IPA Remapping for Dual-Rank DRAM to Minimize Rank Switching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Dual-rank DRAM systems experience increased power consumption and degraded GPU performance due to frequent switching between ranks, which is not optimized in existing technologies.

Innovation Solution

An IPA remapping method that concentrates memory access by monitoring extant mappings, determining offset blocks, and remapping allocated and unallocated blocks to minimize rank switching, using memory management units (MMUs) to optimize memory access.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If dual-rank DRAM is used to increase memory capacity, then memory capacity is improved, but power consumption increases and performance degrades due to frequent rank switching

Engineering Contradiction:

Improvememory capacityVSAvoidpower consumption

Core Design Contradiction:

Quantity of substanceVSUse of energy by moving object

Solution Approach 1:

The patent performs preliminary analysis of application memory access patterns before mapping to physical addresses. The MMU pre-determines which rank will be the primary access target and configures the mapping accordingly, preventing frequent rank switching during runtime. This advance planning allows dual-rank DRAM to be used while maintaining single-rank access efficiency.

Inventive Principle:

Principle #10Preliminary action

2Quantity of substance

If dual-rank DRAM is used to increase memory capacity, then memory capacity is improved, but GPU performance degrades dramatically

Engineering Contradiction:

Improvememory capacityVSAvoidGPU performance

Core Design Contradiction:

Quantity of substanceVSProductivity

Solution Approach 1:

The patent applies different mapping strategies to different application types. For GPU workloads, the MMU identifies graphics-related IPAs and maps them preferentially to a specific rank, creating a localized optimization for GPU access patterns. This ensures that GPU operations access memory from a single rank, eliminating rank switching overhead and restoring GPU performance to single-rank levels.

Inventive Principle:

Principle #3Local quality

3Ease of operation

If extant IPA to PA mapping is used without remapping, then memory allocation simplicity is maintained, but rank switching occurs frequently causing performance degradation

Engineering Contradiction:

Improvememory allocation simplicityVSAvoidmemory access efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The patent introduces an intermediate remapping layer in the MMU that sits between the standard IPA allocation mechanism and the physical DRAM ranks. This intermediary analyzes the allocation pattern and performs intelligent remapping to concentrate accesses on one rank, while transparently maintaining the simplicity of the original IPA allocation interface. Applications continue to use simple sequential IPAs, but the intermediary translates them to PA patterns that avoid rank switching.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12536105B2Method for intermediate physical address (IPA) remapping and electronic device

Publication Date: 2026.01.27 MEDIATEK INC

AI summary

A method for intermediate physical address (IPA) remapping, applicable to a memory having multiple ranks are provided. The method includes the following steps. An extant mapping from IPAs to physical addresses (PAs) of the memory is monitored. A remapping from at least one IPA to at least one PA is executed to concentrate the PAs of the memory used by an application into one of the ranks of the memory.