Many-core computing circuit, stacked chip and fault-tolerant control method
By introducing backup mechanisms of computing units, storage units and data channels into the multi-core chip, the automatic replacement of failed units and channels is achieved, which solves the problem of insufficient read and write access reliability in the multi-core chip structure and improves the quality of chip output Rate.
Patent Information
- Application Number
- CN202110752151.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-02
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2041-07-02
AI Technical Summary
The prior art cannot effectively improve the overall read and write access reliability between PE and memory cells in the multi-core chip structure, resulting in a low yield on the chip.
Using a multi-core computing circuit with fault tolerance, including a computing unit module, a storage unit module and an on-chip bus module. Each module is equipped with an instance and a backup unit. By replacing the failed instance unit and channel, it realizes that Failure repair of computing units, storage units and data channels.
It improves the read and write access reliability between the computing unit and the storage unit, and enhances the chip yield of the multi-core chip structure.
Smart Images

Figure CN113360323B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of many-core chips, and in particular to a many-core computing circuit, a stacked chip and a fault-tolerant control method. Background Art
[0002] The many-core chip structure generally includes multiple PEs (process engines), and each PE accesses different storage units through the NOC (network on chip) to complete the operation. Therefore, the read and write access reliability between the PE and the storage unit in the many-core chip structure can affect the chip yield of the many-core chip structure.
[0003] The existing technology adopts ECC (Error Correcting Code) technology to enable a small number of bit errors to be tolerated when reading and writing between PE and storage units, thereby improving the reliability of read and write communication between PE and storage units, thereby improving the chip yield of the multi-core chip structure.
[0004] However, as the chip area in the many-core chip structure increases, the existing technology cannot effectively improve the overall read and write access between PE and storage units in the many-core chip structure, resulting in a low chip yield of the many-core chip structure.
[0005] Therefore, how to improve the chip yield rate of multi-core chip structures is a technical problem that needs to be solved urgently. Summary of the invention
[0006] The object of the present invention is to provide a many-core computing circuit, a stacked chip and a fault-tolerant control method to improve the chip yield of the many-core chip structure.
[0007] To achieve the above objectives, the embodiments of the present invention provide the following solutions:
[0008] In a first aspect, an embodiment of the present invention provides a multi-core computing circuit with a fault-tolerant function, comprising:
[0009] A computing unit module, comprising at least one instance computing unit and at least one backup computing unit;
[0010] A storage unit module, comprising at least one instance storage unit and at least one backup storage unit;
[0011] The on-chip bus module includes at least one group of data channels; wherein each group of data channels includes a correspondingly set instance channel and a backup channel;
[0012] Each computing unit in the computing unit module is connected to each storage unit in the storage unit module through the corresponding data channel in the on-chip bus module; wherein each computing unit includes at least one instance computing unit and at least one backup computing unit; each storage unit includes at least one instance storage unit and at least one backup storage unit.
[0013] In a possible embodiment, the data channel includes: a first selector and a second selector;
[0014] The first output terminal of the first selector is connected to the first input terminal of the second selector to construct the instance channel;
[0015] The second output terminal of the first selector is connected to the second input terminal of the second selector to construct the backup channel.
[0016] In a possible embodiment, the computing unit includes:
[0017] A status register, used to store information indicating whether the operation unit is in a normal state;
[0018] An enable register, used to store information indicating whether the computing unit is in an activated state;
[0019] The logic identity information register is used to store the logic identity information representing the operation unit when the many-core computing circuit is working.
[0020] In a possible embodiment, the example computing unit includes one or more of a central processing unit, a digital signal processing chip, a graphics processor, and an intellectual property core module;
[0021] The backup computing unit includes one or more of a central processing unit, a digital signal processing chip, a graphics processor and an intellectual property core module.
[0022] In a possible embodiment, each storage unit in the storage unit module includes one or more of a read-only memory and a random access memory.
