A storage and computing architecture FPGA that uses backup backtracking to overcome single-particle upset failures
By introducing backup registers and judgment circuits into the memory and calculation unit of the FPGA, comparing and correcting the data, transient errors caused by single-particle flip are solved, data processing efficiency and computing speed are improved, and it is suitable for application scenarios with dense single-particle flip.
Patent Information
- Application Number
- CN202310769915.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-27
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-06-27
AI Technical Summary
When performing calculation tasks, existing FPGAs have delays and energy losses due to frequent data transfers, which limits data processing efficiency. Especially in application scenarios where single-particle flips are dense, it is prone to transient errors, which affects the accuracy of the calculation results.
Using the backup traceback mechanism, by introducing backup registers and judgment circuits into the calculation unit, comparing the data of the target register and backup register, if an error is found, the operation will be performed again to ensure the accuracy of the operation result, and to reduce data through the global read and write channel. Transmission delay.
It improves the data processing efficiency and computing speed of FPGA, enhances operation reliability, and avoids transient errors caused by single-particle flips. It is suitable for application scenarios with dense single-particle flips.
Smart Images

Figure CN116737470B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of FPGA technology, and in particular to a storage-computing architecture FPGA that utilizes backup backtracking to overcome single-particle upset failures. Background Art
[0002] With the development of high-tech technologies such as the Internet of Things, cloud computing, and artificial intelligence, various data-intensive applications based on hardware platforms have also grown dramatically. FPGA (Field Programmable Gate Array) is a commonly used hardware platform due to its advantages of parallelism and reconfigurability.
[0003] The existing FPGA, like most traditional hardware platforms, adopts the von Neumann architecture. The FPGA contains a large number of resource modules such as CLB, DSP and BRAM. These resource modules are connected by highly configurable interconnects to form computing units and storage units in the FPGA. BRAM is an important on-chip storage unit. On-chip data is mainly stored in BRAM, and on-chip data must be transferred to the computing unit during calculation.
[0004] Therefore, existing FPGAs, like other hardware platforms that use the von Neumann architecture, need to repeatedly transfer large amounts of data between storage units and computing units when performing various computing tasks. Frequent and large-scale data movement not only consumes a large amount of wiring resources and interconnection resources, but also causes huge delays and energy losses, thereby limiting the data processing efficiency of FPGAs. Summary of the invention
[0005] In response to the above problems and technical requirements, the applicant has proposed a storage-computing architecture FPGA that uses backup backtracking to overcome single-particle upset failures. The technical solution of this application is as follows:
[0006] A storage and computing architecture FPGA that overcomes single-particle upset failures by using backup backtracking, wherein the storage and computing architecture FPGA includes a plurality of resource modules arranged according to a predetermined structure, interconnection resources arranged around the resource modules, and a global read-write channel implemented by additional hardware resources. Resource modules located in the same sub-area are connected through the interconnection resources inside the storage and computing architecture FPGA to implement a storage and computing unit. Each storage and computing unit includes a processor and a local storage unit. The local storage unit of each storage and computing unit is used to store instructions and data of the storage and computing unit. The processor in each storage and computing unit is connected to the local storage unit in the same storage and computing unit via the interconnection resources and operates according to the instructions and data stored in the local storage unit. Resource modules in multiple different sub-areas correspondingly implement multiple storage and computing units, and the local storage unit in each storage and computing unit transmits data with the off-chip of the storage and computing architecture FPGA through the global read-write channel.
[0007] There is at least one storage-computing unit including a backup register corresponding to a target register and a determination circuit. The target register is a register in the storage-computing unit that will be affected by a single-particle upset. During the process of the storage-computing unit executing a storage-computing operation, the backup register corresponding to each target register is used to back up and store the register data written into the target register. The determination circuit compares and determines the register data read from the target register and its corresponding backup register, and obtains a determination result when the storage-computing unit ends the current storage-computing operation. When the determination result obtained by the determination circuit indicates that an operation error occurs in the storage-computing unit, the storage-computing unit is triggered to re-execute the current storage-computing operation. Otherwise, it is determined that the storage-computing unit completes the current storage-computing operation.
[0008] Its further technical solution is that when the judgment circuit compares and determines that the register data read out from a target register and its corresponding backup register are consistent, the comparison result corresponding to the target register is determined to be correct register data, otherwise the comparison result corresponding to the target register is determined to be wrong register data; when the storage-calculation unit ends the current storage-calculation operation, the judgment result is obtained according to the comparison results corresponding to all target registers in the storage-calculation unit.
