Fourier transform in-memory operation system and operation method thereof

By splitting the N-dimensional input data into one-dimensional data groups and using discrete Fourier transform operators to process it, the problems of slow Fourier transform operation speed and low efficiency in the prior art are solved, and different data dimensions are achieved efficiently Fourier transform operation.

CN118981593AActive Publication Date: 2024-11-19HUAZHONG UNIV OF SCI & TECH

Patent Information

Application Number
CN202411072004.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-19
Estimated Expiration
2044-08-06

AI Technical Summary

Technical Problem

When the existing in-memory computing system performs Fourier transformation, as the data dimension increases, the computing speed decreases, the computing efficiency is low, and it is difficult to be compatible with different data dimensions.

Method used

By splitting the N-dimensional input data into multiple one-dimensional data groups in each iteration operation, and signaling each one-dimensional data group using a discrete Fourier transform operator, output a current pulse signal, and then converting it into Convert to one-dimensional transform data set and reorganize until the N-dimensional Fourier transform result is obtained.

Benefits of technology

It reduces the computational complexity, improves the computing speed and computing efficiency of Fourier transform, and realizes efficient calculations in different data dimensions while maintaining low power consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981593A_ABST
    Figure CN118981593A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of in-memory computing, and particularly discloses a Fourier transform in-memory computing system and a computing method thereof. The system comprises a preprocessing module used for splitting N-dimensional input data into a plurality of one-dimensional data groups according to a splitting dimension corresponding to the N-dimensional input data in a current iterative operation, and outputting a corresponding voltage pulse signal according to each one-dimensional data group; the operation module is used for outputting a current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group by using a discrete Fourier transform operator; and the post-processing module is used for converting each group of current pulse signals into a corresponding one-dimensional conversion data group and recombining each one-dimensional conversion data group according to the splitting dimension to obtain N-dimensional recombined data of the current iterative operation. According to the invention, low power consumption can be maintained, and the operation speed and the operation efficiency of the in-memory operation system for realizing Fourier transform can be effectively improved at the same time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of in-memory computing technology, and more specifically, relates to a Fourier transform in-memory computing system and a computing method thereof. Background Art

[0002] As a digital signal analysis method, Fourier transform has been widely used in signal analysis, image processing, pattern recognition and other fields. From a mathematical point of view, Fourier transform is a transformation that can transform a signal in the time domain into a signal in the frequency domain. Discrete Fourier transform is used to process discrete signals, transforming discrete signals in the time domain into discrete signals in the frequency domain. Fourier transform can be used not only to process one-dimensional and two-dimensional sequences, but also to process higher-dimensional data, such as three-dimensional image data, time series of three-dimensional data (four-dimensional), etc.

[0003] In the traditional von Neumann computing architecture, the computing module and the storage module are separate. During the Fourier transform operation, data is continuously transmitted between the computing module and the storage module. When the dimension of the processed data increases, the amount of data to be processed will also increase. The calculation of large amounts of data will consume a lot of system power consumption, resulting in system data access delay problems.

[0004] In the prior art, by using a non-volatile memory in-memory computing system, the memory and computing integration can be achieved by utilizing its characteristics of being able to store information and perform computing functions, which can solve the problems of high power consumption and data access delay in the above-mentioned traditional data computing process to a certain extent. However, as the dimension of processed data increases, such as three dimensions or even higher, the computing speed of the existing in-memory computing system to realize Fourier transform will drop significantly due to the rapid increase in the amount of data computing, and the computing efficiency will be low. Summary of the invention

[0005] In view of the defects of the prior art, the purpose of this application is to efficiently realize Fourier transform of different data dimensions, aiming to solve the problems of slow operation speed, low operation efficiency and difficulty in compatibility with different data dimensions in the existing in-memory computing system for implementing Fourier transform operations.

[0006] To achieve the above objectives, in a first aspect, the present application provides a Fourier transform in-memory operation system, comprising:

[0007] A preprocessing module is used to split the N-dimensional input data into multiple one-dimensional data groups according to the splitting dimensions corresponding to the N-dimensional input data in the current iterative operation, and output a corresponding voltage pulse signal according to each of the one-dimensional data groups; the N-dimensional input data includes the original N-dimensional data input in the first iterative operation, or the N-dimensional reorganized data output by the post-processing module in the first N-1 iterative operations;

[0008] A calculation module, used for outputting a current pulse signal corresponding to each one-dimensional data group based on a voltage pulse signal corresponding to each one-dimensional data group using a discrete Fourier transform operator;

[0009] The post-processing module is used to convert each group of the current pulse signals into a corresponding one-dimensional conversion data group, and reorganize each of the one-dimensional conversion data groups according to the splitting dimension to obtain the N-dimensional reorganized data of the current iterative operation; use the N-dimensional reorganized data obtained when the current iterative operation is the N-th iterative operation as the N-dimensional Fourier transform result of the original N-dimensional data, and output the N-dimensional Fourier transform result.

[0010] Optionally, the operation module includes a signal output unit, a real data output unit and an imaginary data output unit of the current pulse signal;

[0011] The real data output unit is used to output the real data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group;

[0012] The imaginary data output unit is used to output the imaginary data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group;

[0013] The signal output unit is used to output the current pulse signal corresponding to each one-dimensional data group based on the real data and the imaginary data of the current pulse signal corresponding to each one-dimensional data group.

[0014] Optionally, the real data output unit and the imaginary data output unit both include a first discrete Fourier transform matrix subunit and a second discrete Fourier transform matrix subunit; the first discrete Fourier transform matrix subunit and the second discrete Fourier transform matrix subunit are both constructed based on a memristor array, and the conductance value of the memristor array is determined based on a discrete Fourier transform operator;

[0015] The first discrete Fourier transform matrix subunit is used to map the real part terms of the discrete Fourier transform operator; the second discrete Fourier transform matrix subunit is used to map the imaginary part terms of the discrete Fourier transform operator.

[0016] Optionally, the real data output unit is specifically used for:

[0017] Inputting the real part data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the first discrete Fourier transform matrix subunit, and outputting the first real part data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0018] Inputting the negative number of the imaginary data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the second discrete Fourier transform matrix subunit, and outputting the second real data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0019] The first real part data and the second real part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the real part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

[0020] Optionally, the imaginary data output unit is specifically used for:

[0021] Input the real part data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the second discrete Fourier transform matrix subunit, and output the first imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0022] Inputting the imaginary data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the first discrete Fourier transform matrix subunit, and outputting the second imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0023] The first imaginary part data and the second imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

[0024] Optionally, for any discrete Fourier transform matrix subunit of the first discrete Fourier transform matrix subunit and the second discrete Fourier transform matrix subunit, the discrete Fourier transform matrix subunit is used to:

[0025] The partial data of the voltage pulse signal corresponding to each of the one-dimensional data groups and the negative number of the partial data of each of the voltage pulse signals are input into the memristor array, so that the partial data of each of the voltage pulse signals and the negative number of the partial data of each of the voltage pulse signals are weightedly summed with the conductance value of the corresponding memristor in the memristor array to obtain the partial data of the current pulse signal corresponding to each of the one-dimensional data groups output by the memristor array; the partial data includes real data or imaginary data.