[0023] In a second aspect, an embodiment of the present invention provides a stacked chip, wherein the many-core chip includes any one of the many-core computing circuits described in the first aspect;
[0024] The computing unit module in the multi-core computing circuit is arranged on a first wafer; the storage unit module in the multi-core computing circuit is arranged on a second wafer; the on-chip bus module is arranged on the first wafer or on the second wafer;
[0025] The first wafer and the second wafer are stacked.
[0026] In a possible embodiment, the on-chip bus module is disposed on a third wafer;
[0027] The first wafer, the second wafer and the third wafer are stacked.
[0028] In a third aspect, an embodiment of the present invention provides a fault-tolerant control method based on any of the many-core computing circuits described in the first aspect, comprising:
[0029] Determine whether there is a failed instance computing unit in the computing unit module;
[0030] If there is a failed instance computing unit, the failed instance computing unit is replaced by a backup computing unit, and the instance computing unit sequence of the computing unit module is updated;
[0031] Determine whether there is a failed instance storage unit in the storage unit module;
[0032] If there is a failed instance storage unit, the failed instance storage unit is replaced by a backup storage unit in the storage unit module, and the instance storage unit sequence corresponding to the instance operation unit sequence is updated to reorganize a storage space with a continuous logical address;
[0033] Determine whether there is a failed instance channel in each group of data channels of the on-chip bus module;
[0034] If there is a failed instance channel, the failed instance channel is replaced by a backup channel corresponding to the failed instance channel, and the data channel corresponding to the failed instance channel is updated.
[0035] In a possible embodiment, the step of replacing the failed instance computing unit with a backup computing unit and updating the instance computing unit sequence of the computing unit module includes:
[0036] In the computing unit mapping relationship of the instance computing unit sequence, the physical identity information of the failed instance computing unit is removed, and the physical identity information of the backup computing unit is added;
[0037] The operation unit mapping relationship is updated according to the physical identity information of the existing operation units in the operation unit mapping relationship; wherein the operation unit mapping relationship includes a mapping relationship between the physical identity information and the logical identity information of each operation unit in the instance operation unit sequence.
[0038] In a possible embodiment, updating the computing unit mapping relationship according to the physical identity information of the existing computing unit in the computing unit mapping relationship includes:
[0039] Replacing the physical identity information of the failed instance computing unit in the computing unit mapping relationship before updating with the physical identity information of the backup computing unit to obtain the updated computing unit mapping relationship;
[0040] or,
[0041] The physical identity information of the existing computing units is used to match the logical identity information of each computing unit preset in the computing unit mapping relationship to obtain the updated computing unit mapping relationship.
[0042] In a possible embodiment, the step of replacing the failed instance storage unit with a backup storage unit in the storage unit module and updating the instance storage unit sequence corresponding to the instance computing unit sequence includes:
[0043] Obtaining physical identity information of the failed instance storage unit;
[0044] When mapping the physical identity information of the storage units in the instance storage unit sequence to the storage unit mapping relationship of the instance storage unit sequence, the physical identity information of the failed instance storage unit is skipped to generate the storage unit mapping relationship by using the physical identity information of the valid instance storage unit; wherein the storage unit mapping relationship includes a mapping relationship between the physical identity information and the logical identity information of each instance storage unit in the storage unit module.
[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0046] In the present invention, the computing unit module is provided with an instance computing unit and a backup computing unit. When the instance computing unit fails to work properly, the backup computing unit can be used to replace the failed instance computing unit, so that the computing unit module can work normally; the storage unit module is provided with an instance storage unit and a backup storage unit. When one or more instance storage units fail to work properly, the storage unit module can still provide a sufficient number of storage units for the computing unit to read and write access; at the same time, each group of data channels in the on-chip bus module is provided with an instance channel and a backup channel. When the instance channel in the data channel fails to work properly, the corresponding backup channel is used to replace the failed instance channel, so that the data channel works normally. The present invention can independently realize the failure repair of the computing unit, the storage unit and the data channel, improve the reliability of the read and write access between the computing unit and the storage unit, and thus improve the chip yield of the multi-core chip structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0048] Figure 1 is a structural schematic diagram of a multi-core computing circuit with fault tolerance provided by an embodiment of the present invention;
[0049] Figure 2 It is a schematic diagram of connection between a computing unit in a computing unit module and a storage unit in a storage unit module provided by an embodiment of the present invention;
[0050] Figure 3 is a structural schematic diagram of a data channel provided by an embodiment of the present invention;
[0051] Figure 4 is a schematic diagram of a stacked chip structure provided by an embodiment of the present invention;
[0052] Figure 5 is a flow chart of a fault-tolerant control method provided by an embodiment of the present invention;
[0053] Figure 6 is a schematic diagram of a computing unit mapping relationship when all instance computing units provided by an embodiment of the present invention are valid;
[0054] Figure 7 yes Figure 6 A schematic diagram of the operation unit mapping relationship after there is a failed instance operation unit in the example;
[0055] Figure 8 yes Figure 6 Schematic diagram of the operation unit mapping relationship after there is a failed instance operation unit in the.
