DRAM-PIM computing architecture and computing system for multi-level computing granularity

Abstract

The invention relates to a DRAM-PIM (Dynamic Random Access Memory-Personal Information Modeling) computing architecture and a computing system for multi-level computing granularity. The DRAM-PIM computing architecture of the multi-level computing granularity comprises a plurality of different storage and computation fusion units; the plurality of different storage and calculation fusion units are used for executing storage and calculation tasks with different storage and calculation levels; wherein the storage and calculation tasks with different storage and calculation levels correspond to different calculation granularities. According to the DRAM-PIM computing architecture for multi-level computing granularity provided by the invention, a plurality of different storage and calculation fusion units are arranged and can be used for executing storage and calculation tasks with different computing granularities, and compared with the prior art that the computing granularity of the computing architecture is fixed and only a part of storage and calculation tasks corresponding to the computing granularity can be executed, the computing granularity of the computing architecture is fixed, so that the computing efficiency is improved. According to the DRAM-PIM computing architecture of the multi-level computing granularity provided by the invention, the storage and calculation tasks of various computing granularities can be executed, and the corresponding storage and calculation fusion units are flexibly adopted to execute the storage and calculation tasks according to different granularities of the storage and calculation tasks, so that the availability of the computing architecture is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a DRAM-PIM computing architecture and computing system with multi-level computing granularity. Background Art

[0002] With the rapid development of artificial intelligence technology, the rapid increase in algorithm complexity and data volume has put forward higher requirements for computing systems. Most computing systems use a storage-computing integrated computing architecture. The core idea of ​​this storage-computing integrated computing architecture is to integrate computing power directly into the storage unit, significantly improving data processing efficiency and overall system energy efficiency by reducing the distance of data migration.

[0003] In the prior art, the computing granularity of the integrated storage and computing architecture is mostly determined according to usage requirements during the design process of the integrated storage and computing architecture.

[0004] However, the computing granularity of such a storage-computing integrated computing architecture is fixed and cannot be applied to different types of computing tasks, resulting in low availability of the computing architecture. Summary of the invention

[0005] Based on this, it is necessary to provide a DRAM-PIM computing architecture and computing system with multi-level computing granularity and high availability to address the above technical issues.

[0006] In a first aspect, the present application provides a DRAM-PIM computing architecture with multi-level computing granularity, the computing architecture comprising a plurality of different storage-computing fusion units;

[0007] The multiple different storage and computing fusion units are used to execute storage and computing tasks at different storage and computing levels;

[0008] Among them, the computing granularity corresponding to the computing tasks of different computing levels is different.

[0009] In one embodiment, the storage and computing fusion unit includes a first storage and computing fusion unit, a second storage and computing fusion unit and a third storage and computing fusion unit; the first storage and computing fusion unit is used to execute a first storage and computing task of a first storage and computing level; the second storage and computing fusion unit is used to execute a second storage and computing task of a second storage and computing level; the third storage and computing fusion unit is used to execute a third storage and computing task of a third storage and computing level; wherein the computing granularity corresponding to the first storage and computing level is smaller than the computing granularity corresponding to the second storage and computing level, and the computing granularity corresponding to the second storage and computing level is smaller than the computing granularity corresponding to the third storage and computing level.

[0010] In one of the embodiments, the first storage and computing fusion unit is specifically used to store the first data of the first storage and computing task and perform computing on the first data.

[0011] In one embodiment, the first storage and computing fusion unit includes multiple first storage units and multiple first computing units corresponding to the multiple first storage units one by one; the first storage unit is used to store the first data; the first computing unit is used to obtain the first data from the first storage unit and perform computing on the first data.

[0012] In one of the embodiments, the second storage-computing fusion unit is specifically used to store the second data of the second storage-computing task and perform computation on the second data.

[0013] In one embodiment, the second storage and computing fusion unit includes multiple second storage units and multiple second computing units corresponding to the multiple second storage units one by one, and each of the second storage units includes multiple first storage units; the second storage unit is used to store the second data; the second computing unit is used to obtain the second data from the second storage unit and perform computing on the second data.

[0014] In one of the embodiments, the third storage and computing fusion unit is specifically used to store the third data of the third storage and computing task and perform computing on the third data.

