Embedded hybrid computing semiconductor memory

By designing a computing storage fusion engine and heterogeneous resource management module in embedded semiconductor memory, dynamically adjusting the data transmission path and computing unit use, the performance bottleneck of the traditional computing and storage separation architecture is solved, and efficient computing, optimization of energy consumption and improvement of resource utilization is achieved.

CN119668524AActive Publication Date: 2025-03-21深圳华芯星半导体有限公司
4 Cites 0 Cited by

Patent Information

Application Number
CN202510179446.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-02-19
Publication Date
2025-03-21
Estimated Expiration
2045-02-19

AI Technical Summary

Technical Problem

The traditional computing and storage separation architecture faces performance bottlenecks, including large data handling overhead, limited interface bandwidth, low computing architecture efficiency and high programming model complexity, which affect system performance and scalability.

Method used

Design an embedded hybrid computing semiconductor memory, including a computing storage fusion engine and a heterogeneous resource management module. The computing storage convergence engine dynamically changes data transmission paths through reconfigurable interconnect networks to reduce data movement. The heterogeneous resource management module dynamically adjusts the use of computing units based on reinforcement learning based on real-time load.

Benefits of technology

By reducing data movement, improving computing efficiency, optimizing energy consumption ratio, reducing system latency, and improving resource utilization and system performance through flexible computing unit allocation and intelligent task scheduling.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses an embedded hybrid computing semiconductor memory, comprising: a computing storage fusion engine comprising a plurality of types of computing units; wherein a reconfigurable interconnection network is formed between the plurality of computing units and the memory, and a data transmission path is allowed to be dynamically changed; and the heterogeneous resource management module is used for dynamically adjusting the use of the computing unit according to the real-time load based on reinforcement learning. In this way, data movement can be reduced to improve the calculation efficiency, the energy consumption ratio is optimized, and the system delay is reduced.
Need to check novelty before this filing date? Find Prior Art