Shared Instruction Prefetch Pipeline for Multicore Fetch Bandwidth

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing multicore processing systems face inefficiencies in instruction prefetching due to the high cost of providing private instruction memory for each core and the complexity of full cache coherent solutions, which are impractical for cross-lane processing units (XPUs).

Innovation Solution

A hardware circuit with a shared instruction memory (TiMem) and independent instruction buffers (iBuf) for each core, utilizing a prefetch pipeline architecture that includes a prefetch unit and instruction router to optimize instruction delivery, reducing redundant requests and improving bandwidth.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If independent instruction streams are fetched for each thread on separate cores, then thread-level parallelism is achieved, but instruction fetch bandwidth is insufficient and performance degrades

Engineering Contradiction:
Improvethread-level parallelism performanceVSAvoidinstruction fetch bandwidth
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent merges the instruction fetch resources of multiple cores into a shared instruction fetch unit. Multiple cores share a common instruction fetch bandwidth pool, allowing coordinated instruction fetching across cores. This resolves the bandwidth insufficiency by combining individual core fetch capabilities into a larger shared resource that can serve multiple threads simultaneously.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The instruction fetch unit is designed to serve multiple cores universally rather than being dedicated to a single core. The same fetch infrastructure can service instructions for any thread on any core, making the system more efficient at utilizing available bandwidth across diverse thread workloads while maintaining the ability to handle thread-level parallelism.

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

2Use of energy by moving object

If conservative prefetching is used to avoid misprediction penalties, then fetch resource utilization is improved, but prefetcher performance is limited

Engineering Contradiction:
Improvefetch resource utilizationVSAvoidprefetcher performance
Core Design Contradiction:
Use of energy by moving objectVSProductivity

Solution Approach 1:

The patent implements feedback mechanisms where the prefetcher monitors actual instruction demand and fetch outcomes. When mispredictions occur or bandwidth is available, the system adjusts prefetching aggressiveness based on observed performance. This feedback loop allows the system to balance between conservative resource usage and aggressive prefetching to maximize overall prefetcher effectiveness without wasting fetch bandwidth.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4320514B1Cooperative instruction prefetch on multicore system
Publication Date: 2026.05.06 GOOGLE LLC
  • EP4320514B1 patent drawingFigure 1
  • EP4320514B1 patent drawingFigure 2
  • EP4320514B1 patent drawingFigure 3

AI summary

Aspects of the disclosure are directed to methods, systems, and apparatuses using an instruction prefetch pipeline architecture that provides good performance without the complexity of a full cache coherent solution deployed in conventional CPUs. The architecture can include components which can be used to construct an instruction prefetch pipeline, including instruction memory (TiMem), instruction buffer (iBuf), a prefetch unit, and an instruction router.