Neural Network Integer Conversion for Low-Power AI Processing

Solution Overview

Problem

Neural network devices face challenges in efficiently processing floating-point numbers due to their complexity and power consumption, which is particularly problematic for low-power devices like smartphones that require high-performance operations.

Innovation Solution

Incorporating a direct memory access (DMA) controller and a data converter within the neural network device to convert floating-point data into integer-type data, allowing the processor to perform neural network operations using integer operations, thereby reducing power consumption and improving processing speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If floating-point data is used for neural network operations, then the precision and range of data representation are improved, but the power consumption and processing complexity increase

Engineering Contradiction:

Improvedata representation precisionVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent changes the data type parameter from floating-point to integer representation. The data converter transforms floating-point input data into integer format, allowing the neural network processor to operate with integer arithmetic instead of floating-point arithmetic, thereby reducing power consumption while maintaining acceptable precision through quantization

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces a data converter as an intermediary component between the memory and the neural network processor. This converter acts as a mediator that transforms floating-point data into integer format before processing, enabling the processor to work with simpler integer arithmetic while the converter handles the complexity of floating-point to integer conversion

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If floating-point data is used for neural network operations, then the range and precision of data representation are improved, but the device complexity increases

Engineering Contradiction:

Improvedata representation precisionVSAvoidprocessing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent changes the data type parameter from floating-point to integer representation. The data converter transforms floating-point input data into integer format, allowing the neural network processor to operate with integer arithmetic instead of floating-point arithmetic, thereby reducing processing complexity while maintaining acceptable precision through quantization

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces a data converter as an intermediary component between the memory and the neural network processor. This converter acts as a mediator that transforms floating-point data into integer format before processing, enabling the processor to work with simpler integer arithmetic while the converter handles the complexity of floating-point to integer conversion

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If integer operations are used for neural network processing, then the power consumption and processing speed are improved, but the ability to represent precise floating-point values deteriorates

Engineering Contradiction:

Improveprocessing speedVSAvoidfloating-point value precision

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent changes the data type parameter from floating-point to integer representation. The data converter transforms floating-point input data into integer format, allowing the neural network processor to operate with integer arithmetic instead of floating-point arithmetic, thereby reducing power consumption while maintaining acceptable precision through quantization

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces a data converter as an intermediary component between the memory and the neural network processor. This converter acts as a mediator that transforms floating-point data into integer format before processing, enabling the processor to work with simpler integer arithmetic while the converter handles the complexity of floating-point to integer conversion

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11429838B2Neural network device for neural network operation, method of operating neural network device, and application processor including the neural network device

Publication Date: 2022.08.30 SAMSUNG ELECTRONICS CO LTD

AI summary

Provided are a neural network device for performing a neural network operation, a method of operating the neural network device, and an application processor including the neural network device. The neural network device includes a direct memory access (DMA) controller configured to receive floating-point data from a memory; a data converter configured to convert the floating-point data received through the DMA controller to integer-type data; and a processor configured to perform a neural network operation based on an integer operation by using the integer-type data provided from the data converter.