Power modeling method and power managing method for calculation in internal storage

Abstract

The invention discloses a power modeling method and a power managing method for calculation in internal storage. The power modeling method comprises the following steps: establishing a BP model, and adopting each power bandwidth BP to represent a relationship between bandwidth and power for calculating PIM in internal storage; the power managing method comprises the following steps: power monitoring sub task limit PAST, processing unit acceleration PUB and power sprint PS; when power supply exceeds power needs, adopting PAST to manage power consumption in PIM; adopting a PUB method while needing to adopt a dynamic regulating power mode; while short-time power overload is required, adopting a PS method to improve a power upper limit value of a power arbiter; and realizing power management for calculation in internal memory. Power obtained by adopting the BP model is consistent with practically measured power, the PAST method can successfully limit power of the PIM, and the PUB can successfully improve performances of a chip; and meanwhile, the PAST method, the PUB method and the PS method can be adopted to generate a more effective energy system, and a power management scheme is reasonably configured, so that performances of PIM are further improved.

Description

technical field [0001] The invention relates to an in-memory computing technology, in particular to a power modeling method and a power management method based on in-memory computing. Background technique [0002] Processing-in-Memory (PIM) is an idea of ​​transferring computation to memory, and is currently widely used to accelerate data-sensitive applications. With many data-centric applications, such as real-time analysis, graph computing, and neural network algorithms, the demand for high-speed, high-bandwidth data retrieval is increasingly strong, and PIM design is gradually becoming a research hotspot in academia and industry. Computational tasks related to PIM (such as word counting, range lookup, etc.) are usually relatively simple but contain a large amount or even all of the data in memory. Therefore, a PIM task can be divided and transferred to multiple memory processor units (such as HMC memory banks and RRAM crossbar arrays), so that they can complete the task ...

AI Technical Summary

Problems solved by technology

This results in additional thermal cost and design complexity, and affects the logic behavior of the PIM

Also, the relationship between power and data processing throughput in PIM has not been well studied

In the absence of a suitable power model guide, the PIM structure may be designed with a mismatched memory bandwidth and power supply relationship, and the power consumption of previous PIM designs may exceed the power supply at peak throughput.

Images