[0026] In a second aspect, the present application provides a computing method applied to any of the aforementioned Fourier transform in-memory computing systems, comprising:

[0027] In the current iterative operation, the N-dimensional input data is split into multiple one-dimensional data groups according to the splitting dimension corresponding to the N-dimensional input data, and a corresponding voltage pulse signal is output according to each one-dimensional data group; the N-dimensional input data is the original N-dimensional data input in the first iterative operation, or the N-dimensional reorganized data output by the post-processing module in the first N-1 iterative operations;

[0028] Using a discrete Fourier transform operator, based on the voltage pulse signal corresponding to each of the one-dimensional data groups, output a current pulse signal corresponding to each of the one-dimensional data groups;

[0029] Each group of the current pulse signals is converted into a corresponding one-dimensional conversion data group, and each of the one-dimensional conversion data groups is reorganized according to the splitting dimension to obtain the N-dimensional reorganized data of the current iterative operation; the N-dimensional reorganized data obtained when the current iterative operation is the N-th iterative operation is used as the N-dimensional Fourier transform result of the original N-dimensional data, and the N-dimensional Fourier transform result is output.

[0030] Optionally, the using a discrete Fourier transform operator to output a current pulse signal corresponding to each one-dimensional data group based on a voltage pulse signal corresponding to each one-dimensional data group comprises:

[0031] Based on the voltage pulse signal corresponding to each of the one-dimensional data groups, output the real part data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0032] Based on the voltage pulse signal corresponding to each of the one-dimensional data groups, outputting the imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0033] Based on the real data and the imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups, the current pulse signal corresponding to each of the one-dimensional data groups is output.

[0034] Optionally, outputting the real data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group includes:

[0035] Based on the real part data of the voltage pulse signal corresponding to each of the one-dimensional data groups, output the first real part data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0036] Outputting second real part data of the current pulse signal corresponding to each one-dimensional data group based on the negative number of the imaginary part data of the voltage pulse signal corresponding to each one-dimensional data group;

[0037] The first real part data and the second real part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the real part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

[0038] Optionally, outputting the imaginary data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group includes:

[0039] Based on the real data of the voltage pulse signal corresponding to each of the one-dimensional data groups, output the first imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0040] Based on the imaginary data of the voltage pulse signal corresponding to each of the one-dimensional data groups, output the second imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups;

[0041] The first imaginary part data and the second imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

[0042] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the prior art:

[0043] The present application provides a Fourier transform in-memory operation system and an operation method thereof, wherein the operation system comprises a preprocessing module, an operation module and a post-processing module, wherein the preprocessing module splits N-dimensional input data into multiple one-dimensional data groups according to corresponding splitting dimensions in each iterative operation, wherein the N-dimensional input data comprises the original N-dimensional data input at the beginning or the intermediate reorganized data output by the post-processing module in the first N-1 iterative operations, and outputs a corresponding voltage pulse signal according to each one-dimensional data group; the operation module further uses a discrete Fourier transform operator to perform signal processing on the voltage pulse signal corresponding to each one-dimensional data group The post-processing module converts each group of current pulse signals into a corresponding one-dimensional conversion data group, and reorganizes each one-dimensional conversion data group to obtain N-dimensional reorganized data, until the iterative operation reaches the final N-th iterative operation, and outputs the N-dimensional Fourier transform result of the original N-dimensional data. By splitting the N-dimensional discrete Fourier transform into multiple one-dimensional discrete Fourier transforms, the complexity of the operation is greatly reduced. While maintaining low power consumption, the operation speed and efficiency of the in-memory operation system to realize Fourier transform can be effectively improved, and Fourier transform operations of different data dimensions can also be efficiently realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is one of the structural schematic diagrams of the Fourier transform in-memory operation system provided in the embodiment of the present application;

[0045] Figure 2 This is one of the processing flow diagrams of the preprocessing module in the Fourier transform in-memory computing system provided in the embodiment of the present application;

[0046] Figure 3 It is a structural schematic diagram of a computing module in a Fourier transform in-memory computing system provided in an embodiment of the present application;

[0047] Figure 4 is a structural schematic diagram of a real part data output unit of a current pulse signal in a calculation module provided in an embodiment of the present application;

[0048] Figure 5 Schematic diagram of the structure of the DFT matrix subunit in the computing module provided in the embodiment of the present application;

[0049] Figure 6 It is a structural schematic diagram of the imaginary data output unit of the current pulse signal in the calculation module provided in the embodiment of the present application;

[0050] Figure 7 This is the second structural diagram of the Fourier transform in-memory operation system provided in the embodiment of the present application;

[0051] Figure 8 This is the second processing flow diagram of the preprocessing module in the Fourier transform in-memory computing system provided by the embodiment of the present application;

[0052] Fig. 9 It is a schematic diagram of the processing flow of a post-processing module in the Fourier transform in-memory computing system provided by an embodiment of the present application;

[0053] Fig.10 It is a flowchart of the operation method of the Fourier transform in-memory operation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] The terms "first" and "second" and the like in the specification and claims herein are used to distinguish different objects rather than to describe a specific order of the objects. For example, a first discrete Fourier transform matrix subunit and a second discrete Fourier transform matrix subunit and the like are used to distinguish different discrete Fourier transform matrix subunits rather than to describe a specific order of the discrete Fourier transform matrix subunits.

[0056] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0057] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0058] Figure 1 is one of the structural diagrams of the Fourier transform in-memory operation system provided in the embodiment of the present application, such as Figure 1 As shown, the system specifically includes: a pre-processing module 110 , a calculation module 120 and a post-processing module 130 .

[0059] The pre-processing module 110 is used to split the N-dimensional input data into multiple one-dimensional data groups according to the splitting dimension corresponding to the N-dimensional input data in the current iterative operation, and output the corresponding voltage pulse signal according to each one-dimensional data group; the N-dimensional input data includes the original N-dimensional data input in the first iterative operation, or the N-dimensional reorganized data output by the post-processing module 130 in the first N-1 iterative operations;

[0060] The operation module 120 is used to output a current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group output by the preprocessing module 110 by using a discrete Fourier transform (DFT) operator;

[0061] The post-processing module 130 is used to convert each current pulse signal output by the operation module 120 into a corresponding one-dimensional conversion data group, and reorganize each one-dimensional conversion data group according to the splitting dimension to obtain N-dimensional reorganized data of the current iterative operation; use the N-dimensional reorganized data obtained when the current iterative operation is the N-th iterative operation as the N-dimensional Fourier transform result of the original N-dimensional data, and output the N-dimensional Fourier transform result.

