Convolution Processing Device Data Segmentation for AI

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

There is a demand for higher-performance computational units that can execute convolution computation efficiently, particularly for AI applications, and existing technologies face challenges in miniaturizing computational units while maintaining processing capabilities.

Innovation Solution

A computation processing device is designed with a first computational unit that performs simultaneously-executable convolution computation on data corresponding to no more than a first maximum number of channels, and a data dividing unit that splits data into sub-divisions with no more than the first maximum number of channels when the data exceeds this limit.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Volume of moving object

If the computational unit is miniaturized to reduce device scale, then the device size is reduced, but the processing capability for convolution computation is limited

Engineering Contradiction:

Improvedevice scaleVSAvoidprocessing capability

Core Design Contradiction:

Volume of moving objectVSProductivity

Solution Approach 1:

The patent divides the input data into multiple data groups, where each group contains data corresponding to no more than a maximum number of channels. This segmentation allows the miniaturized computational unit to process data in manageable portions, achieving the desired computation processing while maintaining a small device scale.

Inventive Principle:

Principle #1Segmentation

2Productivity

If the computational unit processes data with more channels simultaneously, then the processing capability is improved, but the device scale increases

Engineering Contradiction:

Improveprocessing capabilityVSAvoiddevice scale

Core Design Contradiction:

ProductivityVSVolume of moving object

Solution Approach 1:

The patent segments the data into multiple groups, each with no more than a maximum number of channels, allowing the computational unit to maintain a small scale while still processing data with many channels through multiple sequential operations.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a temporal dimension by processing data groups sequentially across multiple time steps. This allows the system to handle data with more channels than the computational unit can process simultaneously, effectively increasing processing capability without increasing device scale.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Productivity

If a single computational unit processes all data channels simultaneously, then the processing capability is maximized, but the device complexity increases

Engineering Contradiction:

Improveprocessing capabilityVSAvoiddevice complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the data into multiple groups with no more than a maximum number of channels each, allowing a single computational unit to process data efficiently without requiring complex parallel architectures. This segmentation approach reduces device complexity while maintaining processing capability.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12326914B2Computation processing device, computation processing method, and configuration program

Publication Date: 2025.06.10 NEC PLATFROMS LTD

AI summary

This computation processing device includes: a first computational unit that executes simultaneously-executable convolution computation on data corresponding to no more than a first maximum number of channels of the convolution computation; and a data dividing unit that divides the data which is subject to the convolution computation into data which has no more than the first maximum number of channels when the number of pieces of data which is subject to the convolution computation exceeds the first maximum number of channels.