Threshold-Voltage Reservoir Circuit for Low-Power Neural Arithmetic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor devices face challenges in achieving low power consumption, high-speed operation, small area occupancy, and high reliability, particularly in neural network applications where existing technologies are limited by high calculation costs and complex layer structures.
Innovation Solution
A semiconductor device design incorporating a matrix arrangement of transistors with varying threshold voltages and channel lengths, utilizing oxide semiconductors, and employing a current mirror circuit to perform product and sum arithmetic operations, allowing for low power consumption and high-speed processing without the need for weight storage circuits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional neural network architectures are used, then pattern recognition and data processing capabilities are achieved, but power consumption increases and operation speed decreases
Solution Approach 1:
The patent extracts and eliminates the weight storage circuit from the neural network architecture, retaining only the essential computational elements (transistors with varying threshold voltages) to perform arithmetic operations, thereby reducing power consumption while maintaining computational capability
Solution Approach 2:
The patent utilizes transistors with different threshold voltages as fixed weights to perform arithmetic operations, changing the electrical parameter (threshold voltage) to encode weight values and enable computation without additional storage circuits
2Productivity
If complex layer structures are implemented for high-performance neural networks, then processing capability improves, but device area increases
Solution Approach 1:
The patent merges the weight storage function with the computational function by using transistors with different threshold voltages to simultaneously store weight information and perform arithmetic operations, eliminating the need for separate weight storage circuits and reducing device area
Solution Approach 2:
The transistor serves multiple functions: it acts as a weight element, a switch, and a computational unit simultaneously, allowing a single component to perform what traditionally required multiple separate components, thereby reducing overall device area
3Measurement precision
If traditional arithmetic circuits are used for neural network computations, then calculation accuracy is maintained, but device complexity increases
Solution Approach 1:
The patent changes the approach to weight representation by using continuous threshold voltage variations in transistors instead of discrete digital values, enabling analog computation that maintains accuracy while simplifying circuit structure
Solution Approach 2:
The patent replaces traditional digital arithmetic circuits with an analog system using current mirrors and transistors, where mathematical operations are performed through physical electrical phenomena rather than sequential logic operations
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
The design achieves low power consumption, high-speed operation, and compact size while maintaining reliability, facilitating efficient arithmetic processing and reducing the complexity of weight calculations in neural network applications.
Implementation Method 1
a third circuit (comparison circuit), having a function of outputting voltage corresponding to a difference between current flowing through the second wiring and current flowing through the third wiring
Data Source
AI summary
A novel semiconductor device is provided. In reservoir computing using an input layer, a reservoir layer, and an output layer, variation in threshold voltage between transistors is used as a weight used for product arithmetic processing. Two transistors are provided in one product arithmetic circuit and data u is supplied to gates of the two transistors. Drain current of each of the transistors is determined by the data u and the threshold voltage of the transistor. The difference between the drain currents corresponds to a product arithmetic result. The difference between the drain currents is converted into voltage to be output. A plurality of product arithmetic circuits are connected in parallel to form a product-sum arithmetic circuit.