[0062] Specifically, the original N-dimensional data described in the embodiment of the present application refers to N-dimensional data to be processed by Fourier transform. For example, for three-dimensional data, the original N-dimensional data may be original three-dimensional image data. It should be noted that N is a positive integer.

[0063] The N-dimensional reorganized data described in the embodiment of the present application refers to N-dimensional data obtained by performing data reorganization processing on the input data through a post-processing module after one iterative operation.

[0064] The N-dimensional input data described in the embodiment of the present application includes the original N-dimensional data input in the first iterative operation, or the N-dimensional reorganized data output by the post-processing module in the first N-1 iterative operations.

[0065] That is to say, the input of the preprocessing module in the first iteration is the original N-dimensional data, and the input of the preprocessing module in the second iteration is the N-dimensional reorganized data output by the post-processing module in the first iteration. The input of the preprocessing module in the third iteration is the N-dimensional reorganized data output by the post-processing module in the second iteration; and so on, the input of the preprocessing module in the Nth iteration is the N-dimensional reorganized data output by the post-processing module in the N-1th iteration.

[0066] It should be noted that the input data is processed once by the pre-processing module, the operation module and the post-processing module in sequence, which is an iterative operation process.

[0067] The splitting dimension described in the embodiment of the present application refers to the dimension in the splitting direction when the preprocessing module splits the N-dimensional input data. For example, for three-dimensional data, it can be divided into dimensions in the three directions of the x-axis, y-axis and z-axis. Therefore, when splitting the three-dimensional data, you can choose to split from the dimensions in the x-axis, y-axis, or z-axis direction.

[0068] The one-dimensional conversion data group described in the embodiment of the present application is a one-dimensional data group obtained by the post-processing module performing numerical mapping on each group of current pulse signals according to the mapping relationship between the input data signal and the voltage pulse signal.

[0069] In an embodiment of the present application, after the original N-dimensional data is input into the system, in the first iterative operation, the original N-dimensional data is first received and processed by the preprocessing module, and the preprocessing module splits the N-dimensional input data into multiple one-dimensional data groups according to the splitting dimension corresponding to the N-dimensional input data.

[0070] like Figure 2 As shown, when the original N-dimensional data is three-dimensional data of size 5×5×5, in the first iterative operation, the preprocessing module can first split the three-dimensional data from the dimension in the x-axis direction, thereby obtaining corresponding 25 one-dimensional data groups of size 1*5.

[0071] Furthermore, the preprocessing module maps each value in each one-dimensional data group to a set of voltage pulse signals of corresponding size through linear conversion, and outputs the voltage pulse signal corresponding to each one-dimensional data group. The pulse width of these voltage pulses remains consistent, but the voltage size of the pulse maintains a linear transformation relationship with the value of the input data, and then these voltage pulse signals will be divided into 25 time series for input. It can be understood that each voltage pulse signal in each set of voltage pulse signals corresponds one-to-one to the value in each one-dimensional data group.

[0072] In an embodiment of the present application, the operation module is pre-constructed by a non-volatile memory and mapped with a DFT operator. Thus, the operation module can perform DFT operation on the voltage pulse signal corresponding to each one-dimensional data group output by the pre-processing module. Combined with the conductivity characteristics of the non-volatile memory, the voltage pulse signal corresponding to each one-dimensional data group can be converted into a corresponding current pulse signal, and the current pulse signal corresponding to each one-dimensional data group can be output to the post-processing module.

[0073] It should be noted that the length of each one-dimensional data group after the original N-dimensional input data is expanded is consistent with the data length of each row of the DFT operator used, so that it can be applied to process input data of various sizes and also helps to improve computing efficiency.

[0074] Furthermore, in an embodiment of the present application, the post-processing module converts each group of current pulse signals into a corresponding one-dimensional conversion data group according to the mapping relationship between the aforementioned N-dimensional input data and the voltage pulse signal, and reorganizes the data of each one-dimensional conversion data group according to the aforementioned splitting dimension to obtain N-dimensional reorganized data of the first iterative operation.

[0075] Next, in the embodiment of the present application, the second iteration operation is performed, and the preprocessing module receives and processes the N-dimensional reorganized data. The preprocessing module splits the N-dimensional reorganized data into multiple one-dimensional data groups according to the splitting dimensions corresponding to the input data. Figure 2 When the original N-dimensional data is three-dimensional data of size 5×5×5, the preprocessing module can continue to split the N-dimensional reorganized data from the dimension in the y-axis direction, thereby also obtaining corresponding 25 one-dimensional data groups of size 1*5.

[0076] Further, in the embodiment of the present application, the pre-processing module, the operation module and the post-processing module perform processing in the manner of the first iterative operation, which will not be described in detail here. Finally, the N-dimensional reorganized data of the second iterative operation can be obtained.

[0077] Next, in the embodiment of the present application, the third iteration operation is entered, and the preprocessing module receives and processes the N-dimensional reorganized data output by the second iteration operation. The preprocessing module splits the N-dimensional reorganized data into multiple one-dimensional data groups according to the splitting dimension corresponding to the input data. Figure 2 When the original N-dimensional data is three-dimensional data of size 5×5×5, the preprocessing module can continue to split the N-dimensional reorganized data from the dimension in the z-axis direction, thereby also obtaining corresponding 25 one-dimensional data groups of size 1*5.

[0078] Further, in the embodiment of the present application, the pre-processing module, the operation module and the post-processing module are processed in the manner of the first iterative operation described above, which will not be repeated here. Finally, the N-dimensional reorganized data of the third iterative operation can be obtained. If the original N-dimensional data is three-dimensional data, the N-dimensional reorganized data output by the third iterative operation is the Fourier transform result of the original three-dimensional data.

[0079] If the original N-dimensional data is higher-dimensional data, then by analogy, the system will continue to calculate in the above-mentioned iterative operation method until the final N-th iterative operation is performed, and the N-dimensional reorganized data of the N-th iterative operation is obtained, and the N-dimensional reorganized data is used as the N-dimensional Fourier transform result of the original N-dimensional data, and the N-dimensional Fourier transform result is output.

