Weighting Device Multi-Level Storage Low Voltage Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network technologies face challenges in developing a weighting device capable of operating at low voltage and embodying multi-level weights, which is essential for simulating biological neural networks effectively.
Innovation Solution
A weighting device is designed with a substrate, transistors, a charge trap material layer, and a switching layer that can switch between low and high resistance states, allowing for the storage and reading of multi-level weights using a low voltage operation method involving selection, write, and erase voltages to control charge trapping and release.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional neural network weighting devices are used, then they can store weight information, but they require high voltage operation and cannot achieve multi-level weights effectively
Solution Approach 1:
The weighting device is divided into two separate transistors: a first transistor for writing weight information by controlling charge injection into the charge trap material layer, and a second transistor for reading weight information by sensing current flow. This segmentation allows each transistor to be optimized for its specific function, enabling low-voltage operation while maintaining reliable multi-level weight storage capability.
Solution Approach 2:
A charge trap material layer is introduced as an intermediary between the first and second transistors. This layer traps charges injected by the first transistor and modulates the threshold voltage of the second transistor, thereby storing weight information. The charge trap material layer enables multi-level weight storage through variable charge trapping, allowing the device to operate at low voltages while preserving weight information reliability.
2Measurement precision
If multi-level weights are implemented, then neural network simulation accuracy improves, but device complexity increases
Solution Approach 1:
Multi-level weight storage is achieved by varying the amount of charge trapped in the charge trap material layer, which continuously modulates the threshold voltage of the second transistor. By controlling the charge quantity parameter, the device can represent multiple weight levels without requiring multiple discrete components, thus achieving high measurement precision with relatively simple device structure.
3Quantity of substance
If charge trap material layer is used for weight storage, then multi-level weights are achieved, but charge leakage becomes a problem
Solution Approach 1:
The patent converts the potential harmful effect of charge leakage into a beneficial mechanism by using the charge trap material layer's inherent charge trapping and releasing characteristics. The layer can trap charges during write operations and release them during read operations, with the amount of trapped charge corresponding to the weight value. This controlled charge management transforms what could be a reliability issue into the core mechanism for multi-level weight storage.
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 device enables efficient storage and reading of multi-level weights, facilitating the operation of neural networks with reduced voltage requirements and preventing charge leakage, thus enhancing the simulation of biological neural networks.
Implementation Method 1
a charge trap material layer arranged on the switching layer and configured to trap or release charges according to a resistance state of the switching layer
Implementation Method 2
a switching layer arranged on the charge trap material layer and configured to switch between a low resistance state and a high resistance state
Data Source
AI summary
Provided are a weighting device that may be driven at a low voltage and is capable of embodying multi-level weights, a neural network, and a method of operating the weighting device. The weighting device includes a switching layer that may switch between a high resistance state and a low resistance state based on a voltage applied thereto and a charge trap material layer that traps or discharges charges according to a resistance state of the switching layer. The weighting device may be used for controlling a weight in a neural network.