Magnetoresistance Multiply Accumulate Device Temperature Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks using magnetoresistance effect elements face performance deterioration due to temperature changes and asymmetric resistance values based on electric current direction, particularly during back-propagation learning, which reduces processing speed and accuracy.
Innovation Solution
A multiply and accumulate calculation device incorporating multiple calculation elements and reference elements with specific magnetization configurations, including magnetization free and fixed layers, and nonmagnetic layers, to detect and correct resistance value changes, ensuring stable operation across temperature variations and current directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If magnetoresistance effect elements are used for neural network calculations, then high-speed processing is achieved, but performance deteriorates due to temperature changes and asymmetric resistance values
Solution Approach 1:
The patent introduces reference elements with fixed resistance values to detect and compensate for temperature-induced parameter changes in magnetoresistance effect elements. By measuring resistance variations in reference elements and applying correction algorithms, the system maintains stable neural network performance despite temperature fluctuations and asymmetric resistance characteristics
Solution Approach 2:
The system implements a feedback mechanism where resistance values of reference elements are continuously measured and used to correct the resistance values of calculation elements. This closed-loop approach compensates for temperature drift and asymmetric resistance effects, ensuring reliable neural network calculations while maintaining high processing speed
2Measurement precision
If tunnel magnetoresistance effect elements are used to achieve higher resistance ratios, then calculation precision improves, but back-propagation learning becomes more complex and processing speed reduces
Solution Approach 1:
The patent introduces reference elements as intermediary components that mediate between the tunnel magnetoresistance effect elements and the control system. These reference elements provide a stable baseline for measuring and correcting resistance variations, simplifying the control of asymmetric resistance values and enabling efficient back-propagation learning while maintaining high calculation precision
3Quantity of substance
If asymmetric resistance values are utilized for weight encoding, then storage capacity increases, but correction operations become more complex
Solution Approach 1:
The patent performs preliminary measurement and correction of resistance values using reference elements before neural network calculations. By pre-characterizing the asymmetric resistance values of both calculation and reference elements, the system establishes correction factors that simplify subsequent weight encoding and decoding operations, maintaining high storage capacity while reducing operational complexity
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 solution minimizes performance deterioration of neural networks by accurately detecting and correcting resistance value changes, maintaining processing speed and accuracy despite temperature fluctuations and electric current direction variations.
Implementation Method 1
a magnetic neuron system using a magnetic domain wall-based variable magnetoresistance element using the fact that a resistance value changes continuously in accordance with a position of a magnetic domain wall within a magnetic recording layer
Implementation Method 2
a resistance value changes continuously in accordance with a position of a magnetic domain wall within a magnetic recording layer
Data Source
AI summary
A multiply and accumulate calculation device includes a multiple calculation unit and a accumulate calculation unit. The multiple calculation unit includes a plurality of multiple calculation elements, which are variable resistance elements, and at least one reference element. The accumulate calculation unit includes an output detector configured to detect a total value of at least outputs from the plurality of multiple calculation elements. Each of the plurality of multiple calculation elements is a magnetoresistance effect element including a magnetized free layer having a magnetic domain wall, a magnetization fixed layer in which a magnetization direction is fixed, and a nonmagnetic layer sandwiched between the magnetized free layer and the magnetized fixed layer. The reference element is a reference magnetoresistance effect element having a magnetization free layer that does not have the magnetic domain wall.