Fixed-Weighting Code Learning Device for Neural Network Area Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional analog-type neural network circuits require a large number of connection units to accommodate both positive and negative weightings, leading to increased circuit area and power consumption, as they need separate resistances for each, which doubles the occupied space.
Innovation Solution
A self-learning-type fixed-weighting-sign learning apparatus with an analog-type neural network circuit that uses separate positive-dedicated and negative-dedicated connection units, controlled by a self-learning mechanism to match input data, reducing the number of connection units and area occupied.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If separate resistances are provided for both positive weighting and negative weighting in each connection unit, then both positive and negative weighting functions can be achieved, but the circuit area (occupied space) is doubled
Solution Approach 1:
The connection units are segmented into two separate groups: positive-dedicated connection units for positive weighting and negative-dedicated connection units for negative weighting. This segmentation allows each group to use only one type of resistance, eliminating the need for dual resistances in each connection unit and thereby reducing the circuit area by half while maintaining both weighting capabilities at the system level.
2Adaptability or versatility
If the number of connection units is increased to accommodate all input-output neuron pairs in a full connection network, then complete neural network functionality is achieved, but the occupied area is further increased
Solution Approach 1:
The full connection neural network is segmented into two separate neural networks: one with positive-dedicated connection units handling positive weighting connections, and another with negative-dedicated connection units handling negative weighting connections. This segmentation maintains the complete n×m connection functionality while reducing the area per connection unit, as each connection unit now requires only one resistance instead of two.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Significantly reduces the area occupied by connection units while maintaining learning and generalization capabilities similar to conventional methods, improving the learning capability and efficiency of the neural network circuit.
Implementation Method 1
all the input-side neurons and all the output-side neurons are connected in a one-to-one manner at connection units each having a resistance value corresponding to weighting for each connection unit
Data Source
AI summary
A neural network circuit is provided with which it is possible to significantly reduce the area occupied by the connection unit of a full connection (FC)-type neural network circuit. An analog-type neural network circuit constitute a learning apparatus having a self-learning function and corresponding to a brain function, wherein the neural network comprises: a plurality (n) of input-side neurons; a plurality (m, and including cases when n=m) of output-side neurons; (n×m) connection units each connecting one input-side neuron and one output-side neuron; and a self-learning control unit, the (n×m) connection units being constituted from connection units corresponding to only the positive weighting function as a brain function, and connection units corresponding to only the negative weighting function as the brain function.