PIM compiler designed through parameters

Through the PIM compiler designed with parameter, the parameters are used to describe the in-memory computing unit matrix and adjust the circuit parameters, which solves the problems of slow compilation speed and high design cost of existing compilers, realizes low-power and low-cost compilation design, and improves the design efficiency of in-memory computing chips.

CN119513026APending Publication Date: 2025-02-25BEIJING PINGXIN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311068887.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-08-24
Publication Date
2025-02-25

AI Technical Summary

Technical Problem

During the compilation process, existing memory compilers have problems such as the compilation speed to be improved and the design cost is high, and there is a lack of compilation tools that meet the design of in-memory computing chips.

Method used

Design a PIM compiler designed through parameter design, set up the in-memory computing main module, use the parameter description in-memory computing unit matrix, adjust the circuit parameters to realize a specific optimization solution supported by multi-scale capacity and computing power, and meet the needs of different PIM IP design tasks.

Benefits of technology

It realizes low-power and low-cost compilation design, improves the overall design efficiency of in-memory computing chips, makes up for the gap in the compiler market, and solves the problems of existing compilers that lack generality and high design cost and low efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119513026A_ABST
    Figure CN119513026A_ABST
Patent Text Reader

Abstract

The invention discloses a PIM compiler designed through parameters in the technical field of compilers, an in-memory calculation main module is arranged in the compiler, and parameters are used for describing an in-memory calculation unit matrix. The parameters comprise the number of storage and calculation matrixes in the storage and calculation main module, the number of sub-matrixes in each matrix, the number of rows of the sub-matrixes, the number of columns of the sub-matrixes, the number of input parallel numbers of the sub-matrixes and the number of digits input each time and calculated with a single sub-matrix; defining the size of a calculation matrix in a memory, the number of corresponding peripheral circuits and a design method according to parameters; according to a performance evaluation result, a specific optimization scheme supported by multi-scale capacity and computing power based on rapid adjustment of circuit parameters is realized so as to meet different PIM IP design task requirements, the PIM compiler designed through parameters is reasonable in structural design, low-power-consumption and low-cost compiling design is realized, and the overall design efficiency of a computing chip in a memory is improved. The method can be applied to the use scene of in-memory computing chip IP design.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of compilers, in particular to a PIM compiler designed through parameters. Background Art

[0002] In recent years, with the rapid development of AI system-related technologies, the demand for compilers has become increasingly urgent. Compilers have changed from a relatively niche research field to a direction that both academia and industry pay close attention to and invest heavily in. In the field of chip design, compilers are an important part of the AI ​​chip software stack. Compilers are software tools used to automatically design and generate integrated circuits. They generate solutions suitable for IP modules based on user needs to meet customers' chip design requirements. These memory circuits include static random access memory (SRAM), read-only memory (ROM), and dynamic random access memory (DRAM).

[0003] During the compilation process, existing memory compilers have problems such as the need to improve the compilation speed and high design cost. There is a lack of compilation tools on the market that meet the design of in-memory computing chips. According to the different requirements of circuit designers, how to use the compiler to quickly generate a set of directly applicable memory computing structures has become a valuable research topic. Traditional compilers have the problems of high compilation design cost and low compilation efficiency. On the other hand, there is a gap in the market for compilers dedicated to the field of in-memory computing, and the commercialization of integrated memory and computing technology is imminent. There is an urgent need for software and hardware to work together to improve design efficiency. To this end, we propose a PIM compiler designed through parameters. Summary of the invention

[0004] The purpose of the present invention is to provide a PIM compiler designed through parameters to solve the problems of high compilation design cost and low compilation efficiency of traditional compilers mentioned in the above background technology. On the other hand, there is a market gap for compilers dedicated to the field of in-memory computing, and the commercialization of integrated storage and computing technology is imminent, and there is an urgent need for software and hardware to work together to improve design efficiency.

[0005] To achieve the above object, the present invention provides the following technical solutions: a PIM compiler designed by parameters, wherein an in-memory calculation main module is set in the compiler, and the in-memory calculation unit matrix is ​​described by parameters, wherein the parameters include the number of memory calculation matrices in the memory calculation main module, the number of sub-matrices in each matrix, the number of rows of the sub-matrix, the number of columns of the sub-matrix, the number of parallel inputs of the sub-matrix, and the number of bits for each input and calculation of a single sub-matrix;

[0006] Define the size of the in-memory computation matrix and the corresponding number and design method of peripheral circuits with parameters;

[0007] Based on the performance evaluation results, a specific optimization solution based on multi-scale capacity and computing power support with rapid adjustment of circuit parameters is implemented to meet the requirements of different PIM IP design tasks;

[0008] Finally, the compilation results of the compiler are directly output.

[0009] Preferably, the parameter definition can evaluate the area, power consumption, and computing power (computing speed) of the storage and computing circuit structure through the definition of parameters, and obtain the performance report of each module to quantitatively describe the trade-off relationship between different performance indicators.

[0010] Preferably, the input parameters can be adjusted according to various PIM design schemes, and a complete PIM design compilation result that meets various needs of designers can be output through matrix size, read-write circuits, address encoding, circuit calculation, and circuit calculation and post-addition tree circuits.

[0011] Preferably, if the application requirements are not met, the input parameters are adjusted, and a performance report of each module is obtained according to the module-level design solution, including area, performance, power consumption, etc.