[0056] Explanation of the figure marks: 100 is a many-core computing circuit, 110 is a computing unit module, 111 is an instance computing unit, 112 is a backup computing unit, 120 is a storage unit module, 121 is an instance storage unit, 122 is a backup storage unit, 130 is an on-chip bus module, 131 is a data channel, 132 is an instance channel, 133 is a backup channel, 134 is a first selector, 135 is a second selector, and 200 is a stacked chip. DETAILED DESCRIPTION
[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention belong to the scope of protection of the embodiments of the present invention.
[0058] See also Figure 1 , Figure 1 The present invention provides a schematic structural diagram of a many-core computing circuit with fault tolerance function. The many-core computing circuit 100 includes: a computing unit module 110 , a storage unit module 120 and an on-chip bus module 130 .
[0059] The computing unit module 110 is provided with at least one instance computing unit 111 and at least one backup computing unit 112 .
[0060] The instance computing unit 111 and the backup computing unit 112 are both a type of computing unit (process engine, PE). The computing units in the many-core computing circuit 100 can access different storage units (instance computing unit 111 and / or backup computing unit 112) through the on-chip bus (network on chip, NOC) to implement preset computing functions.
[0061] In practical applications, the instance computing unit 111 may be implemented by one or more of a central processing unit (CPU), a digital signal processing chip (DSP), a graphics processing unit (GPU), and an intellectual property core module (IP core), and the backup computing unit 112 may also be implemented by one or more of a central processing unit, a digital signal processing chip, a graphics processor, and an intellectual property core module.
[0062] Specifically, the IP core may be a pre-designed circuit function module used in an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).
[0063] Of course, the instance computing unit 111 and the backup computing unit 112 may also select other processor chips according to actual needs, which is not limited here.
[0064] When the many-core computing circuit 100 is operating normally, the instance computing unit 111 needs to be connected to the circuit to implement the preset computing function. When one or more instance computing units 111 fail, the backup computing unit 112 can be connected to the circuit to replace the failed instance computing unit 111 to ensure the normal operation of the many-core computing circuit 100, implement the failure repair function of the computing unit, improve the reliability of the read and write access between the computing unit and the storage unit 121, and thereby improve the yield of the stacked chip 200.
[0065] The storage unit module 120 includes at least one instance storage unit 121 and at least one backup storage unit 122. When the many-core computing circuit 100 is operating normally, the instance computing unit 111 and the backup computing unit 112 can configure 0, 1 or more instance storage units 121 from the instance storage unit 121 and the backup storage unit 122 respectively, so as to implement the read and write access operations or corresponding design functions of the instance computing unit 111.
[0066] The instance storage unit 121 and the backup storage unit 122 may both be one or more of a read-only memory (ROM) and a random access memory (RAM).
[0067] In this embodiment, when one or more instance storage units 121 fail to work properly, the storage unit module 120 can still provide a sufficient number of storage units for the instance computing unit 111 to read and write access, thereby realizing the failure repair function of the storage unit, improving the reliability of the read and write access between the computing unit and the storage unit, and thereby improving the yield of the stacked chip 200.
