Multi-Chip I/O Access Synchronization via Inter-Chip Interconnect

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

As the number of cores in multi-core networking chips increases, managing access to on-chip and attached memory becomes challenging, leading to constrained processing speed and complex memory coherency issues, particularly in communication networks where parallel processing tasks are common.

Innovation Solution

A multi-chip system architecture with an inter-chip interconnect interface that enables efficient cross-chip communications, memory coherence methods, and synchronization techniques to allow multiple chip devices to operate as a single system, sharing resources and managing access to I/O devices in a conflict-free and efficient manner.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If the number of cores in multi-core networking chips increases, then processing power is improved, but managing access to on-chip and attached memory becomes challenging and processing speed is constrained

Engineering Contradiction:
Improveprocessing powerVSAvoidmemory access management complexity
Core Design Contradiction:
PowerVSDevice complexity

Solution Approach 1:

The patent divides the multi-core chip into multiple clusters, with each cluster containing a subset of cores and having its own local memory and I/O devices. This segmentation allows each cluster to manage memory access independently, reducing the overall complexity of memory access management across the entire chip while maintaining high processing power through parallel cluster operations.

Inventive Principle:
Principle #1Segmentation

2Productivity

If multiple chip devices are connected in a multi-chip system, then computational and memory capacities are increased, but synchronizing access to I/O devices becomes complex

Engineering Contradiction:
Improvecomputational capacityVSAvoidI/O access synchronization complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent introduces an inter-chip interconnect interface that acts as an intermediary between multiple chip devices in a multi-chip system. This interface handles the synchronization of I/O device access automatically, allowing multiple chips to operate together as a unified system with increased computational capacity without requiring complex synchronization logic in each individual chip.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If cross-chip communications are enabled, then resource sharing between chip devices is improved, but communication reliability and ordering become challenging

Engineering Contradiction:
Improveresource sharing capabilityVSAvoidcross-chip communication reliability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements preliminary ordering of I/O access requests at the queue level before they are transmitted across chip boundaries. By establishing a predetermined order for access requests and using queueing mechanisms, the system ensures that communications between chip devices are reliable and ordered, preventing data inconsistencies while still enabling extensive resource sharing across the multi-chip system.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS10592459B2Method and system for ordering I/O access in a multi-node environment
Publication Date: 2020.03.17 MARVELL ASIA PTE LTD
  • US10592459B2 patent drawing
  • US10592459B2 patent drawing
  • US10592459B2 patent drawing

AI summary

According to at least one example embodiment, a multi-chip system includes multiple chip devices configured to communicate to each other and share resources, such as I/O devices. According to at least one example embodiment, a method of synchronizing access to an input/output (I/O) device in the multi-chip system comprises initiating, by a first agent of the multi-chip system, a first operation for accessing the I/O device, the first operation is queued, prior to execution by the I/O device, in a queue. Once the first operation is queued, an indication of such queuing is provided. Upon detecting, by a second agent of the multi-chip system, the indication of queuing the first operation in the queue, initiating a second operation to access the I/O device, the second operation is queued subsequent to the first operation in the queue.