[0009] A further technical solution is that when the number of target registers whose comparison results in the storage-calculation unit indicate that the register data is erroneous reaches a predetermined threshold, a determination result indicating that an operation error has occurred in the storage-calculation unit is obtained.
[0010] A further technical solution is that the determination circuit includes a comparator and a counter, the comparator is used to compare the target register and its corresponding backup register, and the counter is used to count the number of target registers whose comparison results are register data errors.
[0011] Its further technical solution is that the judgment circuit includes a comparator, which is used to compare the target register and its corresponding backup register, and the status register of the status unit inside the processor in the storage and calculation unit is used to count the number of target registers whose comparison results are register data errors.
[0012] A further technical solution is to read a predetermined number of bits of a status register of a status unit inside a processor in the storage and calculation unit to obtain the number of target registers whose comparison results are register data errors.
[0013] A further technical solution is that the predetermined threshold is adjustable, and the higher the accuracy requirement of the calculation result obtained by the storage-computing architecture FPGA on the storage-computing unit performing the storage-computing operation, the smaller the predetermined threshold is.
[0014] A further technical solution is to trigger the storage and calculation unit to re-execute the current item storage and calculation operation, including:
[0015] After determining that the judgment result indicates that a calculation error occurs in the storage and calculation unit, after a predetermined delay, the storage and calculation unit is triggered to re-execute the current storage and calculation operation.
[0016] A further technical solution is that when the storage-computing unit performs a storage-computing operation, the storage-computing instruction and the initial data are written into the local storage unit of the storage-computing unit via the global read-write channel, and the processor in the storage-computing unit performs the storage-computing operation based on the initial data according to the storage-computing instruction, and in the process of performing the storage-computing operation, the storage-computing instruction and the initial data stored in the local storage unit remain unchanged;
[0017] The storage and computing unit re-executes the current storage and computing operation, including: the processor in the storage and computing unit re-executes the storage and computing operation based on the initial data according to the storage and computing instruction.
[0018] The beneficial technical effects of this application are:
[0019] The present application discloses a storage-computing architecture FPGA that utilizes backup backtracking to overcome single-particle flip failures. In the storage-computing architecture FPGA, resource modules located in the same sub-area are connected via interconnection resources within the FPGA to implement a storage-computing unit. In addition, for application scenarios with dense single-particle flips, the storage-computing unit in the storage-computing architecture FPGA adds a corresponding backup register for a target register that will be affected by the single-particle flip. In the process of executing storage-computing operations, the register data read from the target register and its corresponding backup register are compared and determined, and when the storage-computing unit ends the current storage-computing operation, it is determined whether to save the operation result or trigger the re-execution of the storage-computing operation. This can avoid operation errors caused by transient errors caused by single-particle flip failures in application scenarios with dense single-particle flips, thereby improving operation reliability when implementing parallel multi-core storage-computing operations through multiple storage-computing units, and has outstanding data processing efficiency, operation speed and operation reliability.
[0020] The storage-computing architecture FPGA of this application is mainly suitable for application scenarios with intensive single-particle flips. After the storage-computing unit completes a storage-computing operation, it determines whether to trigger re-execution, so as to avoid the phenomenon that the normal execution process of the storage-computing operation is affected by the intensive occurrence of single-particle flips.
[0021] The processor and local data storage in each storage and computing unit formed inside the FPGA are in the same sub-area. When executing, the processor only needs to read and write data in the adjacent local data storage, which reduces the wiring resources and interconnection resources consumed by transmitting data between the processor and the local data storage, reduces data transmission delay and energy loss, and improves the data processing efficiency of the FPGA. In addition, the FPGA can realize multi-core parallel storage and computing operations through multiple storage and computing units. The processors of each storage and computing unit will not cause conflicts with each other. Each storage and computing unit can perform operations at full speed, which is not affected by the storage bandwidth, and can further improve the overall data processing efficiency and computing speed of the FPGA. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a schematic diagram of the arrangement of resource modules inside the storage-computing architecture FPGA and a schematic diagram of the sub-areas obtained by division in an example of the present application.
[0023] Figure 2 is based on Figure 1 Schematic diagram of the circuit structure of each storage and computing unit implemented.
[0024] Figure 3 It is a circuit diagram of a target register, a backup register, and a determination circuit in a single storage and calculation unit in one embodiment of the present application.
[0025] Figure 4 This is a flow chart of a method in which a single storage and computing unit implements a storage and computing operation in one embodiment of the present application.
[0026] Figure 5 yes Figure 2 Another circuit structure diagram in which the local storage units of each storage and computing unit are connected to the outside of the FPGA chip.