[0068] The computing unit and the storage unit are connected via an on-chip bus. In this embodiment, the on-chip bus module 130 is provided with at least one group of data channels 131. Each group of data channels 131 includes a corresponding instance channel 132 and a backup channel 133.
[0069] NOC is a commonly used solution for achieving connection between cores on a System on Chip (SoC). Specifically, an internationally open and universal bus structure can be selected, or an on-chip bus can be independently developed according to specific fields.
[0070] A common NOC may be a Wishbone bus or an Avalon bus, or may be at least any one of an address bus, a data bus, and a control bus.
[0071] In this embodiment, each computing unit in the computing unit module 110 is connected to each storage unit (the instance storage unit 121 and the backup storage unit 122) in the storage unit module 120 via a corresponding data channel 131 in the on-chip bus module 130. Figure 2 The figure shows a connection diagram of a computing unit in a computing unit module and a storage unit in a storage unit module provided by an embodiment of the present invention, wherein each computing unit is connected to all storage units (instance storage unit 121 and backup storage unit 122) in the storage unit module 120 through different data channels 131 in the on-chip bus. Each computing unit includes each instance storage unit 121 and backup storage unit 122 in the computing unit module 110; each storage unit includes each instance storage unit 121 and backup storage unit 122 in the storage unit module 120.
[0072] In this embodiment, when the instance channel 132 is valid, the instance computing unit 111 realizes read and write access to the corresponding instance storage unit 121 through the instance channel 132; and when the instance channel 132 fails, the instance computing unit 111 realizes read and write access to the corresponding instance storage unit 121 through the backup channel 133; the failure repair function of the data channel 131 is realized, the reliability of the read and write access between the computing unit and the storage unit is improved, and the yield of the stacked chip 200 is thereby improved.
[0073] The working principle of this embodiment is:
[0074] In this embodiment, the computing unit module 110 is provided with an instance computing unit 111 and a backup computing unit 112. When the instance computing unit 111 fails to work normally, the backup computing unit 112 can be used to replace the failed instance computing unit 111, so that the computing unit module 110 can work normally; the storage unit module 120 is provided with an instance storage unit 121 and a backup storage unit 122. When one or more instance storage units 121 fail to work normally, the storage unit module 120 can still provide a sufficient number of storage units for the computing unit to read and write access; at the same time, each group of data channels 131 in the on-chip bus module 130 is provided with an instance channel 132 and a backup channel 133. When an instance channel 132 in a data channel 131 fails to work normally, the corresponding backup channel 133 is used to replace the failed instance channel 132, so that the data channel 131 can work normally. This embodiment can independently implement failure repair of the computing unit, the instance storage unit 121 and the data channel 131, improves the reliability of read and write access between the computing unit and the instance storage unit 121, and thus improves the chip yield of the stacked chip 200.
[0075] In practical applications, a selector can be used to implement the construction of the data channel 131 in the on-chip bus module 130. Figure 3 FIG. 1 is a schematic diagram of the structure of a data channel provided by an embodiment of the present invention. The data channel 131 includes a first selector 134 and a second selector 135 .
[0076] The first selector 134 may be a DEMUX (data separation selector), and the second selector 135 may be a MUX (multiplexer, data mixing selector).
[0077] Specifically, the first output end of the first selector 134 is connected to the first input end of the second selector 135 to construct the instance channel 132 ; the second output end of the first selector 134 is connected to the second input end of the second selector 135 to construct the backup channel 133 .
[0078] Figure 3 In the example, the NI (network interface) connected to the input end of the first selector 134 is an entrance of the NOC, and the computing unit can access the NOC through the entrance; the NI connected to the output end of the second selector 135 is an exit of the NOC, and the instance storage unit 121 can be connected to the computing unit through the exit. Specifically, the NI can select an AXI (Advanced eXtensible Interface) interface bus.
[0079] In actual applications, in order to improve the failure repair efficiency of the operation unit, in this embodiment, different registers are configured for each operation unit in the operation unit module 110 for calling when repairing the failure of the operation unit.
[0080] The operation unit includes a status register, an enable register and a logic identity information register.
