Ising Optimization Data Processing With Windowed Weight Storage

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In large-scale discrete optimization problems with constraint conditions, the capacity of memory for holding weight coefficients is insufficient, leading to inefficient solution finding.

Innovation Solution

A data processing device that efficiently searches for solutions by storing a part of the weight coefficients and local fields in an internal memory, updating these fields based on external storage, and switching windows of state variables to reduce memory usage and enhance processing efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If all weight coefficients are stored in internal memory, then solution finding efficiency is improved, but memory capacity requirement increases excessively

Engineering Contradiction:
Improvesolution finding efficiencyVSAvoidmemory capacity
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent divides the complete set of weight coefficients into multiple windows, where each window contains weight coefficients for a specific subset of state variables. The internal memory stores only one window at a time, while other windows are stored in external storage. This segmentation allows the system to process large-scale optimization problems without requiring all weight coefficients to be simultaneously available in internal memory, thus resolving the contradiction between solution finding efficiency and memory capacity requirements.

Inventive Principle:
Principle #1Segmentation

2Speed

If weight coefficients are loaded into internal memory for processing, then calculation speed is improved, but memory usage increases

Engineering Contradiction:
Improvecalculation speedVSAvoidmemory usage
Core Design Contradiction:
SpeedVSQuantity of substance

Solution Approach 1:

The patent implements a dynamic window switching mechanism where the internal memory stores different windows of weight coefficients depending on which subset of state variables is currently being processed. The control unit dynamically selects and loads the appropriate window into internal memory based on the optimization problem's current state, allowing fast access to required weight coefficients while minimizing overall memory usage by reusing the same internal memory space for different processing stages.

Inventive Principle:
Principle #15Dynamics

3Loss of time

If the entire weight coefficient matrix is stored in internal memory, then access time is reduced, but device complexity increases

Engineering Contradiction:
Improveaccess timeVSAvoidmemory management complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The patent introduces an external storage device as an intermediary to hold the complete weight coefficient matrix, while internal memory stores only the currently needed window. The control unit manages the coordination between internal and external storage, loading appropriate windows into internal memory as needed. This intermediary approach reduces access time for required weight coefficients while avoiding the complexity of managing a complete in-memory matrix, as the external storage handles the bulk of the data management.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4636654A1Data processing device, data processing method, and program
Publication Date: 2025.10.22 FUJITSU LTD
  • EP4636654A1 patent drawingFigure 1
  • EP4636654A1 patent drawingFigure 2
  • EP4636654A1 patent drawingFigure 3

AI summary

A device that searches for a solution represented by a combination of values of state variables using an Ising-type function including the state variables and terms corresponding to constraint conditions, the device performing: retrieving from a storage device, a part of a first weight coefficient group between the state variables, and a part of a second weight coefficient group between each of the state variables and each of the constraint conditions; and executing, using the part of the first and second weight coefficient groups, first processing determinig whether a change in values of the first state variables that belong to a trial target portion is permitted, and second processing in which a first local field is updated using the first weight coefficients, a second local field is updated using the second weight coefficients, and the first local field corresponding to each of the first state variables is updated.