[0027] Figure 6 yes Figure 2 Another circuit structure diagram in which the local storage units of each storage and computing unit are connected to the outside of the FPGA chip.
[0028] Figure 7 yes Figure 2 Another circuit structure diagram in which the local storage units of each storage and computing unit are connected to the outside of the FPGA chip. DETAILED DESCRIPTION
[0029] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.
[0030] The present application discloses a storage and computing architecture FPGA that utilizes backup backtracking to overcome single-particle upset failures. The storage and computing architecture FPGA, like a conventional FPGA, includes a number of resource modules arranged in a predetermined structure and interconnected resources arranged around the resource modules. The predetermined structure can be arranged according to actual conditions. For example, taking the currently mainstream Column-Based structure as an example, the resource modules are arranged in a row and column structure to form a two-dimensional array architecture, and each column is arranged with resource modules of the same type. Interconnected resources are arranged around each resource module. The types of resource modules contained in the FPGA include configurable logic modules CLB, configurable storage modules BRAM, and configurable multiplication and addition modules DSP. The configurable logic module CLB contains units such as lookup tables and registers, which can be configured to implement various logical functions. The configurable multiplication and addition module DSP contains arithmetic units that can perform multiple operations such as multiplication, addition, and multi-bit logic gates. The configurable storage module BRAM contains block storage units that can implement single- and double-port storage and FIFO storage of various widths and depths. This part of the structure follows the structure inside the existing FPGA, and this application will not elaborate on it.
[0031] All resource modules arranged in a predetermined structure in the storage and computing architecture FPGA of the present application are divided into several sub-areas, each of which contains a certain number of resource modules. Resource modules in the same sub-area are connected through the interconnection resources inside the FPGA to realize a storage and computing unit, and resource modules in multiple different sub-areas correspondingly realize multiple storage and computing units.
[0032] Each storage and computing unit includes its own processor and local storage unit. The processor in each storage and computing unit is connected to the local storage unit in the storage and computing unit via an interconnection resource. The local storage unit in each storage and computing unit is used to store the instructions and data of the storage and computing unit. The processor in each storage and computing unit is used to operate according to the instructions and data stored in the connected local storage unit.
[0033] Each processor includes multiple necessary modules, including a calculation unit (ALU), an instruction address (Program Counter), a control unit (Control), an interrupt unit (Interrupt), a stack address (Stack Pointer) and a status unit (Status). In one embodiment, the instruction sets used by the two storage and calculation units implemented by the resource modules in any two sub-areas are the same or different, and the bit widths of the two storage and calculation units using the same instruction set are the same or different. In the present application, the storage and calculation units formed inside the storage and calculation architecture FPGA can form various typical general-purpose 8-bit processors including RISC, 8051 and 6802.
[0034] In one embodiment, the number and type of resource modules contained in any two sub-areas obtained by division are the same, or in another embodiment, the number and type of resource modules contained in any two sub-areas obtained by division are at least one different, that is, the multiple sub-areas obtained by division may be completely the same or not necessarily the same. In addition, each resource module in the storage-computing architecture FPGA belongs to a sub-area, all resource modules in the storage-computing architecture FPGA are divided into corresponding sub-areas, or some resource modules in the storage-computing architecture FPGA are divided into corresponding sub-areas, that is, the storage-computing architecture FPGA can be divided into sub-areas globally or locally. For example, Figure 1 In the example shown, the resource modules inside the FPGA are divided into 8 sub-areas, each of which contains the same number and type of resource modules. Each area framed by a dotted line represents a sub-area obtained by the division. Figure 1 For the eight sub-areas obtained by dividing the storage-computing architecture FPGA, the storage-computing architecture FPGA implements eight storage-computing units through resource modules in the eight sub-areas, and each storage-computing unit includes a processor and a local storage unit. Figure 2 shown.
[0035] This application does not limit the number and type of resource modules contained in each sub-area, but the number and type of resource modules contained in each sub-area must be sufficient to implement the devices required for a storage and computing unit. The processor in a storage and computing unit is mainly implemented based on DSP, and the storage unit in a storage and computing unit is mainly implemented based on BRAM. In addition, a certain number of CLBs are also required.