[0080] The Fourier transform in-memory operation system of the embodiment of the present application includes a preprocessing module, an operation module and a post-processing module. The preprocessing module splits the N-dimensional input data into multiple one-dimensional data groups according to the corresponding splitting dimension in each iterative operation, wherein the N-dimensional input data includes the original N-dimensional data input at the beginning or the intermediate reorganized data output by the post-processing module in the first N-1 iterative operations, and outputs a corresponding voltage pulse signal according to each one-dimensional data group; the operation module further uses the DFT operator to perform signal processing on the voltage pulse signal corresponding to each one-dimensional data group. , output the corresponding current pulse signal; the post-processing module converts each group of current pulse signals into a corresponding one-dimensional conversion data group, and reorganizes each one-dimensional conversion data group to obtain N-dimensional reorganized data, until the iterative operation is performed to the final N-th iterative operation, and the N-dimensional Fourier transform result of the original N-dimensional data is output. By splitting the N-dimensional DFT into multiple one-dimensional DFTs, the complexity of the operation is greatly reduced. While maintaining low power consumption, the operation speed and efficiency of the in-memory operation system to realize Fourier transform can be effectively improved, and Fourier transform operations of different data dimensions can also be efficiently realized.

[0081] Figure 3 is a schematic diagram of the structure of the operation module in the Fourier transform in-memory operation system provided by the embodiment of the present application, such as Figure 3As shown, the operation module 120 includes a signal output unit 1201, a real data output unit 1202 and an imaginary data output unit 1203 of a current pulse signal;

[0082] The real data output unit 1202 is used to output the real data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group;

[0083] The imaginary data output unit 1203 is used to output the imaginary data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group;

[0084] The signal output unit 1201 is used to output the current pulse signal corresponding to each one-dimensional data group based on the real data and the imaginary data of the current pulse signal corresponding to each one-dimensional data group.

[0085] Specifically, in an embodiment of the present application, the operation module includes a signal output unit, a real data output unit and an imaginary data output unit of the current pulse signal. After operation by the operation module, the real data output unit and the imaginary data output unit of the current pulse signal will respectively output the real part data and the imaginary part data, and the currents corresponding to the same output result will be added respectively, and finally the current pulse signal corresponding to each one-dimensional data group will be obtained.

[0086] Based on the content of the above embodiment, as an optional embodiment, the real data output unit and the imaginary data output unit of the current pulse signal both include a first DFT matrix subunit and a second DFT matrix subunit; the first DFT matrix subunit and the second DFT matrix subunit are both constructed based on a memristor array, and the conductance value of the memristor array is determined based on a DFT operator;

[0087] The first DFT matrix subunit is used to map the real part terms of the DFT operator; the second DFT matrix subunit is used to map the imaginary part terms of the DFT operator.

[0088] Specifically, the first DFT matrix subunit described in the embodiment of the present application includes a DFT matrix composed of a memristor array, which is used to map the real part terms of the DFT operator.

[0089] The second DFT matrix subunit described in the embodiment of the present application also includes a DFT matrix composed of a memristor array, which is used to map the imaginary part terms of the DFT operator.

[0090] In the embodiment of the present application, a memristor is introduced in the process of N-dimensional DFT operation. Memristors have the advantages of small area, low power consumption, high speed, and compatibility with CMOS process, and are a major candidate device for improving computing performance. At the same time, memristors have multi-order conductivity characteristics, which can be used to use memristor conductivity values ​​to store frequently used basic data matrices, so that Ohm's law and Kirchhoff's law can be used to easily realize matrix multiplication calculations and convert output voltage pulse signals into output current data.

[0091] It should be noted that the DFT operator contains real part terms and imaginary part terms. Since the memristor device can only map one value at a time, a memristor array can only map one numerical matrix. Therefore, the real DFT operator and the imaginary DFT operator need to be mapped in different DFT matrices respectively.

[0092] It should also be noted that, in the embodiments of the present application, the memristor array may adopt a 1T1R memristor array structure or a 1S1R memristor array structure.

[0093] In an embodiment of the present application, the real data output unit and the imaginary data output unit of the current pulse signal both include a first DFT matrix subunit and a second DFT matrix subunit. Specifically, the DFT operator is split into real part terms and imaginary part terms, and then mapped in the memristor array respectively, to form the first DFT matrix subunit and the second DFT matrix subunit. The voltage pulses of the real part and the imaginary part of the input are respectively sent to the DFT matrix subunits mapped with the real part and the imaginary part of the DFT operator.

[0094] In a specific embodiment of the present application, the original N-dimensional data is three-dimensional data of a size of 5*5*5, and the size of the memristor array is 10*5, that is, the size of the DFT matrix set in the DFT matrix subunit is 10*5.

[0095] The computing system of the embodiment of the present application uses memristors to construct computing modules, uses multiple memristor arrays for parallel processing, and uses a simple module structure to implement current addition, thereby reducing the multiple data transmission processes in the traditional N-dimensional Fourier transform. This method realizes low-power, low-latency N-dimensional Fourier transform operations.

[0096] In an embodiment of the present application, the DFT operator is split into a real part and a complex part, which are then mapped in a memristor array. The real part and the imaginary part of the input voltage pulse signal are respectively sent to the DFT matrix in the DFT matrix subunit to map the real part and the imaginary part of the DFT operator, and each DFT matrix is ​​composed of a memristor array. Since the calculation of the real part and the imaginary part of the voltage pulse signal requires the first DFT matrix subunit and the second DFT matrix subunit respectively, a total of four memristor arrays are required.

[0097] Since real numbers multiplying real numbers get positive real numbers, real numbers multiplying imaginary numbers get positive imaginary numbers, and imaginary numbers multiplying imaginary numbers get negative real numbers, therefore, the output currents representing positive real numbers and negative real numbers are added to the corresponding positions, and the output currents representing different positive and imaginary numbers are added to the corresponding positions, and the complex input of the voltage pulse signal is converted to the complex DFT operator by DFT.

[0098] Furthermore, in an embodiment of the present application, the real data output unit of the current pulse signal is used to perform calculations and conversions based on the real part of the voltage pulse signal corresponding to each one-dimensional data group, and finally output the real data of the current pulse signal corresponding to each one-dimensional data group.

[0099] Figure 4 is a structural diagram of a real part data output unit of a current pulse signal in a calculation module provided in an embodiment of the present application, such as Figure 4 As shown, the real data output unit 1202 includes a first DFT matrix subunit 1204 and a second DFT matrix subunit 1205. The real data output unit 1202 is specifically used for:

[0100] Input the real part data of the voltage pulse signal corresponding to each one-dimensional data group to the first DFT matrix subunit 1204, and output the first real part data of the current pulse signal corresponding to each one-dimensional data group;

[0101] Inputting the negative number of the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group into the second DFT matrix subunit 1205, and outputting the second real data of the current pulse signal corresponding to each one-dimensional data group;

[0102] The first real part data and the second real part data of the current pulse signal corresponding to each one-dimensional data group are superimposed, and the real part data of the current pulse signal corresponding to each one-dimensional data group is output.

