiment of the present application does not make any special restrictions on this. The first switch is used to control the on / off of the corresponding branch, and can be a MOSFET, a relay or other types of electronic switches. The embodiment of the present application does not make any special restrictions on this. The third resistor is used to determine the amplification ratio of the corresponding branch, and it can be a fixed resistor or a variable resistor. The embodiment of the present application does not make any special restrictions on this.

[0082] Optionally, the third resistor in the target first branch is used to determine the first corresponding ratio.

[0083] For example, assuming that there are three first branches connected in parallel, the third resistors of each branch are R3_1, R3_2, and R3_3, and the corresponding magnifications are A1, A2, and A3. The first control subunit selects the branch where R3_2 is located to be gated according to the operation cycle of the input signal, so as to amplify the input signal. Then the branch where R3_2 is located is the target branch, and A2 is the first corresponding ratio.

[0084] Optionally, the number of the first branches can be set according to the weighted number of the input signal. That is, the configuration of the proportional amplification unit can be dynamically adjusted according to the different weights of the input signal. Based on this, the number of the first branches can be set according to the weighted number of the input signal to achieve more sophisticated signal processing and amplification.

[0085] Reference Figure 3 As an example, the above-mentioned proportional adjustment unit 402 includes a plurality of parallel second branches 4021 and a second control subunit (not shown in the figure), each of the second branches 4021 includes a charging transistor and a second switch; the first end of the charging transistor is used to connect to an external power supply, the second end is connected to the data storage module 50 through the second switch, and the third end is connected to the output end of the operational amplifier 30; the second control subunit is connected to the control end of the second switch in the plurality of second branches 4021, and is used to select the target second branch in the plurality of second branches 4021 according to the operation cycle of the input signal.

[0086] Based on the above description, the embodiment of the present application can change the equivalent charging path of the ratio adjustment unit by selecting different second branches through the second control subunit, thereby achieving different ratio adjustments of the amplified signal array. Specifically, the second control subunit dynamically selects the target second branch according to the operation cycle of the input signal, thereby achieving real-time adjustment of the adjustment ratio. The charging transistor and the second switch in the selected target second branch determine the charging path and ratio adjustment result of the branch.

[0087] In some examples, the second control subunit may dynamically select one or more branches for gating according to the operation cycle of the input signal, thereby achieving different adjustment ratios for the amplified signal array.

[0088] The second control subunit may be a microcontroller or a digital signal processor (DSP), and the control subunit may also implement complex control logic through programming to meet different application requirements.

[0089] The charging transistor is used to control the flow of current and the charging process. It can be a MOSFET or other types of transistors, and the embodiments of the present application do not make specific limitations on this.

[0090] The second switch is used to control the on / off of the corresponding branch, and may be a MOSFET, a relay, or other types of electronic switches. The embodiment of the present application does not make any specific limitation on this.

[0091] The charging transistor in the target second branch is used to determine the second corresponding ratio and output the corresponding operation signal array to the data storage module.

[0092] As a specific example, the ratio adjustment unit includes three second branches connected in parallel, the charging transistors of each branch are Q1, Q2, and Q3, and the corresponding second switches are S1, S2, and S3. The second control subunit can select S1 according to the operation cycle of the input signal, then the second branch where S1 is located is the target second branch, and the corresponding transistor Q1 is used to determine the second corresponding ratio.

[0093] Optionally, at least some of the corresponding charging transistors in the plurality of second branches have different width-to-length ratios. It should be understood that different width-to-length ratios determine that the plurality of charging transistors have different adjustment ratios.

[0094] In an optional example, the number of the second branches can be set according to the weighted number of the input signal. That is, the configuration of the ratio adjustment unit can be dynamically adjusted according to the different weights of the input signal. Based on this, the number of the second branches can be set according to the weighted number of the input signal to achieve more sophisticated signal processing and amplification.

[0095] In some examples, the data storage module 50 may be a capacitor, a first end of which is connected to an end of the second switch away from the charging transistor, and a second end of which is grounded; the capacitor is used to store the operation signal array. The embodiment of the present application uses a capacitor to store the adjusted operation signal array, which can ensure the stability and reliability of the signal in the subsequent processing process.