[0036] The storage and computing unit implemented by the resource module in the sub-area actually includes many registers, such as instruction registers and data registers: when the processor reads instructions from the local storage unit, the instructions in the local storage unit are first written into the instruction register IR, and then written into the processor by the instruction register IR. When the processor reads data from the local storage unit, the data in the local storage unit is first written into the data register DR, and then written into the processor by the data register DR. Correspondingly, when the processor writes data to the local storage unit, it first writes the data into the data register DR, and then writes it into the local storage unit by the data register DR. In addition to the above-mentioned instruction registers and data registers, the storage and computing unit also includes many general registers necessary for the circuit structure. In a complex electromagnetic environment, the registers will be affected by single-particle upsets due to radiation exposure and other reasons. Although the single-particle upset effect is transient and non-destructive, it will change the content of the register, thereby causing persistent functional errors and causing errors in the operation of the storage and computing unit.
[0037] The storage-computing architecture FPGA disclosed in the present application is mainly aimed at application environments with dense single-particle flips, that is, application environments with a large number of electromagnetic radiation particles and a long duration. Therefore, in order to ensure that the storage-computing architecture FPGA can also work reliably in such application environments, there is at least one storage-computing unit in the storage-computing architecture FPGA of the present application, which includes a backup register corresponding to the target register and a judgment circuit, and the judgment circuit connects each target register and its backup register. The target register is a register in the storage-computing unit that will be affected by a single-particle flip. In actual applications, all registers in the storage-computing unit that will be affected by a single-particle flip are used as target registers, and each target register has a corresponding backup register. Please refer to Figure 3 For example, in a storage-computation unit, there are four target registers, namely R1, R2, R3, and R4. The backup register R1b is the backup register of the target register R1, the backup register R2b is the backup register of the target register R2, the backup register R3b is the backup register of the target register R3, and the backup register R4b is the backup register of the target register R4.
[0038] In addition to resource modules and interconnection resources, the storage and computing architecture FPGA of this application also adds additional hardware resources to implement global read and write channels compared to conventional FPGAs. The local storage unit in each storage and computing unit transmits data to the FPGA chip through the global read and write channels, such as Figure 2 shown.
[0039] The storage and computing architecture of the present application forms multiple storage and computing units inside the FPGA, and each storage and computing unit supports independent storage and computing operations in the form of a soft core. Therefore, the FPGA can jointly implement a multi-core storage and computing operation through multiple storage and computing units formed inside, including the following steps, please refer to Figure 4 :
[0040] 1. Write the storage and computing instructions and initial data of each storage and computing unit into the local storage unit of the corresponding storage and computing unit from outside the FPGA chip through the global read and write channel.
[0041] 2. Start each storage and computing unit to perform parallel operations simultaneously. The processor in each storage and computing unit performs storage and computing operations based on the initial data according to the storage and computing instructions in the local storage unit of the storage and computing unit.
[0042] For any storage unit including a target register and a determination circuit, the write ports of all the target registers and their backup memories of the storage unit are connected to the write bus. During the process of the storage unit performing a storage operation, the backup register corresponding to each target register is used to backup and store the register data written into the target register. That is, when the register data is written into the target register via the write bus, it is also written into the corresponding backup register via the write bus. For example, Figure 3 The register data Data_R1 written into the target register R1 is also written into the backup register R1b, the register data Data_R2 written into the target register R2 is also written into the backup register R2b, the register data Data_R3 written into the target register R3 is also written into the backup register R3b, and the register data Data_R4 written into the target register R4 is also written into the backup register R4b.
[0043] In addition, the read ports of the backup registers corresponding to each target register in the storage and computing unit are connected to the determination circuit. For any target register in the storage and computing unit, when reading data from the target register, the determination circuit in the storage and computing unit reads the register data from the target register and its corresponding backup register respectively at the same time, and compares and determines the register data READ_R read from the target register and the register data READ_Rb read from the corresponding backup register.
[0044] Since the target register and its corresponding backup register write the same register data, in the absence of data errors, the register data READ_R read out of the target register and the register data READ_Rb read out of its corresponding backup register should be consistent. However, in an application environment with dense single-particle flips, the content of the register data in a target register or its corresponding backup register may change due to the influence of single-particle flips. Even if both the target register and the backup register are affected by single-particle flips, the content of the register data read out of the two registers will not change equally, thus causing the register data READ_R and READ_Rb read out of the two registers to be inconsistent. Therefore, in one embodiment, when the determination circuit compares and determines that the register data READ_R and READ_Rb read out of a target register and its corresponding backup register are consistent, the comparison result corresponding to the target register is determined to be correct register data, otherwise the comparison result corresponding to the target register is determined to be a register data error. Although in actual application, the data inconsistency may be caused by a data error in the backup register, this application also classifies this situation as a register data error in the target register for data accuracy considerations.