[0103] Specifically, the first real data described in the embodiment of the present application refers to the real data of the voltage pulse signal corresponding to each one-dimensional data group input into the first DFT matrix sub-unit, and the partial real data item of the current pulse signal obtained after being processed by the memristor array in the DFT matrix sub-unit.

[0104] The second real data described in the embodiment of the present application refers to the negative number of the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group input into the second DFT matrix sub-unit, and the partial real data item of the current pulse signal is obtained after being processed by the memristor array in the DFT matrix sub-unit.

[0105] Based on the content of the above embodiment, as an optional embodiment, for any DFT matrix subunit of the first DFT matrix subunit and the second DFT matrix subunit, the DFT matrix subunit is used to:

[0106] The partial data of the voltage pulse signal corresponding to each one-dimensional data group and the negative number of the partial data of each voltage pulse signal are input into the memristor array, so that the partial data of each voltage pulse signal and the negative number of the partial data of each voltage pulse signal are weighted and summed with the conductance value of the corresponding memristor in the memristor array to obtain the partial data of the current pulse signal corresponding to each one-dimensional data group output by the memristor array; the partial data includes real data or imaginary data.

[0107] Specifically, the partial data described in the embodiment of the present application refers to the data item in the mathematical expression of the voltage pulse signal corresponding to the one-dimensional data group, which may specifically include the real part data V of the voltage pulse signal. real Or imaginary data V imag .

[0108] It should be noted that in the embodiment of the present application, the size of the memristor array is equal to twice the size of the DFT operator used, half of the DFT matrix inputs a positive voltage pulse signal, and half of the DFT matrix inputs a negative voltage pulse signal. The conductance value of the memristor array is set according to the size of the DFT operator, and there is a linear mapping relationship between the two, which can be set according to actual application requirements, and this application does not make specific limitations on this.

[0109] In addition, the conductance value of each memristor device in the half matrix to which the negative voltage pulse is input is set to the middle value between the maximum and minimum conductance values, thereby realizing the mapping of the Fourier operator containing positive and negative numbers.

[0110] In an embodiment of the present application, after obtaining the voltage pulse signal corresponding to each one-dimensional data group, the partial data of the voltage pulse signal corresponding to each one-dimensional data group and the negative number of the partial data of each voltage pulse signal are input into the memristor array, and the conductance value G of each memristor in the memristor array based on the DFT operator mapping, the partial data of each voltage pulse signal and the negative number of the partial data of each voltage pulse signal are weighted and summed with the conductance value of the corresponding memristor in the memristor array, so that the partial data of the current pulse signal corresponding to each one-dimensional data group can be obtained.

[0111] For example, the real data of the voltage pulse signal corresponding to each one-dimensional data group and the negative number of the real data of each voltage pulse signal are input into a memristor array that has a mapping relationship with the real part term of the DFT operator, that is, the first DFT matrix subunit. Then, the real data of each voltage pulse signal and the negative number of the real data of each voltage pulse signal are weightedly summed with the conductance value of the corresponding memristor in the memristor array, and the first real data of the corresponding current pulse signal can be obtained.

[0112] Figure 5 is a schematic diagram of the structure of the DFT matrix subunit in the operation module provided in the embodiment of the present application, such as Figure 5 As shown, in an embodiment of the present application, the size of the DFT matrix set in the DFT matrix subunit is 10*5, that is, the size of the memristor array in the DFT matrix is ​​also 10*5, and five positive voltage pulses and five corresponding negative voltage pulses are input simultaneously each time, corresponding to ten rows of the memristor array respectively. After the voltage passes through the memristor array, five columns of current are generated, corresponding to the output result of the matrix multiplication of the one-dimensional data group vector.

[0113] Specifically, assuming that the conductance values ​​of each memristor are G 11 , G 12 , …, G 55 , G′ 11 , G′ 12 , …, G′ 55 , the partial data V of the voltage pulse signal corresponding to the one-dimensional data group i (where i = 1, 2, ..., 5) and the corresponding negative signals -V i The five columns of partial data I corresponding to the current pulse signal can be obtained by inputting them into the DFT matrix respectively and performing weighted summation with the conductance values ​​of the corresponding memristors in the internal memristor array. m ,Right now:

[0114]

[0115] Among them, m=1,2,…,5.

[0116] The computing system of the embodiment of the present application constructs a DFT matrix sub-unit by introducing a memristor array structure, and utilizes newly added memristor devices with a smaller area. Half of the memristor array is designed to process positive voltage pulse signals, and half of the memristor array processes negative voltage pulse signals, thereby realizing the processing of negative values ​​of the DFT operator and ensuring the effectiveness and reliability of the DFT in-memory operation.

[0117] Further, in an embodiment of the present application, the real data output unit inputs the real data of the voltage pulse signal corresponding to each one-dimensional data group into the first DFT matrix subunit, and after the above-mentioned processing of the memristor array in the first DFT matrix subunit, the real data V of the voltage pulse signal corresponding to each one-dimensional data group is real and its corresponding negative signal -V real , which can be compared with the conductance value G of the corresponding memristor real Perform weighted summation and output the first real part data I of the current pulse signal corresponding to each one-dimensional data group real1 .

[0118] Similarly, the real data output unit outputs the negative number -V of the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group. imag Input to the second DFT matrix subunit, the negative number -V of the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group imag and its corresponding negative signal V imag , and the conductance value G of the memristor imag By performing weighted summation, the second real part data of the current pulse signal corresponding to each one-dimensional data group can be output -I real2 Furthermore, for each one-dimensional data group, the first real part data I of the current pulse signal corresponding to the one-dimensional data group real1 and the second real part data -I real2 Perform superposition processing and output the real part data I of the current pulse signal corresponding to each one-dimensional data group real =I real1 -I real2 .

[0119] The computing system of the embodiment of the present application utilizes two different DFT matrix sub-units to process the real data and imaginary data of the voltage pulse signal corresponding to each one-dimensional data group respectively, and implements the real number matrix multiplication operation in the DFT operation in blocks, thereby reducing the complexity of the computing circuit and improving the parallelism of the operation, further improving the effectiveness and reliability of the DFT in-memory operation.

[0120] Furthermore, in an embodiment of the present application, the imaginary data output unit performs calculations and conversions based on the voltage pulse signal corresponding to each one-dimensional data group, and finally outputs the imaginary data of the current pulse signal corresponding to each one-dimensional data group.

