Approximate Computing Control for Runtime Accuracy Switching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing Approximate Computing technologies face challenges in dynamically adjusting calculation accuracy to keep errors within an allowable range that varies based on user subjectivity, leading to inefficiencies in reducing execution time and power consumption.

Innovation Solution

An information processing system with a controller, storage section, and branch determination section that dynamically controls approximation levels and executes either exact or approximate loop bodies with stochastic branch instructions, allowing stepwise control of computational accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If approximate processing is used to reduce execution time and power consumption, then processing speed and energy efficiency are improved, but calculation accuracy deteriorates and errors increase

Engineering Contradiction:

Improveprocessing speedVSAvoidcalculation accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent implements dynamic switching between exact and approximate processing modes based on runtime conditions. The system uses a branch determination section that probabilistically selects between first processing (exact) and second processing (approximate) based on an approximation level, allowing the processing mode to adapt dynamically rather than being static. This resolves the contradiction by enabling the system to use approximate processing when speed is prioritized and exact processing when accuracy is critical.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes the approximation level parameter during calculation execution to control the balance between accuracy and speed. The storage section holds an approximation level that can be modified, and the branch determination section uses this parameter to determine the probability of selecting approximate versus exact processing. By adjusting this parameter, the system can optimize performance for different operational requirements.

Inventive Principle:

Principle #35Parameter changes

2Use of energy by moving object

If approximate processing is used to reduce power consumption, then energy efficiency is improved, but reliability of calculation results deteriorates

Engineering Contradiction:

Improvepower consumptionVSAvoidreliability of calculation results

Core Design Contradiction:

Use of energy by moving objectVSReliability

Solution Approach 1:

The system dynamically adjusts the mix of approximate and exact processing based on power constraints and reliability requirements. When power consumption needs to be reduced, the system increases the probability of selecting approximate processing while maintaining acceptable reliability through controlled error rates. The approximation level can be adjusted to ensure reliability constraints are met while maximizing energy efficiency.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system incorporates feedback mechanisms to monitor calculation results and adjust the approximation level accordingly. The branch determination section uses the approximation level to probabilistically select processing modes, and the system can adjust this level based on feedback about result quality and error rates, ensuring reliability is maintained while optimizing power consumption.

Inventive Principle:

Principle #23Feedback

3Adaptability or versatility

If the approximation level is changed during calculation execution, then adaptability to different accuracy requirements is improved, but system complexity increases

Engineering Contradiction:

Improveadaptability to accuracy requirementsVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments the processing into distinct first processing (exact) and second processing (approximate) paths, with a branch determination section that selects between them. This segmentation allows independent optimization of each processing mode while maintaining adaptability through the branching mechanism. The storage section separately holds the approximation level parameter, further segmenting the complexity management.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The branch determination section serves multiple functions: it probabilistically selects between processing modes, implements the approximation level parameter, and enables dynamic adaptation to different accuracy requirements. This multi-functionality reduces overall system complexity by consolidating control logic into a single component rather than requiring separate control mechanisms for each processing mode.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12604079B2Information processing system and information processing method

Publication Date: 2026.04.14 THE UNIV OF TOKYO

AI summary

An information processing system has a storage unit, a determination unit, and a control unit. The storage unit stores an approximation level related to a calculation. The approximation level can be changed during execution of the calculation. During the execution of the calculation, the determination unit determines whether to carry out a first process or a second process according to a probability determined on the basis of the approximation level stored in the storage unit. The second process is an approximation process of the first process. The control unit carries out control so as to carry out the process determined by the determination unit.