[0045] Since the storage and computing architecture FPGA of the present application is mainly aimed at application environments with dense single-particle flips, dense electromagnetic radiation particles often cause a large number of registers to be affected by single-particle flips and cause data errors. If a register data error is detected in a register, the processor is controlled to suspend the storage and computing operation or restart the storage and computing unit, which will cause the loop to be interrupted continuously, making it difficult to complete the storage and computing operation.
[0046] Therefore, during the storage-calculation process, the comparison results corresponding to each target register are compared and determined respectively and the comparison results are recorded. When the storage-calculation unit ends the current storage-calculation operation, the judgment result is finally obtained according to the comparison results corresponding to all the target registers in the storage-calculation unit. In one embodiment, when the number of target registers whose comparison results in the storage-calculation unit are register data errors reaches a predetermined threshold, a judgment result indicating that a calculation error has occurred in the storage-calculation unit is obtained. In one embodiment, the predetermined threshold is adjustable, and the higher the accuracy requirement of the storage-calculation architecture FPGA for the calculation result obtained by the storage-calculation unit when performing the storage-calculation operation, the smaller the predetermined threshold.
[0047] That is, when the storage-calculation unit finishes the current storage-calculation operation, the determination circuit needs to count the number of target registers whose comparison results are register data errors. One implementation method is that the determination circuit includes a comparator and a counter. The comparator is used to compare the target register and its corresponding backup register, and the counter is used to count the number of target registers whose comparison results are register data errors. That is, the counter is implemented by a separate hardware resource, and the number is counted by the counter, such as Figure 3 As shown, the comparator compares the register data READ_R read from the target register and the register data READ_Rb read from the corresponding backup register. When the two sets of register data are consistent, the comparator outputs 0, otherwise the comparator outputs 1, and the count value of the counter is increased by 1. When the storage and calculation unit ends the current storage and calculation operation, the count value of the counter is read to determine the number of target registers with incorrect register data.
[0048] Another implementation is that the determination circuit includes a comparator, the comparator is used to compare the target register and its corresponding backup register, and the status register of the status unit (Status) inside the processor in the storage and calculation unit is used to count the number of target registers whose comparison results are register data errors. It includes: reading the predetermined number of bits of the status register of the status unit inside the processor in the storage and calculation unit to obtain the number of target registers whose comparison results are register data errors. That is, in this implementation, there is no need to add additional hardware resources, and some status bits of the status register of the status unit (Status) inside the processor are directly reused to realize the counting function, so that the architecture design of the storage and calculation unit can be simplified. For example, when the predetermined threshold is 1, one status bit of the status register of the status unit inside the processor can be directly used. For another example, when the predetermined threshold is 7, 3 status bits of the status register of the status unit inside the processor can be used.
[0049] When the storage unit ends the current storage operation, and the determination result obtained by the determination circuit indicates that the storage unit has an operation error, the storage unit is triggered to re-execute the current storage operation. Since the storage architecture is aimed at application scenarios with dense single-particle flips, in one embodiment, after determining that the determination result indicates that the storage unit has an operation error, the storage unit is triggered to re-execute the current storage operation after a predetermined delay. The predetermined delay time can be set by user, or a random value can be taken within a predetermined time interval. By delaying the predetermined time, it is possible to wait for the current burst of dense electromagnetic radiation particles to pass, thereby avoiding the storage unit being repeatedly triggered to re-execute when the storage unit is triggered to be still affected by the dense electromagnetic radiation particles when re-executing the current storage operation. Since the single-particle flip failure is transient and non-destructive, the register data error caused by the single-particle flip can be corrected when the current storage operation is re-executed to ensure operational reliability. In order to enable the storage and computing unit to re-execute the current storage and computing operation, in the present application, after the storage and computing instructions and initial data of the storage and computing unit are written into the local storage unit through step 1, during the process of the processor executing the storage and computing operation, the storage and computing instructions and initial data stored in the local storage unit remain unchanged, so that the processor can re-execute the current storage and computing operation by re-executing the storage and computing instructions in the local storage unit based on the initial data.
[0050] When the storage unit completes the current storage operation and the determination result obtained by the determination circuit indicates that the storage unit has no operation error, it is determined that the storage unit completes the current storage operation and stores the generated operation result in the local storage unit of the storage unit.
[0051] 3. The calculation results in the local storage unit of each storage unit are read out to the outside of the FPGA chip through the global read and write channel.
[0052] On this basis, according to actual needs, the storage-computing architecture FPGA can also realize multiple storage-computing operations in sequence through the internal storage-computing units until all operations are completed. And the storage-computing architecture FPGA can use all or part of the storage-computing units when realizing any storage-computing operation, and the storage-computing units used when realizing each storage-computing operation can be the same or different.