[0121] Figure 6 is a schematic diagram of the structure of the imaginary data output unit of the current pulse signal in the operation module provided in the embodiment of the present application, such as Figure 6 As shown, the imaginary data output unit 1203 includes a first DFT matrix subunit 1204 and a second DFT matrix subunit 1205, and the imaginary data output unit 1203 is specifically used for:

[0122] Input the real part data of the voltage pulse signal corresponding to each one-dimensional data group to the second DFT matrix subunit 1205, and output the first imaginary part data of the current pulse signal corresponding to each one-dimensional data group;

[0123] Inputting the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group into the first DFT matrix subunit 1204, and outputting the second imaginary data of the current pulse signal corresponding to each one-dimensional data group;

[0124] The first imaginary part data and the second imaginary part data of the current pulse signal corresponding to each one-dimensional data group are superimposed, and the imaginary part data of the current pulse signal corresponding to each one-dimensional data group is output.

[0125] Specifically, the first imaginary data described in the embodiment of the present application refers to the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group input into the second DFT matrix sub-unit, and the partial imaginary data item of the current pulse signal obtained after being processed by the memristor array in the DFT matrix sub-unit.

[0126] The second imaginary data described in the embodiment of the present application refers to the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group input into the first DFT matrix sub-unit, and the partial imaginary data items of the current pulse signal are obtained after being processed by the memristor array in the DFT matrix sub-unit.

[0127] In the embodiment of the present application, in combination with the processing principle of the memristor array in the above-mentioned DFT matrix sub-unit, the imaginary data output unit inputs the real data of the voltage pulse signal corresponding to each one-dimensional data group into the second DFT matrix sub-unit, and after being processed by the memristor array, the real data V of the voltage pulse signal corresponding to each one-dimensional data group is real and its corresponding negative signal -V real , which can be compared with the conductance value G of the corresponding memristor imag Perform weighted summation and output the first imaginary data I of the current pulse signal corresponding to each one-dimensional data group imag1 .

[0128] Similarly, the imaginary data output unit also inputs the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group into the first DFT matrix subunit, and after being processed by the memristor array, the imaginary data V of the voltage pulse signal corresponding to each one-dimensional data group is imag and its corresponding negative signal -V imag , which can be compared with the conductance value G of the corresponding memristor real Perform weighted summation and output the second imaginary data I of the current pulse signal corresponding to each one-dimensional data group imag2Furthermore, the first imaginary data I of the current pulse signal corresponding to each one-dimensional data group is imag1 and the second imaginary data I imag2 Perform superposition processing and output the imaginary data I of the current pulse signal corresponding to each one-dimensional data group imag =I imag1 +I imag2 .

[0129] The computing system of the embodiment of the present application utilizes two different DFT matrix sub-units to process the real data and imaginary data of the voltage pulse signal corresponding to each one-dimensional data group respectively, and implements the imaginary matrix multiplication operation in the DFT operation in blocks, thereby reducing the complexity of the computing circuit and improving the parallelism of the operation, further improving the effectiveness and reliability of the DFT in-memory operation.

[0130] Further, in an embodiment of the present application, after obtaining the real data and imaginary data of the current pulse signal corresponding to each one-dimensional data group, the signal output unit will generate a corresponding current pulse signal based on the real data and imaginary data of the current pulse signal corresponding to each one-dimensional data group, and output the current pulse signal corresponding to each one-dimensional data group.

[0131] The computing system of the embodiment of the present application splits the N-dimensional DFT operation into two parts, namely the real part and the narrative part. The output data processed by the computing module includes the real part and the imaginary part. The current pulse signals corresponding to the same output result will be added separately, and the DFT matrix multiplication operation of the complex data of the pulse signal is realized in blocks, thereby improving the parallelism and computing efficiency of the DFT in-memory operation.

[0132] Figure 7 FIG. 2 is a second structural diagram of the Fourier transform in-memory operation system provided in an embodiment of the present application. Figure 7As shown, in an embodiment of the present application, taking the DFT operation of three-dimensional data as an example, it includes three pre-processing processes, three DFT operation processes and three post-processing processes. Among them, the input is an original three-dimensional data of size 5×5×5, which is processed by the pre-processing module to obtain 25 one-dimensional data groups of size 5×1, each one-dimensional data group constitutes a vector, thereby forming a one-dimensional vector group; these vectors are then mapped to voltage pulse signals of corresponding sizes, and used as inputs of the operation module. The operation module includes a DFT matrix subunit containing 4 DFT matrices of size 10×5. After these input voltage pulse signals pass through the DFT matrix subunit constructed by the memristor array in the operation module, the corresponding current pulse signals are output. Furthermore, the output current pulse signal is input into the post-processing module for post-processing, and the output current pulse signal is converted into the actual corresponding one-dimensional vector according to the mapping relationship between the input signal and the input voltage pulse, and then restored to a three-dimensional reconstructed data of size 5×5×5.

[0133] The three-dimensional reconstructed data outputted for the first two times are data matrices in the intermediate state, which need to be input into the preprocessing module again for preprocessing. The three-dimensional reconstructed data outputted for the third time are the three-dimensional output data in the frequency domain corresponding to the original three-dimensional data in the time domain, which is the three-dimensional DFT result.

[0134] In the embodiment of the present application, the N-dimensional DFT is split into multiple one-dimensional DFTs, which greatly reduces the complexity of the operation. At the same time, the proposed splitting and mapping method is applicable to DFT operations of different dimensions and has universal applicability.

[0135] Figure 8 FIG. 2 is a second schematic diagram of a processing flow of a preprocessing module in a Fourier transform in-memory computing system provided in an embodiment of the present application. Figure 8As shown, in an embodiment of the present application, the preprocessing module splits the input original three-dimensional data of size 5×5×5 into 25 one-dimensional data groups of size 5×1, and the corresponding labels of each one-dimensional data group are {[(1,1),(1,2),(1,3),(1,4),(1,5)],[(2,1),(2,2),(2,3),(2,4),(2,5)],…,[(25,1),(25,2),(25,3),(25,4),(25,5)]}, and then, through linear conversion, the numerical values ​​in each one-dimensional data group are mapped to voltage pulse signals of corresponding sizes, so as to obtain 25 groups of voltage pulse signals, each group of voltage pulse signals consists of 5 voltage pulse signals, and these 25 groups of voltage pulse signals are output to the operation module for processing. Here, the labels corresponding to the 25 groups of voltage pulse signals are {[Vin(1,1), Vin(1,2), Vin(1,3), Vin(1,4), Vin(1,5)],…,[Vin(25,1), Vin(25,2), Vin(25,3), Vin(25,4), Vin(25,5)]}, where Vin(1,1), Vin(1,2), Vin(1,3), Vin(1,4), Vin(1,5) respectively represent the 5 voltage pulse signals in the first group of voltage pulse signals.

