Memory Port Segmentation for High Bandwidth

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current DRAM designs that share the same port for row and column address signals lead to inefficiencies, such as incomplete data line occupation and reduced memory speed, especially in high-bandwidth applications, due to the need for pre-fetch and burst techniques, which can cause data wastage and poor flexibility.

Innovation Solution

Separating row and column address ports allows for simultaneous transmission of address signals, ensuring the data line is always fully occupied by enabling continuous data transmission without waiting for read commands to finish, thereby improving memory speed and performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If row and column address ports are shared, then device complexity is reduced, but memory speed and data transmission efficiency deteriorate

Engineering Contradiction:
Improveport structureVSAvoidmemory speed
Core Design Contradiction:
Device complexityVSSpeed

Solution Approach 1:

The command port is segmented into two independent sub-ports: a row address port (RAS) and a column address port (CAS). This segmentation allows simultaneous reception of row and column address signals without mutual interference, resolving the bottleneck where sequential address transmission limited memory speed. The segmentation maintains relatively simple device architecture while significantly improving data transmission efficiency.

Inventive Principle:
Principle #1Segmentation

2Duration of action of moving object

If pre-fetch and burst techniques are used, then data transmission continuity is improved, but data line occupation efficiency deteriorates due to data wastage

Engineering Contradiction:
Improvedata transmission continuityVSAvoiddata line occupation efficiency
Core Design Contradiction:
Duration of action of moving objectVSProductivity

Solution Approach 1:

The memory system dynamically receives row and column address signals at different times through separate ports, enabling flexible timing control without being constrained by fixed pre-fetch or burst patterns. This dynamic approach allows the data line to be occupied continuously with valid data only, eliminating the data wastage inherent in static pre-fetch and burst techniques while maintaining transmission continuity.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4033326B1memory
Publication Date: 2024.02.07 CHANGXIN MEMORY TECH INC
  • EP4033326B1 patent drawingFigure 1~2
  • EP4033326B1 patent drawingFigure 3
  • EP4033326B1 patent drawingFigure 4

AI summary

A memory includes a memory chip (100), a command port (102) and a data port (103). The memory chip (100) includes at least one channel (10). Each channel (10) includes a plurality of banks (101), and the banks (101) are configured to perform read and write operations alternately. The command port (102) is configured to receive command signals at a preset edge of a command clock, and the command signals are configured to control the read and write operations of the banks (101). The data port (103) is configured to receive data signals to be written into the banks (101) or transmit data signals at preset edges of a data clock. The command port (102) includes a row address port (112) and a column address port (122). The row address port (112) is configured to receive a row address signal at a position of a target memory cell, and the column address port (122) is configured to receive a column address signal at a position of the target memory cell.