[0053] Since the processor and local storage unit of each storage unit are in the same sub-area, the processor only needs to read and write the storage instructions and initial data in the adjacent local storage unit during execution, thereby reducing the wiring resources and interconnection resources consumed by data transmission between the processor and the local storage unit, reducing data transmission delay and energy loss, and improving the data processing efficiency of the storage architecture FPGA. In addition, in a multi-core scenario, if multiple computing cores read and write the same main memory at the same time, the read and write bandwidth of the main memory will form a bottleneck, restricting the data processing efficiency. This problem will be particularly obvious in high-performance computing. For FPGAs that support artificial intelligence or machine learning applications, the FPGA generally has a large number of relatively simple computing cores, and the number of computing cores will exceed 100. The demand for the read and write bandwidth of the main memory is greater, and the problem is more prominent. The storage architecture FPGA of the present application is also a multi-core scenario. Since the processors in each storage unit read and write local storage units respectively, instead of all processors uniformly reading and writing the same memory, the processors of each storage unit will not cause conflicts with each other, and each storage unit can perform operations at full speed, which is not affected by the storage bandwidth, which is conducive to improving the overall data processing efficiency and computing speed. This makes the storage-computing architecture FPGA of the present application applicable to scenarios involving a large number of storage-computing units. Even if the number of storage-computing units exceeds 100, it can still have a high computing rate, thereby better supporting artificial intelligence and machine learning scenarios.
[0054] In one embodiment, when the predetermined structure adopted by the arrangement of resource modules inside the storage and computing architecture FPGA of the present application is a Column-Based structure, the BRAM module occupies one or more columns, and the local storage units in each storage and computing unit are implemented based on the BRAM module. Therefore, the BRAM module used to implement the local storage unit in each storage and computing unit is located in one or more columns, and the local storage units in the storage and computing units and several BRAM modules located in the same column are used to implement data transmission with the storage and computing architecture FPGA chip via the same global read-write channel. The storage and computing architecture FPGA chip can exchange data with any local storage unit connected to the global read-write channel via each global read-write channel. Based on this, the present application provides the following two methods to implement data transmission between the local storage unit in each storage and computing unit and the storage and computing architecture FPGA chip:
[0055] In one embodiment, several BRAM modules located in the same column and used to implement local storage units in the storage and computing units are connected to an on-chip read / write interface of the FPGA through the same global read / write channel to perform data transmission with the outside of the FPGA, and different global read / write channels are connected to different on-chip read / write interfaces of the FPGA. Figure 2In the example shown, eight storage and computing units are formed inside the storage and computing architecture FPGA. The local storage units in storage and computing units 1 to 4 are implemented by BRAM modules located in the same column. The local storage units of these four storage and computing units are connected to the on-chip read-write interface 1 of the storage and computing architecture FPGA through the global read-write channel 1 to realize data transmission with the outside of the storage and computing architecture FPGA. The local storage units in storage and computing units 5 to 8 are implemented by BRAM modules located in the same column. The local storage units of these four storage and computing units are connected to the on-chip read-write interface 2 of the storage and computing architecture FPGA through the global read-write channel 2 to realize data transmission with the outside of the FPGA.
[0056] The method provided in the above embodiment often needs to occupy multiple on-chip read / write interfaces of the FPGA, and the on-chip read / write interface resources of the FPGA are relatively precious. Therefore, in another embodiment, only one on-chip read / write interface of the storage-computing architecture FPGA is needed to realize the local storage unit in the soft-core processor, and several BRAM modules located in the same column are connected to an active end of the multiplexer MUX through the same global read / write channel, and the fixed end of the multiplexer MUX is connected to the on-chip read / write interface of the FPGA. By selecting each active end of the multiplexer MUX, each global read / write channel can be selected, so that data can be exchanged with any local storage unit connected to the global read / write channel outside the FPGA chip via the selected global read / write channel. For example, in Figure 5 In the structure diagram shown, it corresponds to Figure 2 , the local storage units in storage and computing units 1 to 4 are connected to an active end of the multiplexer MUX through the global read-write channel 1, and the local storage units in storage and computing units 5 to 8 are connected to the other active end of the multiplexer MUX through the global read-write channel 2. The fixed end of the multiplexer MUX is connected to the on-chip read-write interface of the storage and computing architecture FPGA to realize data transmission with the outside of the chip. Compared with Figure 2 The method of this embodiment only needs to occupy one on-chip read / write interface of the FPGA. It should be noted that Figure 5 Mainly to show the structure of the local storage unit connected to the FPGA chip outside the storage and computing architecture, the check circuit and parity register in each storage and computing unit are omitted, and the subsequent figures are similarly omitted.