[0136] Fig. 9 is a schematic diagram of the processing flow of the post-processing module in the Fourier transform in-memory computing system provided by the embodiment of the present application, such as Fig. 9 As shown, in the embodiment of the present application, the post-processing module processes 25 groups of current pulse signals output by the operation module, wherein the labels corresponding to the 25 groups of current pulse signals are {[Iout(1,1), Iout(1,2), Iout(1,3), Iout(1,4), Iout(1,5)], …, [Iout(25,1), Iout(25,2), Iout(25,3), Iout(25,4), Iout(25,5)]}; through linear The 25 groups of current pulse signals are converted into one-dimensional data groups of corresponding sizes, and their corresponding labels are {[(1,1),(1,2),(1,3),(1,4),(1,5)],[(2,1),(2,2),(2,3),(2,4),(2,5)],…,[(25,1),(25,2),(25,3),(25,4),(25,5)]}, and then reversely reorganized according to the split dimensions to output three-dimensional reorganized data of size 5×5×5.

[0137] The following describes the operation method of the Fourier transform in-memory operation system provided by the present invention. The operation method of the Fourier transform in-memory operation system described below and the Fourier transform in-memory operation system described above can be referred to each other.

[0138] Fig.10 is a flow chart of a method for computing a Fourier transform in-memory computing system provided in an embodiment of the present application. The method can be applied to any of the aforementioned Fourier transform in-memory computing systems, such as Fig.10 As shown, the method includes: step 1010, step 1020 and step 1030.

[0139] Wherein, step 1010, in the current iterative operation, according to the splitting dimension corresponding to the N-dimensional input data, the N-dimensional input data is split into multiple one-dimensional data groups, and the corresponding voltage pulse signal is output according to each one-dimensional data group; wherein, the N-dimensional input data in the first iterative operation is the original N-dimensional data to be processed; in other iterative operations except the first iterative operation, the N-dimensional input data is the N-dimensional reorganized data output by the post-processing module;

[0140] Step 1020, using a DFT operator to output a current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group;

[0141] Step 1030, convert each group of current pulse signals into a corresponding one-dimensional conversion data group, and reorganize each one-dimensional conversion data group according to the split dimension to obtain N-dimensional reorganized data of the current iterative operation; use the N-dimensional reorganized data obtained when the current iterative operation is the N-th iterative operation as the N-dimensional Fourier transform result of the original N-dimensional data, and output the N-dimensional Fourier transform result.

[0142] It should be understood that the method in the above embodiment is applied to the aforementioned Fourier transform in-memory operation system, and its implementation principle and technical effect are similar to those described in the above system. The detailed process of the method can refer to the corresponding description process in the above system, which will not be repeated here.

[0143] The operation method of the Fourier transform in-memory operation system of the embodiment of the present application is as follows: in each iterative operation, the preprocessing module splits the N-dimensional input data into multiple one-dimensional data groups according to the corresponding splitting dimension, wherein the N-dimensional input data includes the original N-dimensional data input at the beginning or the intermediate reorganized data output by the post-processing module in the first N-1 iterative operations, and outputs the corresponding voltage pulse signal according to each one-dimensional data group; the operation module further uses the DFT operator to perform signal processing on the voltage pulse signal corresponding to each one-dimensional data group, and outputs the corresponding current pulse signal; the post-processing module converts each group of current pulse signals into a corresponding one-dimensional conversion data group, and reorganizes each one-dimensional conversion data group to obtain N-dimensional reorganized data, until the iterative operation is performed to the last Nth iterative operation, and the N-dimensional Fourier transform result of the original N-dimensional data is output. By splitting the N-dimensional DFT into multiple one-dimensional DFTs, the complexity of the operation is greatly reduced, and the operation speed and efficiency of the in-memory operation system to realize Fourier transform can be effectively improved while maintaining low power consumption, and Fourier transform operations of different data dimensions can also be efficiently realized.

[0144] Based on the content of the above embodiment, as an optional embodiment, step 1020, using a DFT operator, based on the voltage pulse signal corresponding to each one-dimensional data group, outputs the current pulse signal corresponding to each one-dimensional data group, including:

[0145] Based on the voltage pulse signal corresponding to each one-dimensional data group, output the real part data of the current pulse signal corresponding to each one-dimensional data group;

[0146] Based on the voltage pulse signal corresponding to each one-dimensional data group, output the imaginary part data of the current pulse signal corresponding to each one-dimensional data group;

[0147] Based on the real data and imaginary data of the current pulse signal corresponding to each one-dimensional data group, the current pulse signal corresponding to each one-dimensional data group is output.

[0148] Based on the content of the above embodiment, as an optional embodiment, based on the voltage pulse signal corresponding to each one-dimensional data group, outputting the real part data of the current pulse signal corresponding to each one-dimensional data group includes:

[0149] Based on the real part data of the voltage pulse signal corresponding to each one-dimensional data group, output the first real part data of the current pulse signal corresponding to each one-dimensional data group;

[0150] Outputting second real part data of the current pulse signal corresponding to each one-dimensional data group based on the negative number of the imaginary part data of the voltage pulse signal corresponding to each one-dimensional data group;

[0151] The first real part data and the second real part data of the current pulse signal corresponding to each one-dimensional data group are superimposed, and the real part data of the current pulse signal corresponding to each one-dimensional data group is output.

[0152] Based on the content of the above embodiment, as an optional embodiment, based on the voltage pulse signal corresponding to each one-dimensional data group, outputting the imaginary part data of the current pulse signal corresponding to each one-dimensional data group includes:

[0153] Based on the real data of the voltage pulse signal corresponding to each one-dimensional data group, output the first imaginary data of the current pulse signal corresponding to each one-dimensional data group;

[0154] Based on the imaginary data of the voltage pulse signal corresponding to each one-dimensional data group, output the second imaginary data of the current pulse signal corresponding to each one-dimensional data group;

[0155] The first imaginary part data and the second imaginary part data of the current pulse signal corresponding to each one-dimensional data group are superimposed, and the imaginary part data of the current pulse signal corresponding to each one-dimensional data group is output.

[0156] It should be understood that expressions such as "including" and "may include" that may be used in the present application indicate the existence of the disclosed functions, operations, or constituent elements, and do not limit one or more additional functions, operations, and constituent elements. In the present application, terms such as "including" and / or "having" may be interpreted as indicating specific characteristics, numbers, operations, constituent elements, components, or combinations thereof, but may not be interpreted as excluding the existence or possibility of adding one or more other characteristics, numbers, operations, constituent elements, components, or combinations thereof.