[0012] Preferably, a variety of PIM design solutions are provided, and performance analysis, area analysis and complete design solutions are comprehensively given based on the split results, and the compiler results are output.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: the PIM compiler designed by parameters can flexibly adjust the parameters of the in-memory computing compiler, generate performance reports for each module, and output complete PIM design compilation results that meet the various needs of designers, which fills the gap in compilers in the IP design market of in-memory computing chips, solves the problems of lack of versatility of existing compilers, high design cost and low design efficiency, realizes low-power and low-cost compilation design, and improves the overall design efficiency of in-memory computing chips. It can be applied to the use scenarios of in-memory computing chip IP design. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the structure of the PIM compiler of the present invention;

[0015] Figure 2 This is a schematic diagram of the PIM structure of the present invention;

[0016] Figure 3 This is a schematic diagram of the in-memory computing storage matrix structure of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Embodiment 1:

[0019] See also Figure 1-3 The present invention provides a technical solution: a PIM compiler designed by parameters, in which an in-memory calculation main module is set in the compiler, and the in-memory calculation unit matrix is ​​described by parameters, and the parameters include the number of memory calculation matrices in the memory calculation main module, the number of sub-matrices in each matrix, the number of rows of the sub-matrix, the number of columns of the sub-matrix, the number of parallel inputs of the sub-matrix, and the number of bits for each input and calculation of a single sub-matrix;

[0020] Define the size of the in-memory computation matrix and the corresponding number and design method of peripheral circuits with parameters;

[0021] Based on the performance evaluation results, a specific optimization solution based on multi-scale capacity and computing power support with rapid adjustment of circuit parameters is implemented to meet the requirements of different PIM IP design tasks;

[0022] Finally, the compilation results of the compiler are directly output.

[0023] The parameter definition described above can evaluate the area, power consumption, and computing power (computing speed) of the storage and computing circuit structure, and obtain performance reports of each module to quantitatively describe the trade-off relationship between different performance indicators.

[0024] The input parameters can be adjusted according to various PIM design schemes, and the complete PIM design compilation results that meet the designer's various needs can be output through matrix size, read and write circuits, address encoding, circuit calculation, and circuit calculation and addition tree circuit.

[0025] If the application requirements are not met, the input parameters are adjusted and the performance report of each module, including area, performance, and power consumption, is obtained according to the module-level design plan.

[0026] Provide a variety of PIM design solutions, give comprehensive performance analysis, area analysis and complete design solutions based on the split results, and output compiler results

[0027] By compiling and running, a module-level design solution is generated. The performance of each module can be adjusted and optimized according to the designer's application requirements. Finally, a complete design solution is generated for in-memory computing IP. Compilation analysis and optimization can be reused. Under the in-memory computing structure design framework, the software running time can be greatly shortened, the power consumption of the compilation process can be reduced, and less memory can be occupied to save space resources, thereby significantly reducing the entire computing cost and improving the overall design efficiency of the in-memory computing chip. By designing an in-memory computing compiler with flexible parameters, a performance report of each module is generated, and a complete PIM design compilation result that meets the designer's various needs is output, which fills the gap in compilers in the in-memory computing chip IP design market, solves the problems of lack of versatility of existing compilers, high design cost and low design efficiency, realizes low-power and low-cost compilation design, improves the overall design efficiency of the in-memory computing chip, and can be applied to the use scenarios of in-memory computing chip IP design.

[0028] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0029] Although the present invention has been described above with reference to the embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the various features in the embodiments disclosed by the present invention may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A PIM compiler designed by parameters, characterized in that: The compiler sets an in-memory calculation main module, and uses parameters to describe the in-memory calculation unit matrix. The parameters include the number of memory calculation matrices in the memory calculation main module, the number of sub-matrices in each matrix, the number of rows of the sub-matrix, the number of columns of the sub-matrix, the number of parallel inputs of the sub-matrix, and the number of bits for each input and calculation of a single sub-matrix; Define the size of the in-memory computation matrix and the corresponding number and design method of peripheral circuits with parameters; Based on the performance evaluation results, a specific optimization solution based on multi-scale capacity and computing power support with rapid adjustment of circuit parameters is implemented to meet the requirements of different PIM IP design tasks; Finally, the compilation results of the compiler are directly output.

2. A parameter-designed PIM compiler according to claim 1, characterized in that: The parameter definition described above can evaluate the area, power consumption, and computing power (computing speed) of the storage and computing circuit structure, and obtain performance reports of each module to quantitatively describe the trade-off relationship between different performance indicators.

3. The parameter-designed PIM compiler according to claim 1, wherein: The input parameters can be adjusted according to various PIM design schemes, and the complete PIM design compilation results that meet the designer's various needs can be output through matrix size, read and write circuits, address encoding, circuit calculation, and circuit calculation and addition tree circuit.

4. The parameter-designed PIM compiler according to claim 1, wherein: If the application requirements are not met, the input parameters are adjusted and the performance report of each module, including area, performance, and power consumption, is obtained according to the module-level design plan.

5. The parameter-designed PIM compiler according to claim 1, wherein: Provides a variety of PIM design solutions, gives comprehensive performance analysis, area analysis and complete design solutions based on the split results, and outputs compiler results.