[0057] According to the traditional von Neumann architecture, the instructions and data in each storage unit are mixed and stored. However, in order to improve the data processing efficiency, in one embodiment, according to the Harvard structure, the instructions and data of each storage unit are stored separately and independently. Then the local storage unit in a storage and computing unit includes an instruction local storage unit and a data local storage unit. The instruction local storage unit is used to store the instructions of the storage and computing unit, and the data local storage unit is used to store the data of the storage and computing unit, so that the instructions and data do not interfere with each other when the processor operates, thereby improving the data processing efficiency. The instruction local storage unit is implemented by at least one BRAM module, and the data local storage unit is implemented by at least one BRAM module. Therefore, based on this, a common sub-area includes at least one DSP, two BRAMs, and 300 CLBs.
[0058] When the local storage unit in each storage and computing unit includes an instruction local storage unit and a data local storage unit, the storage size of the instruction local storage units of any two storage and computing units is the same or different, and the storage size of the data local storage units of any two storage and computing units is the same or different.
[0059] When the local storage unit in each storage unit includes an instruction local storage unit and a data local storage unit: (1) In one embodiment, the instruction local storage unit and the data local storage unit in a storage unit share a global read / write channel for data transmission, and the instruction local storage unit and the data local storage unit in the storage unit and the BRAM modules located in the same column are used to transmit data with the storage unit architecture FPGA chip via the same global read / write channel. In this embodiment, the global read / write channel can be used as follows: Figure 2 The method of the embodiment shown is connected to the on-chip read / write interface of the storage computing architecture FPGA, or according to Figure 5 The method of the embodiment shown is connected to an active terminal of the multiplexer MUX. Figure 2 Taking the structure shown in the figure as an example, Figure 6As shown, the instruction local storage units and data local storage units in storage and computing units 1 to 4 are connected to the global read-write channel 1, and the instruction local storage units and data local storage units in storage and computing units 5 to 8 are connected to the global read-write channel 2. (2) Or in another embodiment, the instruction local storage units and data local storage units use their own global read-write channels respectively, and the instruction local storage units in a storage and computing unit transmit instructions via the instruction global read-write channel, and the data local storage units transmit data via the data global read-write channel. Then, the instruction local storage units in the storage and computing units and several BRAM modules in the same column transmit data to the outside of the storage and computing architecture FPGA chip via the same instruction global read-write channel, and the data local storage units in the storage and computing units and several BRAM modules in the same column transmit data to the outside of the storage and computing architecture FPGA chip via the same data global read-write channel. Similarly in this embodiment, each instruction global read-write channel and data global read-write channel can be further configured according to Figure 2 The method of the embodiment shown is connected to the on-chip read / write interface of the storage computing architecture FPGA, or according to Figure 5 The method of the embodiment shown is connected to an active terminal of the multiplexer MUX. Figure 5 Taking the structure shown in the figure as an example, Figure 7 The instruction local storage units in storage and computing units 1 to 4 are all connected to the instruction global read and write channel 1, the data local storage units in storage and computing units 1 to 4 are all connected to the data global read and write channel 1, the instruction local storage units in storage and computing units 5 to 8 are all connected to the instruction global read and write channel 2, the data local storage units in storage and computing units 5 to 8 are all connected to the data global read and write channel 2, the instruction global read and write channel 1, the data global read and write channel 1, the instruction global read and write channel 2 and the data global read and write channel 2 are respectively connected to an active end of the multiplexer MUX.
[0060] In another embodiment, the resource module in each sub-area is also used to implement the control signal and status signal of a corresponding storage and computing unit, the control signal is used to control the operating status of the storage and computing unit, and the status signal is used to display the operating status of the storage and computing unit. The control signal and status signal of the storage and computing unit are connected to other circuit structures inside the FPGA, or the control signal and status signal of the storage and computing unit are connected to the outside of the FPGA. That is, the control signal and status signal required by each storage and computing unit when performing storage and computing are implemented by the resource module in the corresponding sub-area.
[0061] The control signal of each storage and computing unit includes at least one of an enable signal, a clock signal, a reset signal and an interrupt signal. The enable signal is used to control the corresponding storage and computing unit to execute instructions or stop executing instructions. The reset signal is used to reset the storage and computing unit. The interrupt signal provides external interrupt capability. The control signals of any two storage and computing units are the same or different.