[0081] The status register is used to store information indicating whether the operation unit is in a normal state. For example, 0 is used to represent that the operation unit can work normally and is in a valid state; 1 is used to represent that the operation unit is not working normally and is in a failed state.
[0082] The enable register is used to store information representing whether the operation unit is in an activated state, for example: 0 is used to represent that the operation unit can be connected to the many-core computing circuit 100 and is in an activated state; 1 is used to represent that the operation unit cannot be connected to the many-core computing circuit 100 and is in an inactivated state.
[0083] The logical identity information register is used to store the logical identity information (Logic Identity Document, Logic ID) of the computing unit when the many-core computing circuit 100 is working. The Logic ID of the computing unit is used to be called by the controller in the many-core computing circuit 100 to implement the preset computing function. Corresponding to the Logic ID of the computing unit is the physical identity information (Physical Identity Document, Physical ID) of the computing unit, which can be a unique code to refer to the computing unit.
[0084] This embodiment can retrieve and / or modify the information in the above register to achieve failure repair of the operation unit.
[0085] Based on the same inventive concept as the method, an embodiment of the present invention further provides a stacked chip, such as Figure 4 FIG. 2 is a schematic diagram of the structure of the stacked chip. The stacked chip 200 may be provided with one or more of the many-core computing circuits 100 described above.
[0086] The computing unit module 110 in the many-core computing circuit 100 is set on the first wafer; the storage unit module 120 in the many-core computing circuit 100 is set on the second wafer; the on-chip bus module 130 in the many-core computing circuit 100 is set on the first wafer or the second wafer; the first wafer and the second wafer are stacked.
[0087] Of course, the on-chip bus module 130 can be disposed on the third wafer; in this case, the first wafer, the second wafer and the third wafer are stacked.
[0088] In the stacked chip 200, since the many-core computing circuit 100 therein can independently realize failure repair of the computing unit, the instance storage unit 121 and the data channel 131, the reliability of the read and write access between the computing unit and the storage unit is improved, thereby improving the chip yield of the stacked chip 200.
[0089] Based on the same inventive concept as the method, an embodiment of the present invention further provides a fault-tolerant control method, which is applied to any of the many-core computing circuits 100 described above. Figure 5 The flowchart of the embodiment of the method is shown, which specifically includes steps 11 to 16.
[0090] Step 11 , determining whether there is a failed instance computing unit 111 in the computing unit module 110 .
[0091] Specifically, the valid instance storage unit 121 can be used to return detection information to the target instance operation unit sequence. When the target instance operation unit sequence cannot receive the detection information or cannot correctly receive the detection information, it can be determined that the target instance operation unit sequence is invalid and cannot work normally.
[0092] Step 12 : If there is a failed instance computing unit 111 , the failed instance computing unit 111 is replaced by the backup computing unit 112 , and the instance computing unit sequence of the computing unit module 110 is updated.
[0093] In practical applications, a main control chip may be provided in the stacked chip 200 to obtain status information of each instance computing unit 111 in the many-core computing circuit 100, and configure these instance computing units 111 to form an instance computing unit sequence to implement the preset computing function of the many-core computing circuit 100. Of course, a self-check module may also be configured in the computing unit to obtain status information of the computing unit and configure the computing unit.
[0094] When one or more instance computing units 111 fail, this step can activate the backup computing unit 112 in the computing unit and cut out the failed instance computing unit 111, so that the backup computing unit 112 replaces the failed instance computing unit 111 to perform corresponding work, thereby ensuring the normal operation of the many-core computing circuit 100.
[0095] In practical applications, this embodiment implements the failure repair function of the operation unit by updating the operation unit mapping relationship, which specifically includes steps 21 and 22.
[0096] Step 21 , in the computing unit mapping relationship of the instance computing unit sequence, remove the physical identity information of the failed instance computing unit 111 , and add the physical identity information of the backup computing unit 112 .
