Dynamic Processing Element Cluster Topology for Neural Network Efficiency

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing neural network processor architectures face inefficiencies due to fixed computational clusters, which struggle to adapt to the changing computational dimensions of convolutional neural network layers, leading to low computational efficiency.

Innovation Solution

A neural network calculation method that determines calculation parameters for convolutional neural network layers and configures a data topological relationship between processing element clusters to form a reconstructed processing element cluster set, optimizing the convolution operation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a fixed computational cluster is used, then the processor structure is simple and stable, but the computational efficiency decreases when calculating convolutional neural networks with changing computational dimensions

Engineering Contradiction:

Improvecomputational efficiencyVSAvoidprocessor architecture complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent implements dynamic reconstruction of the processing element cluster set based on calculation parameters. The controller dynamically configures the number of processing element clusters and their data topological relationships according to the specific computational dimensions of different convolutional neural network layers, transforming the fixed architecture into a dynamic one that adapts to varying computational requirements.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters of the processing element cluster set by reconstructing it with different numbers of clusters and different data topological relationships based on calculation parameters such as input channel size, output channel size, and feature map dimensions. This parameter adaptation allows the processor to optimize computational efficiency for different network layer configurations.

Inventive Principle:

Principle #35Parameter changes

2Productivity

If the number of processing element clusters is increased to handle larger computational dimensions, then computational efficiency improves, but device complexity and resource allocation difficulty increase

Engineering Contradiction:

Improvecomputational efficiencyVSAvoidadaptability to different computational dimensions

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal processing element cluster set that can handle different computational dimensions through dynamic reconstruction. The same physical processing elements can be reconfigured into different cluster numbers and topological relationships to accommodate various input channel sizes, output channel sizes, and feature map dimensions, making the processor universally adaptable to different convolutional neural network configurations.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system dynamically adjusts the number of processing element clusters and their data topological relationships based on the specific calculation parameters of each convolutional neural network layer, enabling the processor to adapt to varying computational dimensions without requiring separate hardware for each configuration.

Inventive Principle:

Principle #15Dynamics

3Adaptability or versatility

If a fixed data topological relationship is used, then the processor architecture is simpler, but the adaptability to different calculation parameters decreases

Engineering Contradiction:

Improveadaptability to calculation parametersVSAvoiddata topological configuration complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements dynamic configuration of data topological relationships between processing element clusters based on calculation parameters. The controller reconstructs the data topological relationship to match the specific dimensions of input channels, output channels, and feature maps, transforming a static topological structure into a dynamic one that adapts to different computational requirements.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes the data topological relationship parameters between processing element clusters according to calculation parameters such as input channel size, output channel size, and feature map dimensions. This allows the system to optimize data flow and computational efficiency for different convolutional neural network layer configurations.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250139416A1Neural network computing method, processor and device thereof

Publication Date: 2025.05.01 LENOVO (BEIJING) LTD

AI summary

A neural network calculation method, including: determining calculation parameters of convolutional neural network layers; based on the calculation parameters, configuring a data topological relationship between a plurality of processing element clusters in a processing element cluster set of a processor as a target data topological relationship, to form a reconstructed processing element cluster set; and inputting input-feature-map data of the convolutional neural network layers and convolution kernel data of the convolutional neural network layers into the reconstructed processing element cluster set for convolution operation based on the calculation parameters to obtain output feature map data.