Reconfigurable Neural Network Engine for Resource-Constrained Self-Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network systems face challenges in minimizing circuit resources for self-learning mechanisms and are limited by fixed network configurations, making them unsuitable for real-time processing and sequential program processing.
Innovation Solution
A neural network system incorporating a von Neumann-type microprocessor that operates in multiple modes to recalculate or update weight and network configuration information, minimizing circuit resources and enabling reconfiguration for various purposes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed network configuration is used in neural network systems, then circuit resources are reduced, but adaptability to different processing tasks deteriorates
Solution Approach 1:
The patent implements a reconfigurable neural network engine that can dynamically change its network configuration between different processing modes. The system transitions from a fixed configuration to a dynamic one where the network topology and processing pathways can be reconfigured based on the specific task requirements, allowing the same hardware to adapt to different neural network architectures without requiring separate dedicated circuits for each configuration.
Solution Approach 2:
The patent creates a universal neural network engine that can perform multiple functions through reconfiguration. The same hardware infrastructure supports both real-time parallel processing and sequential program processing by changing the network configuration, eliminating the need for separate specialized circuits for different processing tasks and thereby reducing overall circuit resource requirements while maintaining high adaptability.
2Extent of automation
If dedicated circuit resources are allocated for self-learning mechanisms, then learning capability is improved, but device complexity increases
Solution Approach 1:
The patent merges the self-learning functionality with the existing neural network processing engine. Instead of implementing separate dedicated circuits for learning mechanisms, the system integrates weight update and network reconfiguration functions into the same processing engine that performs inference, allowing learning and processing to share the same hardware resources and reducing overall device complexity.
Solution Approach 2:
The neural network engine performs self-learning and self-reconfiguration using its own internal resources. The system can update its weight parameters and reconfigure its network topology autonomously without requiring external control circuits, enabling the learning mechanism to serve itself and thereby reducing the need for additional dedicated circuit resources.
3Speed
If hardware implementation is used for real-time processing, then processing speed is improved, but flexibility for sequential program processing deteriorates
Solution Approach 1:
The patent implements a dynamic operation mode system that allows the neural network engine to switch between real-time parallel processing mode and sequential program processing mode. The same hardware infrastructure can be reconfigured to support different processing paradigms, enabling fast parallel inference when needed while also supporting sequential program execution for tasks requiring structured control flow, thereby achieving both speed and flexibility.
Data Source
AI summary
A neural network system that can minimize circuit resources for constituting a self-learning mechanism and be reconfigured into network configurations suitable for various purposes includes a neural network engine that operates in a first and a second operation mode and performs an operation representing a characteristic determined by setting network configuration information and weight information with respect to the network configuration, and a von Neumann-type microprocessor that is connected to the neural network engine and performs a cooperative operation in accordance with the first or the second operation mode together with the neural network engine. The von Neumann-type microprocessor recalculates the weight information or remakes the configuration information as a cooperative operation according to the first operation mode, and sets or updates the configuration information or the weight information set in the neural network engine, as a cooperative operation according to the second operation mode.