[0015] In one embodiment, the third storage and computing fusion unit includes the first storage and computing fusion unit, the second storage and computing fusion unit and a third computing unit; the first storage and computing fusion unit and the second storage and computing fusion unit are used to store the third data when the storage and computing task is the third storage and computing task; the third computing unit is used to obtain the third data from the first storage and computing fusion unit and the second storage and computing fusion unit, and perform computing on the third data.

[0016] In one embodiment, the third storage and computing fusion unit includes the first storage and computing fusion unit, the second storage and computing fusion unit and a third computing unit; the first storage and computing fusion unit and the second storage and computing fusion unit are used to store the third data and perform computing processing on the third data when the storage and computing task is the third storage and computing task to obtain a first processing result and a second processing result; the third computing unit is used to perform data aggregation processing on the first processing result and the second processing result.

[0017] In a second aspect, the present application provides a computing system, which includes a DRAM-PIM computing architecture with multi-level computing granularity as described in any embodiment of the first aspect.

[0018] The above-mentioned multi-level computing granularity DRAM-PIM computing architecture and computing system, the multi-level computing granularity DRAM-PIM computing architecture includes a plurality of different storage and computing fusion units; the plurality of different storage and computing fusion units are used to execute storage and computing tasks of different storage and computing levels; wherein the storage and computing tasks of different storage and computing levels correspond to different computing granularities. The multi-level computing granularity DRAM-PIM computing architecture provided in the present application is provided with a plurality of different storage and computing fusion units, which can be used to execute storage and computing tasks of different computing granularities. Compared with the computing architecture in the prior art, which has a fixed computing granularity and can only execute part of the storage and computing tasks corresponding to its computing granularity, the multi-level computing granularity DRAM-PIM computing architecture provided in the present application can execute storage and computing tasks of various computing granularities, and flexibly adopt the corresponding storage and computing fusion units to execute the storage and computing tasks according to the different granularities of the storage and computing tasks, thereby effectively improving the availability of the multi-level computing granularity DRAM-PIM computing architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A structural block diagram of a DRAM-PIM computing architecture with multi-level computing granularity in one embodiment;

[0021] Figure 2 A structural block diagram of a DRAM-PIM computing architecture with multi-level computing granularity in another embodiment;

[0022] Figure 3 A structural block diagram of a DRAM-PIM computing architecture with multi-level computing granularity in another embodiment. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0024] With the rapid development of artificial intelligence technology, the rapid increase in algorithm complexity and data volume has put forward higher requirements for computing systems. Most computing systems use a storage-computing integrated computing architecture. The core idea of ​​this storage-computing integrated computing architecture is to integrate computing power directly into the storage unit, significantly improving data processing efficiency and overall system energy efficiency by reducing the distance of data migration.

[0025] In the prior art, the computing granularity of the storage-computing integrated computing architecture is mostly determined according to the use requirements during the design process of the storage-computing integrated computing architecture. That is, the designed storage-computing integrated computing architecture is only suitable for executing fixed storage-computing tasks corresponding to its computing granularity.

[0026] However, in actual application scenarios, if the storage-computing integrated computing architecture is used to perform storage-computing tasks with a lower computing granularity than the original one, it will lead to waste of resources. If the storage-computing integrated computing architecture is used to perform storage-computing tasks with a higher computing granularity than the original one, it may not be able to meet the usage requirements. Therefore, the computing granularity of such a storage-computing integrated computing architecture is fixed and cannot be applied to different types of computing tasks, which in turn leads to low availability of the computing architecture.

[0027] In view of this, the present application provides a DRAM-PIM computing architecture with multi-level computing granularity, which includes multiple different storage-computing fusion units, and different storage-computing fusion units can be used to execute storage-computing tasks with different computing granularities. That is, the computing granularity of the DRAM-PIM computing architecture with multi-level computing granularity is not fixed, and can be used to execute storage-computing tasks with various computing granularities, thereby effectively improving the availability of the DRAM-PIM computing architecture with multi-level computing granularity.

[0028] In an exemplary embodiment, Figure 1 As shown, a DRAM-PIM computing architecture 100 with multi-level computing granularity is provided. The DRAM-PIM computing architecture 100 with multi-level computing granularity includes a plurality of different storage-computing fusion units 101 .

[0029] Among them, the multiple different storage and computing fusion units 101 are used to execute storage and computing tasks with different storage and computing levels; the storage and computing tasks with different storage and computing levels correspond to different computing granularities.

