Neural Network MAC Array with Zero-Weight Skipping for Power Control

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing electronic devices face challenges in efficiently processing input data using neural networks due to the need for mathematical operations between weights and input data, which are not optimized for various device situations, leading to suboptimal performance and power efficiency.

Innovation Solution

The electronic device incorporates a switch and a multiply and accumulate (MAC) array with a zero weight skip unit and MAC activation control unit to selectively perform convolution operations based on device conditions, such as resolution, power consumption, and battery state, thereby optimizing weight processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If data is pre-processed before being input into the neural network, then recognition accuracy is improved, but processing time is increased

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-processing data before neural network input to improve recognition accuracy. The system performs data collection, pre-processing (including normalization, filtering, and feature extraction), and then inputs the processed data into the neural network for recognition, achieving better accuracy through advance preparation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent segments the data processing workflow into distinct modules: data collection module, pre-processing module, neural network recognition module, and result output module. This segmentation allows parallel processing and optimization of each stage independently, reducing overall processing time while maintaining accuracy.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If a neural network recognition module is used, then recognition accuracy is improved, but device complexity is increased

Engineering Contradiction:
Improverecognition accuracyVSAvoiddevice complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements a unified neural network recognition module that handles multiple recognition tasks (voice recognition, image recognition, text recognition) through a single integrated system. This multi-functional approach improves recognition accuracy across different modalities while avoiding the complexity of separate dedicated systems for each task.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent introduces a pre-processing module as an intermediary between data collection and neural network recognition. This intermediary layer prepares and standardizes input data, enabling the neural network to focus on pattern recognition rather than data cleaning, thereby improving accuracy while managing system complexity through clear separation of concerns.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If data is collected from multiple sources, then recognition accuracy is improved, but energy consumption is increased

Engineering Contradiction:
Improverecognition accuracyVSAvoidenergy consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent merges multiple data sources (microphone, camera, sensor, database) into a unified data collection module that operates under centralized control. The system intelligently activates only the necessary data sources based on the specific recognition task, combining their inputs after preprocessing to achieve improved accuracy while minimizing energy consumption by avoiding unnecessary data collection.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3867825B1Method for processing data using neural network and electronic device for supporting the same
Publication Date: 2026.04.29 SAMSUNG ELECTRONICS CO LTD
  • EP3867825B1 patent drawingFigure 1
  • EP3867825B1 patent drawingFigure 2~3
  • EP3867825B1 patent drawingFigure 4~5

AI summary

An electronic device is provided. The electronic device includes a switch configured to select a mode of the electronic device, a multiply and accumulate (MAC) array configured to include a plurality of MAC units, and at least one processor configured to include a zero weight skip unit for confirming a first weight having a value of '0' among weights related with input data, and for forwarding a second weight not having a value of '0' among the weights, to the MAC array. The at least one processor is configured to acquire the input data, acquire the weights, select the mode of the electronic device by using the switch, in response to a first mode of the electronic device being selected, perform convolution operations between the input data and the second weight forwarded to the MAC array through the zero weight skip unit, and, in response to a second mode of the electronic device being selected, perform convolution operations between the input data and the weights forwarded to the MAC array.