[0096] The current charging current of the data storage module satisfies: .

[0097] in, It is the equivalent transconductance of the proportional adjustment unit after being selected in the current operation cycle.

[0098] It is worth noting that the charging current of the above data storage module is described by taking the charging current obtained after calculation of a row in the storage array module as an example.

[0099] Based on the above structure, refer to Figure 3 The storage-computing integrated circuit may further include: a digital-to-analog conversion module 60 , an analog-to-digital conversion module 70 and a matching resistor 80 .

[0100] The digital-to-analog conversion module 60 is connected to the input end of the storage array module 10 and the input end of the reference array module 20, and is used to modulate the digital input signal into an analog voltage signal, and input it into the storage array module 10 and the reference array module 20, wherein the analog voltage signal is the input signal. The modulation method of the digital-to-analog conversion module 60 includes but is not limited to amplitude modulation and width modulation.

[0101] The digital-to-analog conversion module 60 may be a DAC (digital-to-analog converter) for converting a digital input signal into an analog voltage signal. The DAC may be a high-precision DAC to ensure the accuracy of the converted analog voltage signal.

[0102] Optionally, the analog-to-digital conversion module 70 is connected to the data storage module 50 and is used to convert the calculation results stored in the data storage module 50 into digital signals for output.

[0103] The analog-to-digital conversion module 70 may be a high-precision ADC (Analog-to-Digital Converter), which is used to convert the operation result (analog signal) stored in the data storage module into a digital signal for output. The ADC may be a high-precision ADC to ensure the accuracy of the converted digital signal.

[0104] One end of the matching resistor 80 is connected to the first input end of the operational amplifier 30, and the other end is grounded. The matching resistor 80 is used to match the input impedance to ensure the accuracy and stability of signal processing. The resistance value of the matching resistor is determined by the total equivalent resistance value of the storage array module, the total equivalent resistance value of the reference array module, and the equivalent resistance value of the proportional amplification unit after being selected.

[0105] In some examples, the resistance of the matching resistor satisfies ; R’ is the resistance value of the matching resistor, R f is the equivalent resistance after the proportional amplification unit is selected according to the current operation cycle; R data is the total equivalent resistance value of the storage array module; R ref is the total equivalent resistance value of the reference array module.

[0106] That is, the resistance of the matching resistor 80 connected in parallel with the total equivalent resistance of the memory array module 10 is equal to the resistance of the total equivalent resistance of the reference array module 20 connected in parallel with the equivalent resistance of the proportional amplification unit 401 after being selected.

[0107] The following section uses a specific embodiment to provide a detailed introduction to the storage-computation-in-one computing circuit provided in this application.

[0108] Reference Figure 3 , the storage-computation-in-one computing circuit in the embodiment of the present application includes:

[0109] The digital-to-analog conversion module 60 is used to modulate the digital input and convert it into analog voltage and input it into the storage array module and the reference array module for calculation. The modulation method includes but is not limited to amplitude modulation and width modulation.

[0110] The storage array module 10 is a non-volatile memory device array, which is used to store different weight data values ​​and represent resistors in different resistance states for calculation.

[0111] The reference array module 20 is a non-volatile memory device reference array module, characterized by a fixed resistance state resistor, and is used to eliminate the influence of the low switching ratio of the non-volatile memory device on the calculation accuracy.

[0112] The operational amplifier 30 may be a high-gain, low-power operational amplifier, and is used to perform ideal addition and subtraction operations and generate a comparison signal array, which is a voltage signal, while also ensuring low power consumption.

[0113] The proportional amplification module (including the proportional amplification unit 401 and the proportional adjustment unit 402) is used to set the amplification factor after the operational amplifier 30 performs the operation, and can perform weighting of different proportions in different operation cycles. Specifically, the proportional amplification unit 401 is used to perform corresponding gating according to different calculation cycles, and is used to set the voltage amplification factor after the operational amplifier 30 performs addition and subtraction operations. The proportional adjustment unit 402 can modulate the current value charged to the capacitor in different operation cycles, and combined with the proportional amplification unit, it can realize multi-bit input and weight modulation.

[0114] The storage module 50 is used to store operation current results of different proportions corresponding to different operation cycles (corresponding to signals of different operation cycles in the operation signal array).

