Physical-Address Cache Banking for Parallel Access Without Aliasing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing multi-bank caches based on virtual addresses require each bank to connect to all memory channels, leading to increased interconnection costs, power consumption, and degraded processor performance.

Innovation Solution

Partition the cache space into banks based on physical addresses, allowing parallel access and reducing the need for each bank to connect to only one memory channel, with MMU or TLB lookup before writing data to eliminate aliasing issues.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If each bank connects to all memory channels in virtual address-based multi-bank caches, then parallel access capability is achieved, but interconnection complexity and power consumption increase

Engineering Contradiction:
Improveparallel access capabilityVSAvoidinterconnection complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the cache into banks based on physical addresses rather than virtual addresses, which fundamentally changes the interconnection requirements. This physical address-based segmentation allows each bank to connect to only one memory channel, reducing interconnection complexity while maintaining parallel access capability through the physical address mapping structure.

Inventive Principle:
Principle #1Segmentation

2Productivity

If each bank connects to all memory channels to support parallel access, then data read/write operations can proceed in parallel, but power consumption increases

Engineering Contradiction:
Improvedata read/write parallelismVSAvoidcache power consumption
Core Design Contradiction:
ProductivityVSUse of energy by stationary object

Solution Approach 1:

By segmenting the cache into physically-addressed banks, the patent enables parallel data read/write operations across different banks without requiring each bank to maintain multiple memory channel connections. This reduces the overall power consumption of the cache subsystem while preserving parallel processing capability.

Inventive Principle:
Principle #1Segmentation

3Loss of time

If virtual address-based bank partitioning is used, then cache hit access time is reduced, but aliasing issues occur during data writing

Engineering Contradiction:
Improvecache hit access timeVSAvoiddata writing accuracy
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The patent inverts the traditional approach by using physical addresses for bank partitioning instead of virtual addresses. This inversion eliminates the aliasing problem that occurs with virtual address-based partitioning, as physical addresses uniquely identify memory locations without translation ambiguity. The MMU or TLB performs address translation after the bank selection is made, ensuring accurate data writing while maintaining fast cache access.

Inventive Principle:
Principle #13The other way round (Inversion)

Data Source

PatentEP4654025A1Caching method and apparatus, electronic device, storage medium, and program product
Publication Date: 2025.11.26 MOORE THREADS TECH CO LTD
  • EP4654025A1 patent drawingFigure 1
  • EP4654025A1 patent drawingFigure 2
  • EP4654025A1 patent drawingFigure 3~4

AI summary

A cache method and apparatus, an electronic device, a storage medium, and a program product. The cache method includes: partitioning a cache space into a plurality of banks based on physical addresses, wherein the plurality of banks allow parallel access (S11); and in response to a data query request, performing a read/write operation on at least one of the plurality of banks (S12).