[0030] Optionally, the DRAM-PIM computing architecture 100 refers to a storage-computing integrated computing architecture. The storage-computing integrated computing architecture is a computing architecture that directly integrates computing processing capabilities into the storage unit, which can greatly reduce the delay and energy consumption caused by data migration.

[0031] The storage-computing fusion unit 101 refers to a basic component for integrating computing and storage functions to form a storage-computing integrated computing architecture.

[0032] The computational granularity refers to the degree of subdivision of the task when performing the computational task. It can be used to describe the size and frequency of the computational task being decomposed into smaller units. The computational granularity can include coarse granularity and fine granularity. The coarse granularity computational task can be a large-scale matrix multiplication. The fine granularity computational task can be pixel-level image processing.

[0033] In one possible implementation, a plurality of different storage-computing fusion units 101 may be provided in the computing architecture 100. The plurality of different storage-computing fusion units 101 may be used to execute storage-computing tasks of different storage-computing levels. Storage-computing tasks of different storage-computing levels have corresponding computing granularities that are also different.

[0034] Exemplarily, assuming that two different storage and computing fusion units 101 can be set in the computing architecture 100, for example, a first storage and computing fusion unit and a second storage and computing fusion unit, the first storage and computing fusion unit and the second storage and computing fusion unit can be used to execute storage and computing tasks at two different storage and computing levels, for example, storage and computing tasks at the first storage and computing level and storage and computing tasks at the second storage and computing level, then the computing granularity corresponding to the storage and computing tasks at the first storage and computing level is different from the computing granularity corresponding to the storage and computing tasks at the second storage and computing level.

[0035] Exemplarily, assuming that four different storage and computing fusion units 101 can be provided in the computing architecture 100, for example, a first storage and computing fusion unit, a second storage and computing fusion unit, a third storage and computing fusion unit and a fourth storage and computing fusion unit, then the first storage and computing fusion unit, the second storage and computing fusion unit, the third storage and computing fusion unit and the fourth storage and computing fusion unit can be used to execute storage and computing tasks of four different storage and computing levels, for example, storage and computing tasks of the first storage and computing level, storage and computing tasks of the second storage and computing level, storage and computing tasks of the third storage and computing level and storage and computing tasks of the fourth storage and computing level, then the computing granularity corresponding to the storage and computing tasks of the first storage and computing level, the computing granularity corresponding to the storage and computing tasks of the second storage and computing level, the computing granularity corresponding to the storage and computing tasks of the third storage and computing level and the computing granularity corresponding to the storage and computing tasks of the fourth storage and computing level are all different.

[0036] The above-mentioned multi-level computing granularity DRAM-PIM computing architecture includes a plurality of different storage-computing fusion units; the plurality of different storage-computing fusion units are used to execute storage-computing tasks of different storage-computing levels; wherein the storage-computing tasks of different storage-computing levels correspond to different computing granularities. The multi-level computing granularity DRAM-PIM computing architecture provided in the present application is provided with a plurality of different storage-computing fusion units, which can be used to execute storage-computing tasks of different computing granularities. Compared with the computing architecture in the prior art in which the computing granularity is fixed and only part of the storage-computing tasks corresponding to the computing granularity can be executed, the multi-level computing granularity DRAM-PIM computing architecture provided in the present application can execute storage-computing tasks of various computing granularities, and flexibly adopt the corresponding storage-computing fusion units to execute the storage-computing tasks according to the different granularities of the storage-computing tasks, thereby effectively improving the availability of the multi-level computing granularity DRAM-PIM computing architecture.

[0037] In an exemplary embodiment, Figure 2 As shown, the storage and computing fusion unit 101 includes a first storage and computing fusion unit 1011 , a second storage and computing fusion unit 1012 and a third storage and computing fusion unit 1013 .

[0038] Among them, the first storage computing fusion unit 1011 is used to execute the first storage computing task of the first storage computing level; the second storage computing fusion unit 1012 is used to execute the second storage computing task of the second storage computing level; the third storage computing fusion unit 1013 is used to execute the third storage computing task of the third storage computing level; the computing granularity corresponding to the first storage computing level is smaller than the computing granularity corresponding to the second storage computing level, and the computing granularity corresponding to the second storage computing level is smaller than the computing granularity corresponding to the third storage computing level.