[0097] Specifically, when the many-core computing circuit 100 is working, the logical identity information (Logic Identity Document, Logic ID) of the computing unit is used to be called by the controller in the many-core computing circuit 100 to implement the preset computing function. Corresponding to the Logic ID of the computing unit is the physical identity information (Physical Identity Document, Physical ID) of the computing unit, which can be a unique code to refer to the computing unit.
[0098] The operation unit mapping relationship may include a mapping relationship between the Physical ID and the Logic ID of each operation unit in the example operation unit sequence, and the controller may call the actual operation unit (Physical ID) through the Logic ID of the operation unit.
[0099] In this step, the Physical ID of the failed instance computing unit 111 is removed, and the Physical ID of the backup computing unit 112 is added, so that the operation of using the backup computing unit 112 to replace the failed instance computing unit 111 is realized.
[0100] Step 22: Update the computing unit mapping relationship according to the physical identity information of the existing computing units in the computing unit mapping relationship.
[0101] The operation unit mapping relationship includes a mapping relationship between physical identity information and logical identity information of each operation unit in the instance operation unit sequence.
[0102] Specifically, after step 21, the operation units in the instance operation unit sequence have changed, so after the Physical IDs of the existing operation units in the operation unit sequence are assigned Logic IDs, the operation unit mapping relationship can be updated.
[0103] Here, this embodiment provides two updating solutions for the mapping relationship of the operation units.
[0104] Option 1 includes:
[0105] Step 31 : Replace the physical identity information of the failed instance computing unit 111 in the computing unit mapping relationship before updating with the physical identity information of the backup computing unit 112 , to obtain the updated computing unit mapping relationship.
[0106] Specifically, in this step, only the part of the operation unit mapping relationship involving the failed instance operation unit 111 is changed, and the Physical ID of the failed instance operation unit 111 is directly replaced with the PhysicalID of the backup operation unit 112. The Logic ID corresponding to the failed instance operation unit 111 in the operation unit mapping relationship remains unchanged, and the Physical ID and Logic ID of other parts remain unchanged, thereby completing the update of the operation unit mapping relationship.
[0107] like Figure 6 FIG. 2 is a schematic diagram of a computing unit mapping relationship when all the computing units provided in the example of the present invention are valid. Figure 7 Shown Figure 6Schematic diagram of the operation unit mapping relationship after there is a failed instance operation unit in the.
[0108] Option 2 includes:
[0109] Step 41, using the existing physical identity information of the computing unit, matching the logical identity information of each computing unit preset in the computing unit mapping relationship, to obtain the updated computing unit mapping relationship.
[0110] Specifically, in this step, some or all of the instance operation units 111 in the operation unit mapping relationship are changed, and the Physical ID corresponding to the existing operation units in the instance operation unit sequence (that is, the instance operation unit sequence after removing the failed instance operation unit 111 and adding the backup operation unit 112) is re-matched with the preset Logic ID of each operation unit, thereby realizing the update of the operation unit mapping relationship.
[0111] like Figure 8 Shown Figure 6 Schematic diagram of the operation unit mapping relationship after there is a failed instance operation unit in the.
[0112] Step 13 , determining whether there is a failed instance storage unit 121 in the storage unit module 120 .
[0113] Specifically, a valid computing unit may be used to send stimulus information to a target instance storage unit. When the target instance storage unit cannot receive the stimulus information or cannot correctly respond to the stimulus information, the target instance storage unit may be deemed to be invalid and unable to work normally.
[0114] Step 14, if there is a failed instance storage unit 121, use the backup storage unit 122 in the storage unit module 120 to replace it, update the instance storage unit sequence corresponding to the instance operation unit sequence, and reorganize the storage space with continuous logical addresses.
[0115] Specifically, this step can skip the failed instance storage unit 121, so that the computing unit does not access the failed instance storage unit 121 when working.
[0116] Here, this embodiment provides a specific implementation process of step 14, including steps 51 to 52.
[0117] Step 51 : Acquire the physical identity information of the failed instance storage unit 121 .
