Neural Network Processing Unit With Kernel Dependency Scheduling

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Interactions between a central processing unit (CPU) and other processing units during neural network processing cause performance degradation and increased power consumption in electronic devices.

Innovation Solution

A processing unit that generates additional commands based on the type of kernel, allowing a consuming kernel to reference data produced by a producing kernel without direct interaction between the CPU and the processing unit, through a coordinator and command processor.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the CPU interacts with other processing units during neural network processing, then data transfer can be performed, but performance degradation occurs and power consumption increases

Engineering Contradiction:

Improveneural network processing speedVSAvoidpower consumption

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The patent introduces additional commands as intermediaries between the producing kernel and consuming kernel. These commands enable data dependency management and synchronization without requiring direct CPU intervention, thereby reducing CPU-processing unit interactions while maintaining processing speed and reducing power consumption

Inventive Principle:

Principle #24Intermediary (Mediator)

2Productivity

If the CPU interacts with other processing units during neural network processing, then data transfer can be performed, but performance degradation occurs

Engineering Contradiction:

Improveneural network processing speedVSAvoidprocessing time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent implements preliminary actions by generating additional commands in advance that manage data dependencies and synchronization between kernels. This allows the processing unit to autonomously coordinate data transfer and execution without waiting for CPU instructions, reducing processing time and improving performance

Inventive Principle:

Principle #10Preliminary action

3Loss of energy

If additional commands are generated based on kernel type, then interactions between CPU and processing units are reduced, but device complexity increases

Engineering Contradiction:

Improvepower consumptionVSAvoidcommand processing complexity

Core Design Contradiction:

Loss of energyVSDevice complexity

Solution Approach 1:

The patent applies local quality by generating different additional commands based on the specific type of kernel (producing or consuming). This targeted approach reduces CPU interactions only where needed while maintaining simple command processing for other cases, thereby reducing power consumption without significantly increasing overall device complexity

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20260064472A1Processing unit for processing neural network, electronic device including the same, and host processor

Publication Date: 2026.03.05 SAMSUNG ELECTRONICS CO LTD

AI summary

A processing unit for processing a neural network includes a coordinator configured to receive a kernel command from a host processor, determine a type of kernel corresponding to the kernel command, and generate an additional command different according to the type of kernel, the kernel command being used to execute kernels required to process the neural network, and a command processor configured to schedule such that the data produced by the producing kernel is referred to when the consuming kernel is executed based on the kernel command and the additional command.