[0039] In one possible implementation, the storage and computing fusion unit 101 may include a first storage and computing fusion unit 1011, a second storage and computing fusion unit 1012 and a third storage and computing fusion unit 1013. As described above, different storage and computing fusion units are used to execute storage and computing tasks of different storage and computing levels. The first storage and computing fusion unit 1011 is used to execute a first storage and computing task of a first storage and computing level, the second storage and computing fusion unit 1012 is used to execute a second storage and computing task of a second storage and computing level, and the third storage and computing fusion unit 1013 is used to execute a third storage and computing task of a third storage and computing level.

[0040] Furthermore, the computing granularities corresponding to the computing tasks of different computing levels are different. The computing granularity corresponding to the first computing level is smaller than the computing granularity corresponding to the second computing level, and the computing granularity corresponding to the second computing level is smaller than the computing granularity corresponding to the third computing level. That is, the computing granularity corresponding to the computing tasks of the first computing level is the smallest, the computing granularity corresponding to the computing tasks of the second computing level is medium, and the computing granularity corresponding to the computing tasks of the third computing level is the largest.

[0041] In an exemplary embodiment, the first storage and computing fusion unit is specifically used to store the first data of the first storage and computing task and perform computing processing on the first data.

[0042] In an exemplary embodiment, Figure 3 As shown, the first storage and computing fusion unit 1011 includes a plurality of first storage units 201 and a plurality of first computing units 202 corresponding one-to-one to the plurality of first storage units 201 .

[0043] The first storage unit 201 is used to store the first data; the first calculation unit 202 is used to obtain the first data from the first storage unit 201 and perform calculation processing on the first data.

[0044] In one possible implementation, when the storage-computing task is a first storage-computing task, the first data of the first storage-computing task can be stored in multiple first storage units 201 in the first storage-computing fusion unit 1011 in the DRAM-PIM computing architecture 100 with multi-level computing granularity. When the first data needs to be processed, the multiple first computing units 202 in the first storage-computing fusion unit 1011 will read the first data from the corresponding first storage unit 201 and perform computing processing.

[0045] In an exemplary embodiment, the second storage-computing fusion unit is specifically used to store the second data of the second storage-computing task and perform computation on the second data.

[0046] In an exemplary embodiment, Figure 3 As shown, the second storage and computing fusion unit 1012 includes multiple second storage units 301 and multiple second computing units 302 corresponding to the multiple second storage units 301 one by one, and each of the second storage units 301 includes multiple first storage units 201.

[0047] The second storage unit 301 is used to store the second data; the second calculation unit 302 is used to obtain the second data from the second storage unit 301 and perform calculation processing on the second data.

[0048] In one possible implementation, when the storage-computing task is a second storage-computing task, the second data of the second storage-computing task can be stored in multiple second storage units 301 in the second storage-computing fusion unit 1012 in the DRAM-PIM computing architecture 100 with multi-level computing granularity. When the second data needs to be processed, the multiple second computing units 302 in the second storage-computing fusion unit 1012 will read the second data from the corresponding second storage unit 301 and perform computing processing.

[0049] As described above, the second storage unit includes multiple first storage units, that is, it can be understood that in the first storage and computing fusion unit, each first storage unit corresponds to a first computing unit, and in the second storage and computing fusion unit, multiple first storage units constitute a second storage unit, and each second storage unit corresponds to a second computing unit, that is, in the second storage and computing fusion unit, multiple first storage units correspond to one second computing unit.

[0050] In an exemplary embodiment, the third storage and computing fusion unit is specifically used to store the third data of the third storage and computing task and perform computing on the third data.

[0051] In an exemplary embodiment, Figure 3 As shown, the third storage and computing fusion unit 1013 includes the first storage and computing fusion unit 1011 , the second storage and computing fusion unit 1012 and the third computing unit 401 .

[0052] Among them, the first storage and computing fusion unit 1011 and the second storage and computing fusion unit 1012 are used to store the third data when the storage and computing task is the third storage and computing task; the third computing unit 401 is used to obtain the third data from the first storage and computing fusion unit 1011 and the second storage and computing fusion unit 1012, and perform computing processing on the third data.