[0118] Specifically, when the many-core computing circuit 100 is working, the logical identity information (Logic Identity Document, Logic ID) of the instance storage unit 121 and the backup storage unit 122 is used to be called by the controller in the many-core computing circuit 100 to implement the read and write access function between the computing unit and the instance storage unit 121. Corresponding to the LogicID of the instance storage unit 121 is the physical identity information (Physical Identity Document, PhysicalID) of the instance storage unit 121, which can be a unique code to refer to the instance storage unit 121.
[0119] The storage unit mapping relationship may include a mapping relationship between the PhysicalID and the Logic ID of each instance storage unit 121 in the instance storage unit sequence. The controller may call the actual instance storage unit 121 (Physical ID) through the Logic ID of the instance storage unit 121 .
[0120] Step 52, when mapping the physical identity information of the storage units in the instance storage unit sequence to the storage unit mapping relationship of the instance storage unit sequence, skip the physical identity information of the failed instance storage unit 121 to use the physical identity information of the valid backup storage unit 122 to generate the storage unit mapping relationship.
[0121] The storage unit mapping relationship includes a mapping relationship between physical identity information and logical identity information of each instance storage unit 121 and backup storage unit 122 in the storage unit module 120 .
[0122] Specifically, when generating the storage unit mapping relationship, this step first removes the Physical ID of the failed instance storage unit 121 and skips the Physical ID of the failed instance storage unit 121, so that the computing unit does not access the failed instance storage unit 121 when working.
[0123] Step 15 , determining whether there is a failed instance channel 132 in each group of data channels 131 of the on-chip bus module 130 .
[0124] Specifically, when the instance channel 132 cannot transmit data or cannot transmit data correctly, the instance channel 132 is deemed to be invalid.
[0125] Step 16: If there is a failed instance channel 132 , the failed instance channel 132 is replaced by the backup channel 133 corresponding to the failed instance channel 132 , and the data channel 131 corresponding to the failed instance channel 132 is updated.
[0126] Specifically, it can be based on Figure 3 The data channel 131 shown is used to implement the switching between the instance channel 132 and the backup channel 133 .
[0127] The technical solution provided in the embodiments of the present invention has at least the following technical effects or advantages:
[0128] In the embodiment of the present invention, the computing unit module is provided with an instance computing unit and a backup computing unit. When the instance computing unit fails to work properly, the backup computing unit can be used to replace the failed instance computing unit, so that the computing unit module can work normally; the storage unit module is provided with an instance storage unit and a backup storage unit. When one or more instance storage units fail to work properly, the storage unit module can still provide a sufficient number of storage units for the computing unit to read and write access; at the same time, each group of data channels in the on-chip bus module is provided with an instance channel 132 and a backup channel. When the instance channel in the data channel fails to work properly, the corresponding backup channel is used to replace the failed instance channel, so that the data channel works normally. The embodiment of the present invention can independently realize the failure repair of the computing unit, the storage unit and the data channel, improve the reliability of the read and write access between the computing unit and the storage unit, and thus improve the chip yield of the multi-core chip structure.
[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0130] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0131] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0132] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0133] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.
[0134] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A fault-tolerant control method for a many-core computing circuit, characterized in that: include: Determine whether there is a failed instance computing unit in the computing unit module; If there is a failed instance computing unit, the failed instance computing unit is replaced by a backup computing unit, and the instance computing unit sequence of the computing unit module is updated; Determine whether there is a failed instance storage unit in the storage unit module; If there is a failed instance storage unit, the failed instance storage unit is replaced by a backup storage unit in the storage unit module, and the instance storage unit sequence corresponding to the instance operation unit sequence is updated to reorganize a storage space with a continuous logical address; Determine whether there is a failed instance channel in each group of data channels of the on-chip bus module; If there is a failed instance channel, the failed instance channel is replaced by a backup channel corresponding to the failed instance channel, and the data channel corresponding to the failed instance channel is updated; The step of replacing the failed instance computing unit with a backup computing unit and updating the instance computing unit sequence of the computing unit module includes: In the computing unit mapping relationship of the instance computing unit sequence, the physical identity information of the failed instance computing unit is removed, and the physical identity information of the backup computing unit is added; The operation unit mapping relationship is updated according to the physical identity information of the existing operation units in the operation unit mapping relationship; wherein the operation unit mapping relationship includes a mapping relationship between the physical identity information and the logical identity information of each operation unit in the instance operation unit sequence.
