DRAM Row Merging via Hardware Mapping Table

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Dynamic Random Access Memory (DRAM) row refreshes consume a significant portion of power, and existing software-based solutions for deduplicating identical data across DRAM rows are computationally expensive and limited in execution, especially when total system memory falls below a threshold.

Innovation Solution

A hardware-based approach for DRAM row merging, where a memory controller identifies rows with identical data and maps them to a reserved row, excluding these rows from refresh operations by using a mapping table and bitmap to indicate which rows should be skipped during refresh, thereby reducing the need for frequent data copying and computational overhead.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If software-based deduplication is used to identify and merge identical DRAM rows, then data redundancy is reduced, but computational overhead increases significantly

Engineering Contradiction:
ImproveDRAM refresh power consumptionVSAvoidcomputational overhead
Core Design Contradiction:
Loss of energyVSDevice complexity

Solution Approach 1:

The patent replaces software-based computational deduplication with hardware-based mechanisms including a mapping table and bitmap stored in the memory controller. The hardware automatically tracks row mappings and identifies duplicate rows through bitwise operations on the bitmap, eliminating the need for expensive software computation while maintaining deduplication functionality

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The memory controller performs automatic row merging and tracking without requiring external software intervention. The bitmap and mapping table enable the system to self-identify duplicate rows and self-manage the merging process, reducing computational overhead while achieving energy savings from reduced refresh operations

Inventive Principle:
Principle #25Self-service

2Loss of energy

If software-based deduplication solutions are implemented, then identical data rows are merged, but the solutions are limited in execution when total system memory falls below a threshold

Engineering Contradiction:
ImproveDRAM refresh power consumptionVSAvoidexecution flexibility
Core Design Contradiction:
Loss of energyVSAdaptability or versatility

Solution Approach 1:

The hardware-based mapping table and bitmap mechanism provides universal functionality that operates independently of system memory conditions. The same hardware structure handles both small and large memory systems without requiring different software implementations, making the solution adaptable across various memory configurations and thresholds

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Reliability

If all DRAM rows are refreshed to ensure data integrity, then data reliability is maintained, but power consumption increases

Engineering Contradiction:
Improvedata integrityVSAvoidpower consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent extracts and identifies duplicate rows from the set of all DRAM rows using the bitmap and mapping table. By separating identical data into a single representative row and excluding duplicates from refresh operations, the system maintains data integrity for unique data while eliminating unnecessary refresh power consumption for redundant data

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The refresh operation is differentiated by row: unique rows receive full refresh treatment to maintain data integrity, while duplicate rows are excluded from refresh since their data is maintained in the representative row. This local differentiation optimizes power consumption while preserving reliability where needed

Inventive Principle:
Principle #3Local quality

4Speed

If multiple identical DRAM rows are maintained, then data accessibility is improved, but the number of refresh operations increases

Engineering Contradiction:
Improvedata accessibilityVSAvoidrefresh power consumption
Core Design Contradiction:
SpeedVSLoss of energy

Solution Approach 1:

The mapping table acts as an intermediary that translates access requests to merged rows into accesses of the representative row. This intermediary structure maintains fast data accessibility by providing direct mapping paths while enabling the system to refresh only the representative row, thereby reducing total refresh operations and power consumption

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS11721384B2Hardware-assisted dynamic random access memory (DRAM) row merging
Publication Date: 2023.08.08 ADVANCED MICRO DEVICES INC
  • US11721384B2 patent drawing
  • US11721384B2 patent drawing
  • US11721384B2 patent drawing

AI summary

Hardware-assisted Dynamic Random Access Memory (DRAM) row merging, including: identifying, by a memory controller, in a DRAM module, a plurality of rows storing identical data; storing, in a mapping table, data mapping one or more rows of the plurality of rows to another row; and excluding the one or more rows from a refresh the DRAM module.