[0115] The analog-to-digital conversion module 70 converts the capacitor voltage signal after the operation to obtain a digital signal output. Specifically, it can be to convert the capacitor voltage signal after multiple cycles of accumulation operation to obtain a digital signal output.

[0116] In this embodiment, the modulated analog voltage signal is used as an input quantity and simultaneously input into the storage array module and the reference array module; through the configuration of the proportional amplification module, data accumulation with different proportional weights can be achieved, thereby improving the calculation accuracy.

[0117] Among them, the output result of the operational amplifier 30 is affected by the proportional amplification unit, and the two ends of the proportional amplification unit 401 are connected to the negative input and output ends of the operational amplifier 30; at the same time, the negative input end of the operational amplifier 30 is connected to the reference array module 20, the positive input end is connected to the storage array module 10, and the output end is connected to the proportional adjustment unit 402.

[0118] The proportional amplification unit is composed of multiple first branches, each of which can include a third resistor with a fixed value and a selection switch. The value of the third resistor is based on the resistance of the corresponding storage device in the storage array module and the ratio that needs to be adjusted in the current operation cycle, and is selected by the first control subunit according to the current operation cycle. The specific number of branches can be adjusted according to the required weighted number.

[0119] The ratio adjustment unit is composed of multiple second branches, each of which includes a charging transistor and a selection switch. The width-to-length ratio of the charging transistor is determined by the ratio to be adjusted and the amplification ratio of the ratio amplification unit, and is selected according to the current operation cycle by the control signal generated by the second control subunit. The specific number of branches can be adjusted according to the required weighted number.

[0120] Reference Figure 4 In a specific example, the output of the operational amplifier 30 is determined by the input analog voltage value , the storage array module 10, and the third resistor in the first branch selected in the proportional amplification unit 401. Taking a row in an array as an example (the total output can be obtained by the superposition theorem), the current signal of the comparison signal array output by the operational amplifier 30 satisfies:

[0121] ;

[0122] in R f is the equivalent resistance of the proportional amplification unit 401 after being selected according to the current operation cycle; V in is the input analog voltage value; R data’ The row resistance value in the memory array module that operates with the input signal in the current operation cycle may represent the weight of the stored data; R ref’ In the current operation cycle, the row resistance value (203) in the reference array module that operates with the input signal represents the data of the reference column.

[0123] Reference Figure 4 , the charging current of the capacitor is determined by the output of the operational amplifier The equivalent transconductance after the proportional adjustment unit 402 is selected Determine jointly, its value is .

[0124] The charging current ratio of each operation cycle is determined by the equivalent resistance after the proportional amplifier unit 401 is selected. R f The equivalent transconductance after the proportional adjustment unit 402 is selected , the input signal V in ; The row resistance value in the storage array module that operates with the input signal R data’ ; The row resistance value in the non-volatile device reference array that operates with the input signal R ref’ It is determined that this value changes according to the different branches that are selected.

[0125] Reference Figure 5 , the matching resistor 80 is used for resistance matching, one end of which is connected to the input of the operational amplifier 30 and the other end is grounded. Its size should satisfy the equation .in, R’ is the matching resistor, R f is the equivalent resistance after the proportional amplification unit 401 is selected; Rdata is the total equivalent resistance value of the non-volatile device storage array module; R ref The total equivalent resistance value of the reference array module for non-volatile devices;

