Ising Model Optimization Device Circuit Architecture

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

As the number of spin bits increases in optimization devices using Ising models, the circuit scale expands, leading to increased circuit area and longer signal paths, which results in decreased operation speed due to signal transmission delays.

Innovation Solution

The optimization device includes a processor that calculates energy changes for bit inversions in an Ising model, selects bits to invert, and outputs signals directly to an energy calculation unit, reducing the need for additional signal paths and thus minimizing delays.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the number of spin bits is increased to solve larger optimization problems, then the problem-solving capability is improved, but the circuit area and signal path length increase, resulting in decreased operation speed

Engineering Contradiction:
Improveproblem-solving capabilityVSAvoidoperation speed
Core Design Contradiction:
Adaptability or versatilityVSSpeed

Solution Approach 1:

The patent divides the optimization device into multiple independent blocks, each containing spin bits and associated circuitry. This segmentation allows parallel processing of different subsets of spin bits, reducing the critical signal path length while maintaining the ability to handle large optimization problems through coordinated operation of multiple blocks.

Inventive Principle:
Principle #1Segmentation

2Quantity of substance

If the circuit scale is expanded to accommodate more spin bits, then the optimization capacity is improved, but the signal transmission delay increases due to longer signal paths

Engineering Contradiction:
Improvenumber of spin bitsVSAvoidsignal transmission delay
Core Design Contradiction:
Quantity of substanceVSLoss of time

Solution Approach 1:

The patent transitions from a one-dimensional linear arrangement of spin bits to a two-dimensional block-based architecture. By organizing spin bits into multiple blocks arranged in a grid-like structure, the signal path length is reduced as signals only need to traverse within or between adjacent blocks rather than across the entire circuit, effectively adding a spatial dimension to the architecture.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Ease of operation

If additional signal paths are added to manage bit inversion signals across the circuit, then the control capability is improved, but the circuit complexity and signal transmission delay increase

Engineering Contradiction:
Improvecontrol capabilityVSAvoidcircuit complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent combines the control logic for bit inversion directly within each block, merging the functions of local inversion decision-making and execution. This eliminates the need for separate centralized control signal paths that would span the entire circuit, reducing circuit complexity while maintaining full control capability through distributed autonomous block operation.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentUS11443090B2Optimization device and method of controlling optimization device
Publication Date: 2022.09.13 FUJITSU LTD
  • US11443090B2 patent drawing
  • US11443090B2 patent drawing
  • US11443090B2 patent drawing

AI summary

An optimization device includes: a memory; and a processor configured to: calculate, as bit operations, when any bit in a bit string representing a state of an Ising model is inverted, an energy change value of the Ising model based on a coefficient indicating magnitude of an interaction between an own bit and the inverted bit in the bit string; output a first signal indicating inversion availability of the own bit according to the energy change value and a second signal indicating the energy change value; select the bit to be inverted in the bit string and the energy change value corresponding to the bit based on the first signal and the second signal; output a fourth signal indicating the selected energy change value; and calculate energy of the Ising model based on the energy change value indicated by the fourth signal.