[0053] In one possible implementation, when the storage and computing task is the third storage and computing task, the third data of the third storage and computing task can be stored in the first storage and computing fusion unit 1011 and the second storage and computing fusion unit 1012 in the DRAM-PIM computing architecture 100 with multi-level computing granularity. When the third data needs to be processed, the third computing unit 401 in the third storage and computing fusion unit 1013 reads the third data from the first storage and computing fusion unit 1011 and the second storage and computing fusion unit 1012, and performs computing processing.

[0054] That is, when the storage-computing task is the third storage-computing task, the first storage-computing fusion unit and the second storage-computing fusion unit can be used only to store the third data, and the computing unit of the third storage-computing fusion unit performs computing and processing on the third data. As described above, the first storage-computing fusion unit includes multiple first storage units, the second storage-computing fusion unit also includes multiple first storage units, and the third storage-computing fusion unit includes only one computing unit, that is, one computing unit calculates the data stored in multiple first storage units.

[0055] In an exemplary embodiment, Figure 3 As shown, the third storage and computing fusion unit 1013 includes the first storage and computing fusion unit 1011 , the second storage and computing fusion unit 1012 and the third computing unit 401 .

[0056] Among them, the first storage and computing fusion unit 1011 and the second storage and computing fusion unit 1012 are used to store the third data and perform calculation processing on the third data to obtain the first processing result and the second processing result when the storage and computing task is the third storage and computing task; the third calculation unit 401 is used to perform data aggregation processing on the first processing result and the second processing result.

[0057] In another possible implementation, when the storage and computing task is the third storage and computing task, the third data of the third storage and computing task can be stored in the first storage and computing fusion unit 1011 and the second storage and computing fusion unit 1012 in the DRAM-PIM computing architecture 100 with multi-level computing granularity. When the third data needs to be processed, the first storage and computing fusion unit 1011 performs computational processing on the stored data to obtain a first processing result, and the second storage and computing fusion unit 1012 performs computational processing on the stored data to obtain a second processing result, and the third computing unit 401 in the third fusion unit 1013 only needs to obtain the first processing result and the second processing result, and aggregate the first processing result and the second processing result to obtain the calculation result of the third storage and computing task.

[0058] In an exemplary embodiment, the present application also provides a DRAM-PIM computing system with multi-level computing granularity, and the DRAM-PIM computing system with multi-level computing granularity includes the DRAM-PIM computing architecture with multi-level computing granularity of any of the above embodiments.

[0059] In one embodiment, the multi-level computing granularity DRAM-PIM computing architecture includes multiple different storage and computing fusion units; the multiple different storage and computing fusion units are used to execute storage and computing tasks of different storage and computing levels; wherein the storage and computing tasks of different storage and computing levels correspond to different computing granularities.

[0060] In one embodiment, the storage and computing fusion unit includes a first storage and computing fusion unit, a second storage and computing fusion unit and a third storage and computing fusion unit; the first storage and computing fusion unit is used to execute a first storage and computing task of a first storage and computing level; the second storage and computing fusion unit is used to execute a second storage and computing task of a second storage and computing level; the third storage and computing fusion unit is used to execute a third storage and computing task of a third storage and computing level; wherein the computing granularity corresponding to the first storage and computing level is smaller than the computing granularity corresponding to the second storage and computing level, and the computing granularity corresponding to the second storage and computing level is smaller than the computing granularity corresponding to the third storage and computing level.

[0061] In one embodiment, the first storage and computing fusion unit is specifically used to store the first data of the first storage and computing task and perform computing processing on the first data.

[0062] In one embodiment, the first storage and computing fusion unit includes multiple first storage units and multiple first computing units corresponding to the multiple first storage units one by one; the first storage unit is used to store the first data; the first computing unit is used to obtain the first data from the first storage unit and perform computing on the first data.

[0063] In one embodiment, the second storage and computing fusion unit is specifically used to store the second data of the second storage and computing task and perform computing processing on the second data.

[0064] In one embodiment, the second storage and computing fusion unit includes multiple second storage units and multiple second computing units corresponding to the multiple second storage units one by one, and each of the second storage units includes multiple first storage units; the second storage unit is used to store the second data; the second computing unit is used to obtain the second data from the second storage unit and perform computing on the second data.

[0065] In one embodiment, the third storage and computing fusion unit is specifically used to store the third data of the third storage and computing task and perform computing processing on the third data.