[0062] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.
Claims
1. A storage-computing architecture FPGA that uses backup backtracking to overcome single-particle upset failures, characterized in that: The storage and computing architecture FPGA includes a plurality of resource modules arranged in a predetermined structure, interconnection resources arranged around the resource modules, and a global read-write channel realized by additional hardware resources. The resource modules located in the same sub-area are connected through the interconnection resources inside the storage and computing architecture FPGA to realize a storage and computing unit. Each storage and computing unit includes a processor and a local storage unit. The local storage unit of each storage and computing unit is used to store instructions and data of the storage and computing unit. The processor in each storage and computing unit is connected to the local storage unit in the same storage and computing unit via the interconnection resources and operates according to the instructions and data stored in the local storage unit. The resource modules in multiple different sub-areas correspondingly realize multiple storage and computing units, and the local storage unit in each storage and computing unit transmits data with the off-chip of the storage and computing architecture FPGA through the global read-write channel. There is at least one storage unit including a backup register corresponding to a target register and a determination circuit, wherein the target register is a register in the storage unit that will be affected by a single particle upset, and during the process of the storage unit performing a storage operation, the backup register corresponding to each target register is used to back up and store register data written into the target register, and the determination circuit compares and determines the register data read from the target register and its corresponding backup register, and obtains a determination result when the storage unit ends the current storage operation, and when the determination result obtained by the determination circuit indicates that an operation error occurs in the storage unit, the storage unit is triggered to re-execute the current storage operation, otherwise it is determined that the storage unit completes the current storage operation; When the determination circuit determines that the register data read out of a target register and its corresponding backup register are consistent, the comparison result corresponding to the target register is determined to be correct register data, otherwise the comparison result corresponding to the target register is determined to be wrong register data; when the storage-calculation unit ends the current storage-calculation operation, the determination result is obtained according to the comparison results corresponding to all target registers in the storage-calculation unit; When the number of target registers whose comparison results in the storage-calculation unit indicate that the register data is wrong reaches a predetermined threshold, a determination result indicating that an operation error occurs in the storage-calculation unit is obtained.
2. The storage-computing architecture FPGA according to claim 1, characterized in that: The determination circuit comprises a comparator and a counter. The comparator is used to compare a target register and its corresponding backup register. The counter is used to count the number of target registers whose comparison results are register data errors.
3. The storage-computing architecture FPGA according to claim 1, characterized in that: The determination circuit includes a comparator, which is used to compare the target register and its corresponding backup register. The status register of the status unit inside the processor in the storage and calculation unit is used to count the number of target registers whose comparison results are register data errors.
4. The storage-computing architecture FPGA according to claim 3, characterized in that: The predetermined number of bits of the status register of the status unit inside the processor in the storage and calculation unit is read to obtain the number of target registers whose comparison results are register data errors.
5. The storage-computing architecture FPGA according to claim 1, characterized in that: The predetermined threshold is adjustable, and the higher the accuracy requirement of the calculation result obtained by the storage-computing architecture FPGA on the storage-computing unit performing the storage-computing calculation, the smaller the predetermined threshold is.
6. The storage-computing architecture FPGA according to claim 1, characterized in that: The triggering the storage and calculation unit to re-execute the current item storage and calculation operation includes: After determining that the judgment result indicates that the storage and calculation unit has an operation error, after a predetermined delay, the storage and calculation unit is triggered to re-execute the current storage and calculation operation.
7. The storage-computing architecture FPGA according to claim 6, characterized in that: When the storage and computing unit performs a storage and computing operation, the storage and computing instruction and the initial data are written into the local storage unit of the storage and computing unit via the global read and write channel, and the processor in the storage and computing unit performs the storage and computing operation based on the initial data according to the storage and computing instruction, and in the process of performing the storage and computing operation, the storage and computing instruction and the initial data stored in the local storage unit remain unchanged; The storage and computing unit re-executing the current storage and computing operation includes: a processor in the storage and computing unit re-executing the storage and computing operation based on the initial data according to the storage and computing instruction.
Citation Information
Patent Citations
Cycle granularity-based interstage register backup apparatus and method for dual pipeline
CN107168826A
Storage and calculation integrated data reading and decoding circuit and storage and calculation integrated memory
CN113192544A
Field programmable gate array (FPGA) with automatic error checking and correcting function for programmable logic module
CN113608918A
Single event upset error correction method and device, electronic equipment and readable storage medium
CN115543683A
Two redundant assembly linies based on checkpoint technique
CN206833419U
