3D Express Link Architecture for NoC Latency Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional 2D mesh network-on-chip (NoC) designs suffer from latency and congestion issues due to high average data-hop counts and hotspots, which hinder performance, bandwidth, and power, performance, and area (PPA) targets.

Innovation Solution

The implementation of a multi-dimensional (3D) express link architecture that extends the topology of a first die using the metal layers of a second die, adding express links to a 2D mesh to couple non-neighbor router logic through a multi-layered structure, thereby reducing latency and increasing bisection bandwidth.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If 2D mesh topology is used for scalable processors, then the design can be tiled and replicated for given PPA targets, but latency increases due to paths crossing significant number of routers

Engineering Contradiction:
ImprovebandwidthVSAvoidlatency
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent introduces a third dimension to the traditional 2D mesh network by adding vertical routing paths through multiple layers. This allows data to travel through the Z-axis (up/down between layers) rather than being constrained to horizontal paths, significantly reducing the number of routers data packets must traverse and thereby reducing latency while maintaining scalability.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The network is segmented into multiple layers with dedicated routing paths for different traffic types. By dividing the network into hierarchical levels (core routers in one layer, edge routers in another), the patent creates specialized pathways that reduce congestion and improve overall bandwidth while maintaining manageable complexity.

Inventive Principle:
Principle #1Segmentation

2Adaptability or versatility

If 2D mesh topology is used, then tiled design can be replicated, but congestion occurs in the middle of the mesh structure

Engineering Contradiction:
ImprovescalabilityVSAvoidbandwidth
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

By adding vertical routing paths through multiple layers, the patent distributes traffic across three dimensions rather than concentrating it in the center of a 2D mesh. This dimensional expansion provides alternative pathways around congestion points, maintaining high bandwidth utilization while preserving the scalable tiled architecture.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent introduces intermediate routing layers that act as mediators between edge routers and core routers. These intermediate structures provide alternative pathways that bypass congested regions, allowing data to route around bottlenecks while maintaining the overall mesh topology and its scalability benefits.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Quantity of substance

If more routers are added to increase logic and wires for computation, then storage and communication capacity increases, but path length and latency increase

Engineering Contradiction:
Improvelogic capacityVSAvoidlatency
Core Design Contradiction:
Quantity of substanceVSLoss of time

Solution Approach 1:

The patent resolves this contradiction by organizing routers in a three-dimensional spatial arrangement where vertical shortcuts provide direct pathways between distant nodes. This allows the network to scale in logic capacity by adding more routers while maintaining low latency through the Z-axis routing paths that bypass the need to traverse through intermediate routers in the X-Y plane.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS12316537B2Multi-dimensional routing architecture
Publication Date: 2025.05.27 ARM LTD
  • US12316537B2 patent drawing
  • US12316537B2 patent drawing
  • US12316537B2 patent drawing

AI summary

Various implementations described herein refer to a device having a multi-layered logic structure with multiple layers including a first layer and a second layer arranged vertically in a stacked configuration. The device may have a first network that links nodes together in the first layer. The device may have a second network that links the nodes in the first layer together by way of the second layer so as to reduce latency related to data transfer between the nodes.