[0066] In one embodiment, the third storage and computing fusion unit includes the first storage and computing fusion unit, the second storage and computing fusion unit and a third computing unit; the first storage and computing fusion unit and the second storage and computing fusion unit are used to store the third data when the storage and computing task is the third storage and computing task; the third computing unit is used to obtain the third data from the first storage and computing fusion unit and the second storage and computing fusion unit, and perform computing on the third data.

[0067] In one embodiment, the third storage and computing fusion unit includes the first storage and computing fusion unit, the second storage and computing fusion unit and a third computing unit; the first storage and computing fusion unit and the second storage and computing fusion unit are used to store the third data and perform computing processing on the third data when the storage and computing task is the third storage and computing task to obtain a first processing result and a second processing result; the third computing unit is used to perform data aggregation processing on the first processing result and the second processing result.

[0068] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0069] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A DRAM-PIM computing architecture with multi-level computing granularity, characterized in that: The computing architecture includes a plurality of different storage and computing fusion units; The multiple different storage and computing fusion units are used to execute storage and computing tasks at different storage and computing levels; Among them, the computing tasks with different computing levels correspond to different computing granularities.

2. The DRAM-PIM computing architecture with multi-level computing granularity according to claim 1, characterized in that: The storage and computing fusion unit includes a first storage and computing fusion unit, a second storage and computing fusion unit and a third storage and computing fusion unit; The first storage and computing fusion unit is used to execute a first storage and computing task of a first storage and computing level; the second storage and computing fusion unit is used to execute a second storage and computing task of a second storage and computing level; the third storage and computing fusion unit is used to execute a third storage and computing task of a third storage and computing level; Among them, the calculation granularity corresponding to the first storage computing level is smaller than the calculation granularity corresponding to the second storage computing level, and the calculation granularity corresponding to the second storage computing level is smaller than the calculation granularity corresponding to the third storage computing level.

3. The DRAM-PIM computing architecture with multi-level computing granularity according to claim 2, characterized in that: The first storage and computing fusion unit is specifically used to store the first data of the first storage and computing task and perform computing processing on the first data.

4. The multi-level computing granularity DRAM-PIM computing architecture according to claim 3, characterized in that: The first storage and computing fusion unit includes a plurality of first storage units and a plurality of first computing units corresponding one-to-one to the plurality of first storage units; The first storage unit is used to store the first data; The first calculation unit is used to obtain the first data from the first storage unit and perform calculation processing on the first data.

5. The DRAM-PIM computing architecture with multi-level computing granularity according to claim 2, characterized in that: The second storage and computing fusion unit is specifically used to store the second data of the second storage and computing task and perform computing processing on the second data.

6. The DRAM-PIM computing architecture with multi-level computing granularity according to claim 5, characterized in that: The second storage and calculation fusion unit includes a plurality of second storage units and a plurality of second calculation units corresponding to the plurality of second storage units one by one, and each of the second storage units includes a plurality of the first storage units; The second storage unit is used to store the second data; The second calculation unit is used to obtain the second data from the second storage unit and perform calculation processing on the second data.

7. The multi-level computing granularity DRAM-PIM computing architecture according to claim 2, characterized in that: The third storage and computing fusion unit is specifically used to store the third data of the third storage and computing task and perform computing processing on the third data.

8. The DRAM-PIM computing architecture with multi-level computing granularity according to claim 7, characterized in that: The third storage and computing fusion unit includes the first storage and computing fusion unit, the second storage and computing fusion unit and a third computing unit; The first storage-computing fusion unit and the second storage-computing fusion unit are configured to store the third data when the storage-computing task is the third storage-computing task; The third computing unit is used to obtain the third data from the first storage and computing fusion unit and the second storage and computing fusion unit, and perform computing processing on the third data.

9. The DRAM-PIM computing architecture with multi-level computing granularity according to claim 7, characterized in that: The third storage and computing fusion unit includes the first storage and computing fusion unit, the second storage and computing fusion unit and a third computing unit; The first storage-computation fusion unit and the second storage-computation fusion unit are used to store the third data and perform calculation processing on the third data to obtain a first processing result and a second processing result when the storage-computation task is the third storage-computation task; The third computing unit is used to perform data aggregation processing on the first processing result and the second processing result.

10. A DRAM-PIM computing system with multi-level computing granularity, characterized in that: The computing system includes a DRAM-PIM computing architecture with multi-level computing granularity as described in any one of claims 1 to 9.

Images