Semiconductor Variable-Resistance Elements for Reliable Neural Weight Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Variable resistance elements in semiconductor devices experience reliability issues due to variance in analog resistance values, which affects the reliability of neural network computations.
Innovation Solution
A semiconductor device with variable resistance elements that include a first electrode, a second electrode, and a variable resistance layer sandwiched between them, where the filament shape varies according to neural network weights, allowing for continuous resistance values to be stored and controlled through electric pulses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If variable resistance elements store analog resistance values as neural network weights, then the device can perform neural network computations, but the reliability of the semiconductor device is reduced due to variance in the analog resistance values
Solution Approach 1:
The patent applies parameter changes by controlling the forming process of filaments in variable resistance elements to reduce variance in resistance values. Specifically, the method involves adjusting forming conditions such as voltage, current, or pulse duration to create more uniform filament structures, thereby reducing the spread (variance) of resistance values across multiple elements while maintaining their analog nature for neural network weight storage
2Adaptability or versatility
If variable resistance elements with analog resistance values are used, then neural network weights can be stored, but variance in resistance values reduces device reliability
Solution Approach 1:
The patent modifies the resistance value distribution by changing the forming parameters of the variable resistance elements. This includes adjusting electrical stress conditions, pulse widths, or material composition during the filament formation process to achieve a narrower distribution of resistance values, thereby improving reliability while preserving the analog weight storage function
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
This approach reduces variance in resistance values, enhancing the reliability and data retention characteristics of the semiconductor device, enabling more accurate neural network computations with reduced power consumption.
Implementation Method 1
a variable resistance layer that is sandwiched between the first electrode and the second electrode and that stores a resistance value that is continuously variable
Data Source
AI summary
A semiconductor device includes variable resistance elements on a semiconductor substrate. Each of the variable resistance elements includes a first electrode, a second electrode, and a variable resistance layer that is sandwiched between the first electrode and the second electrode and that stores a resistance value that is continuously variable. The variable resistance layer includes a filament whose shape differs according to a neural network weight, and stores, as an analog value, the resistance value that is variable.