2. The fault-tolerant control method according to claim 1, characterized in that: The updating of the computing unit mapping relationship according to the physical identity information of the existing computing unit in the computing unit mapping relationship includes: Replacing the physical identity information of the failed instance computing unit in the computing unit mapping relationship before updating with the physical identity information of the backup computing unit to obtain the updated computing unit mapping relationship; or, The physical identity information of the existing computing units is used to match the logical identity information of each computing unit preset in the computing unit mapping relationship to obtain the updated computing unit mapping relationship.
3. The fault-tolerant control method according to claim 1, characterized in that: The method of replacing the failed instance storage unit with a backup storage unit in the storage unit module and updating the instance storage unit sequence corresponding to the instance computing unit sequence includes: Obtaining physical identity information of the failed instance storage unit; When mapping the physical identity information of the storage units in the instance storage unit sequence to the storage unit mapping relationship of the instance storage unit sequence, the physical identity information of the failed instance storage unit is skipped to generate the storage unit mapping relationship by using the physical identity information of the valid instance storage unit; wherein the storage unit mapping relationship includes a mapping relationship between the physical identity information and the logical identity information of each instance storage unit in the storage unit module.
4. A many-core computing circuit with fault tolerance function, the many-core computing circuit is used to implement the fault tolerance control method of the many-core computing circuit according to any one of claims 1 to 3, characterized in that: include: A computing unit module, comprising at least one instance computing unit and at least one backup computing unit; A storage unit module, comprising at least one instance storage unit and at least one backup storage unit; The on-chip bus module includes at least one group of data channels; wherein each group of data channels includes a correspondingly set instance channel and a backup channel; Each computing unit in the computing unit module is connected to each storage unit in the storage unit module through the corresponding data channel in the on-chip bus module; wherein each computing unit includes at least one instance computing unit and at least one backup computing unit; each storage unit includes at least one instance storage unit and at least one backup storage unit.
5. The many-core computing circuit according to claim 4, characterized in that: The data channel comprises: a first selector and a second selector; The first output terminal of the first selector is connected to the first input terminal of the second selector to construct the instance channel; The second output terminal of the first selector is connected to the second input terminal of the second selector to construct the backup channel.
6. The many-core computing circuit according to claim 4, characterized in that: The computing unit comprises: A status register, used to store information indicating whether the operation unit is in a normal state; An enable register, used to store information indicating whether the computing unit is in an activated state; The logic identity information register is used to store the logic identity information representing the operation unit when the many-core computing circuit is working.
7. The many-core computing circuit according to any one of claims 4 to 6, characterized in that: The example computing unit includes one or more of a central processing unit, a digital signal processing chip, a graphics processor, and an intellectual property core module; The backup computing unit includes one or more of a central processing unit, a digital signal processing chip, a graphics processor and an intellectual property core module.
8. The many-core computing circuit according to any one of claims 4 to 6, characterized in that: Each storage unit in the storage unit module includes one or more of a read-only memory and a random access memory.
9. A stacked chip, characterized in that: The many-core chip comprises the many-core computing circuit as claimed in any one of claims 4 to 8; The computing unit module in the multi-core computing circuit is arranged on a first wafer; the storage unit module in the multi-core computing circuit is arranged on a second wafer; The on-chip bus module in the many-core computing circuit is arranged on the first wafer or the second wafer; The first wafer and the second wafer are stacked.
10. The stacked chip according to claim 9, characterized in that: The on-chip bus module is arranged on the third wafer; The first wafer, the second wafer and the third wafer are stacked.
Citation Information
Patent Citations
Self-test self-repair mechanism and structure of multi-operation unit / multi-core / many-core system
CN101751317A
Big data operation acceleration system and chip
CN108536642A
Many-core computing circuit and stacked chip
CN215910890U
Method for self-testing and self-repairing on chip
WO2010066207A1
