An in-memory computing device and method based on ping-pong structure

Abstract

The purpose of the present invention is to provide an in-memory computing device and method based on a ping-pong structure, which belongs to the technical field of in-memory computing. Based on the in-memory computing macro with a ping-pong structure, the device designs appropriate specific device building modules including an input data cache module, an in-memory computing core module, a top-level control module, a memory control module, and a configuration module. The data scheduling scheme is used to complete the processing and updating of data by the device.

Description

technical field [0001] The invention belongs to the technical field of in-memory computing, and in particular relates to an in-memory computing device and method based on a ping-pong structure. Background technique [0002] With the development of neural network algorithms, the increasingly complex network structure and huge amount of parameters have brought great challenges to processors in terms of computing power and power consumption. As a potential solution, in-memory computing technology has attracted widespread attention in recent years. Computing in memory (CIM) is an emerging non-Von Neumann (Non Von Neumann) architecture technology that reduces data movement by building storage units with computing capabilities and one-time large-scale multiply-add operations. It can realize energy-efficient and large-computing hardware solutions with energy consumption per calculation. [0003] However, limited by the current hardware scale and preparation process, the in-memory ...

AI Technical Summary

Problems solved by technology

However, although this solution provides in-memory computing macros with ping-pong functions, it does not consider the adapted data scheduling method and specific device solutions when such computing macros are applied to digital systems, resulting in limited performance improvement of ping-pong in-memory computing macros.

Images