Ising Device Network Routing for Scalable Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for solving multivariable optimization problems using neural networks, such as Ising devices, face challenges in hardware implementation due to complex wiring and high resource requirements, making it difficult to realize large-scale calculations efficiently.
Innovation Solution
An information processing apparatus comprising multiple Ising devices connected via common buses, with a control device that manages neuron circuits, memory for connection information, and a router to facilitate efficient updating and noise management, allowing for scalable operations without the need for complex wiring or multiple chip configurations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple chips or processors are connected to establish a neural network for large-scale operations, then the calculation capability for large-scale optimization problems is improved, but the wiring becomes complicated and the preparation time and effort increase significantly
Solution Approach 1:
The system is divided into multiple Ising devices, each functioning as an independent neuron circuit unit. Each Ising device contains multiple neuron circuits that can independently process information, allowing the system to scale by simply adding more modular units rather than creating complex interconnections between traditional processors.
Solution Approach 2:
Each Ising device is designed with universal functionality to serve multiple purposes: it can act as both a neuron circuit and a processing unit, and can communicate with other Ising devices through standard interfaces. This multi-functionality eliminates the need for specialized wiring configurations for different types of connections.
2Productivity
If multiple chips or processors are connected to establish a neural network for large-scale operations, then the calculation capability for large-scale optimization problems is improved, but the preparation time and effort increase significantly
Solution Approach 1:
The system performs preliminary configuration by pre-establishing the connection topology and weight values in the memory of each Ising device before execution. The control device prepares the problem data and configures the neural network parameters in advance, so that when the calculation starts, the hardware is already ready to execute the optimization problem without requiring complex real-time setup.
3Adaptability or versatility
If different circuit configurations are used for chips located on the periphery or inside the array, then the neural network can accommodate varying connection requirements, but the manufacturing complexity and resource requirements increase
Solution Approach 1:
All Ising devices in the array use the same homogeneous circuit configuration and can perform the same neuron circuit functions. The system achieves adaptability not through hardware differentiation but through software-controlled connection routing and weight configuration, allowing any Ising device to connect to any other device regardless of physical position.
Data Source
AI summary
An individual ising device connected to common buses includes neuron circuits, a memory, and a router. The memory holds connection destination information per neuron circuit. An individual item of connection destination information includes first address information identifying one of a plurality of connection destination neuron circuits of a neuron circuit and second address information identifying a first ising device including at least one of the connection destination neuron circuits, the first and second address information being correlated. The router supplies, when an output signal of a connection destination neuron circuit changes and when the router receives the first address information identifying this connection destination neuron circuit via a bus, an update signal based on the changed output signal to one of the neuron circuits in accordance with the received first address information, the previously specified second address information identifying the first ising device, and the connection destination information.