[0157] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A Fourier transform in-memory computing system, characterized in that: include: A preprocessing module is used to split the N-dimensional input data into multiple one-dimensional data groups according to the splitting dimensions corresponding to the N-dimensional input data in the current iterative operation, and output a corresponding voltage pulse signal according to each of the one-dimensional data groups; the N-dimensional input data includes the original N-dimensional data input in the first iterative operation, or the N-dimensional reorganized data output by the post-processing module in the first N-1 iterative operations; A calculation module, used for outputting a current pulse signal corresponding to each one-dimensional data group based on a voltage pulse signal corresponding to each one-dimensional data group using a discrete Fourier transform operator; The post-processing module is used to convert each group of the current pulse signals into a corresponding one-dimensional conversion data group, and reorganize each of the one-dimensional conversion data groups according to the splitting dimension to obtain the N-dimensional reorganized data of the current iterative operation; use the N-dimensional reorganized data obtained when the current iterative operation is the N-th iterative operation as the N-dimensional Fourier transform result of the original N-dimensional data, and output the N-dimensional Fourier transform result.

2. The Fourier transform in-memory operation system according to claim 1, characterized in that: The operation module includes a signal output unit, a real data output unit and an imaginary data output unit of the current pulse signal; The real data output unit is used to output the real data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group; The imaginary data output unit is used to output the imaginary data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group; The signal output unit is used to output the current pulse signal corresponding to each one-dimensional data group based on the real data and the imaginary data of the current pulse signal corresponding to each one-dimensional data group.

3. The Fourier transform in-memory operation system according to claim 2, characterized in that: The real data output unit and the imaginary data output unit both include a first discrete Fourier transform matrix subunit and a second discrete Fourier transform matrix subunit; the first discrete Fourier transform matrix subunit and the second discrete Fourier transform matrix subunit are both constructed based on a memristor array, and the conductance value of the memristor array is determined based on a discrete Fourier transform operator; The first discrete Fourier transform matrix subunit is used to map the real part terms of the discrete Fourier transform operator; The second discrete Fourier transform matrix subunit is used to map the imaginary part terms of the discrete Fourier transform operator.

4. The Fourier transform in-memory operation system according to claim 3, characterized in that: The real data output unit is specifically used for: Inputting the real part data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the first discrete Fourier transform matrix subunit, and outputting the first real part data of the current pulse signal corresponding to each of the one-dimensional data groups; Inputting the negative number of the imaginary data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the second discrete Fourier transform matrix subunit, and outputting the second real data of the current pulse signal corresponding to each of the one-dimensional data groups; The first real part data and the second real part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the real part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

5. The Fourier transform in-memory operation system according to claim 3, characterized in that: The imaginary data output unit is specifically used for: Input the real part data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the second discrete Fourier transform matrix subunit, and output the first imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups; Inputting the imaginary data of the voltage pulse signal corresponding to each of the one-dimensional data groups into the first discrete Fourier transform matrix subunit, and outputting the second imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups; The first imaginary part data and the second imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

6. The Fourier transform in-memory operation system according to any one of claims 3 to 5, characterized in that: For any discrete Fourier transform matrix subunit of the first discrete Fourier transform matrix subunit and the second discrete Fourier transform matrix subunit, the discrete Fourier transform matrix subunit is used to: The partial data of the voltage pulse signal corresponding to each of the one-dimensional data groups and the negative number of the partial data of each of the voltage pulse signals are input into the memristor array, so that the partial data of each of the voltage pulse signals and the negative number of the partial data of each of the voltage pulse signals are weightedly summed with the conductance value of the corresponding memristor in the memristor array to obtain the partial data of the current pulse signal corresponding to each of the one-dimensional data groups output by the memristor array; the partial data includes real data or imaginary data.

7. A computing method applied to the Fourier transform in-memory computing system according to any one of claims 1 to 6, characterized in that: include: In the current iterative operation, the N-dimensional input data is split into multiple one-dimensional data groups according to the splitting dimension corresponding to the N-dimensional input data, and a corresponding voltage pulse signal is output according to each one-dimensional data group; the N-dimensional input data is the original N-dimensional data input in the first iterative operation, or the N-dimensional reorganized data output by the post-processing module in the first N-1 iterative operations; Using a discrete Fourier transform operator, based on the voltage pulse signal corresponding to each of the one-dimensional data groups, output a current pulse signal corresponding to each of the one-dimensional data groups; Each group of the current pulse signals is converted into a corresponding one-dimensional conversion data group, and each of the one-dimensional conversion data groups is reorganized according to the splitting dimension to obtain the N-dimensional reorganized data of the current iterative operation; the N-dimensional reorganized data obtained when the current iterative operation is the N-th iterative operation is used as the N-dimensional Fourier transform result of the original N-dimensional data, and the N-dimensional Fourier transform result is output.

8. The operation method of the Fourier transform in-memory operation system according to claim 7, characterized in that: The method of using a discrete Fourier transform operator to output a current pulse signal corresponding to each one-dimensional data group based on a voltage pulse signal corresponding to each one-dimensional data group comprises: Based on the voltage pulse signal corresponding to each of the one-dimensional data groups, output the real part data of the current pulse signal corresponding to each of the one-dimensional data groups; Based on the voltage pulse signal corresponding to each of the one-dimensional data groups, outputting the imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups; Based on the real data and the imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups, the current pulse signal corresponding to each of the one-dimensional data groups is output.

9. The operation method of the Fourier transform in-memory operation system according to claim 8, characterized in that: The step of outputting the real part data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group comprises: Based on the real part data of the voltage pulse signal corresponding to each of the one-dimensional data groups, output the first real part data of the current pulse signal corresponding to each of the one-dimensional data groups; Outputting second real part data of the current pulse signal corresponding to each one-dimensional data group based on the negative number of the imaginary part data of the voltage pulse signal corresponding to each one-dimensional data group; The first real part data and the second real part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the real part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

10. The operation method of the Fourier transform in-memory operation system according to claim 9, characterized in that: The step of outputting imaginary data of the current pulse signal corresponding to each one-dimensional data group based on the voltage pulse signal corresponding to each one-dimensional data group comprises: Based on the real data of the voltage pulse signal corresponding to each of the one-dimensional data groups, output the first imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups; Based on the imaginary data of the voltage pulse signal corresponding to each of the one-dimensional data groups, output the second imaginary data of the current pulse signal corresponding to each of the one-dimensional data groups; The first imaginary part data and the second imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups are superimposed, and the imaginary part data of the current pulse signal corresponding to each of the one-dimensional data groups is output.

Citation Information

Patent Citations

  • Bidimensional compressed sensing image acquisition and reconstruction method based on discrete cosine transformation (DCT) and discrete Fourier transformation (DFT)

    CN103347189A

  • Three-dimensional Fourier transformation calculation method and device

    CN107544941A

  • Quantization and inverse quantization method in large-scale MIMO channel state information feedback

    CN110311718A

  • Hardware implementation of discrete Fourier correlation transforms

    CN115994565A

  • Signal processing system, method, product, equipment and medium

    CN118332267A

Cited By

  • WHT-based extensible in-memory image processing system and method

    CN119784600A

  • A scalable in-memory image processing system and method based on WHT

    CN119784600B