Neural Network Apparatus with Short-Cut Paths for Traffic Congestion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neuromorphic processors with large-scale transfer paths consume excessive power, leading to inefficient operation due to traffic congestion during parallel calculation and learning processes in neural networks.
Innovation Solution
A neural network apparatus with a tree path and short-cut paths connecting cores and routers, allowing data to be transferred efficiently and reducing congestion by providing alternative routes for data transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a large scale of transfer path is provided to avoid traffic congestion during parallel calculation and learning processes, then data transfer efficiency is improved, but power consumption increases significantly
Solution Approach 1:
The patent implements dynamic route selection where routers adaptively choose between tree paths and short-cut paths based on real-time traffic conditions. The system dynamically adjusts the data transfer route to balance load and minimize power consumption while maintaining efficient data transfer during parallel calculation and learning processes
Solution Approach 2:
The transfer path is segmented into two types: tree paths for normal data transfer and short-cut paths for bypassing congested areas. This segmentation allows the system to divide traffic flow and select appropriate paths based on current network conditions, reducing overall power consumption while maintaining productivity
2Productivity
If a large scale of transfer path is provided to support parallel forward and backward propagation processes, then processing capability is improved, but device complexity increases
Solution Approach 1:
The patent merges the tree path structure with short-cut paths to create a unified transfer system. This combination allows the system to maintain a relatively simple base structure while adding selective shortcuts, thereby improving parallel processing capability without proportionally increasing device complexity
3Use of energy by moving object
If data is transferred along tree path structure, then power consumption is reduced, but traffic congestion occurs during parallel processes
Solution Approach 1:
The short-cut paths act as intermediaries that bypass congested tree path segments. When congestion is detected in the tree path structure, data can be rerouted through short-cut paths, maintaining transfer efficiency without significantly increasing power consumption compared to using only tree paths
Data Source
AI summary
According to an embodiment, a neural network apparatus includes cores, routers, a tree path, and a short-cut path. The cores are provided according to leaves in a tree structure, each core serving as a circuit that performs calculation or processing for part of elements of the neural network. The routers are provided according to nodes other than the leaves in the tree structure. The tree path connects the cores and the routers such that data is transferred along the tree structure. The short-cut path connects part of the routers such that data is transferred on a route differing from the tree path. The routers transmit data output from each core to any of the cores serving as a transmission destination on one of routes in the tree path and the short-cut path such that the calculation or the processing is performed according to a structure of the neural network.