[0126] Based on the above description, compared with the prior art, the embodiments of the present application can realize the multiplication and addition operations of input numbers and weight numbers through a single operational amplifier, and simultaneously complete the subtraction operations of the storage array module and the reference array module, and perform weighted summation of data of different cycles in different proportions through a proportional amplification unit and a proportional adjustment unit, and the circuit overhead is relatively small.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A storage and calculation integrated circuit, characterized in that: The circuit comprises: a storage array module, a reference array module, an operational amplifier, a proportional amplification module and a data storage module; The first input terminal of the operational amplifier is connected to the output terminal of the storage array module, the second input terminal is connected to the output terminal of the reference array module and the first terminal of the proportional amplification module, the output terminal of the operational amplifier is connected to the second terminal of the proportional amplification module, and the third terminal of the proportional amplification module is connected to the data storage module; The operational amplifier is used for generating an output signal array based on an input signal according to the storage array module, and the reference array module is used for generating a comparison signal array based on a reference signal array generated by the input signal, wherein the storage array module stores a plurality of first resistors characterized by different resistance states, and the output signal array is a signal array obtained by operating the input signal and the first resistor; the reference array module stores a plurality of second resistors corresponding to the first resistors one by one, and the reference signal array is a signal array obtained by operating the input signal and the second resistor; The proportional amplification module includes a proportional amplification unit and a proportional adjustment unit; The first end of the proportional amplification unit is connected to the second input end of the operational amplifier, and the second end is connected to the output end of the operational amplifier, and is used to amplify the comparison signal array by a first corresponding ratio according to the operation cycle of the input signal to obtain an amplified signal array; The first end of the ratio adjustment unit is connected to the output end of the operational amplifier, and the second end is connected to the storage module, and is used to adjust the amplified signal array by a second corresponding ratio according to the operation cycle of the input signal to obtain an operation signal array; The data storage module is used to store the operation signal arrays of different operation cycles.

2. The storage-computation-in-one computing circuit according to claim 1, characterized in that: The proportional amplification unit includes a plurality of first branches and a first control subunit connected in parallel, each of the first branches includes a third resistor and a first switch; One end of the third resistor is connected to the second input end of the operational amplifier, and the other end is connected to the output end of the operational amplifier through the first switch, and the first control subunit is connected to the control end of the first switch in the plurality of first branches, and is used to select a target first branch in the plurality of first branches according to the operation cycle of the input signal; The third resistor in the target first branch is used to determine the first corresponding ratio.

3. The storage-computation-in-one computing circuit according to claim 2, characterized in that: The circuit further comprises a matching resistor, one end of which is connected to the first input terminal of the operational amplifier and the other end of which is grounded; The resistance value of the matching resistor is determined by the resistance value of the total equivalent resistor of the storage array module, the resistance value of the total equivalent resistor of the reference array module, and the resistance value of the equivalent resistor after the proportional amplification unit is enabled.

4. The storage-computation-in-one computing circuit according to claim 1, characterized in that: The ratio adjustment unit includes a plurality of second branches connected in parallel and a second control subunit, each of the second branches includes a charging transistor and a second switch; The first end of the charging transistor is used to connect to an external power supply, the second end is connected to the storage module through the second switch, and the third end is connected to the output end of the operational amplifier; the second control subunit is connected to the control end of the second switch in the plurality of second branches, and is used to select a target second branch in the plurality of second branches according to the operation cycle of the input signal; The charging transistor in the target second branch is used to determine the second corresponding ratio and output a corresponding operation signal to the storage module.

5. The storage-computation-in-one computing circuit according to claim 4, characterized in that: The data storage module comprises a capacitor, a first end of the capacitor is connected to an end of the second switch away from the charging transistor, and a second end is grounded; the capacitor is used to store the operation signal array.

6. The storage-computation-in-one computing circuit according to claim 5, characterized in that: The circuit further comprises an analog-to-digital conversion module, which is connected to the first end of the capacitor and is used for converting the voltage of the capacitor into a digital signal.

7. The storage-computation-in-one computing circuit according to any one of claims 1 to 6, characterized in that: The circuit also includes a digital-to-analog conversion module, which is connected to the input end of the storage array module and the input end of the reference array module, and is used to modulate a digital input signal into an analog voltage signal and input it to the storage array and the reference array, wherein the analog voltage signal is the input signal.

8. The storage-computation-in-one computing circuit according to any one of claims 1 to 3, characterized in that: The current signal in the comparison signal array is determined by the input signal, the equivalent resistance of the proportional amplification unit after being selected according to the current operation cycle, the row resistance value of the storage array module and the input signal in the current operation cycle, and the row resistance value of the reference array module and the input signal in the current operation cycle.

9. The storage-computation-in-one computing circuit according to claim 8, characterized in that: The current charging current of the data storage module is determined by the current signal in the comparison signal array and the equivalent transconductance after the proportional adjustment unit is selected in the current operation cycle.