Arithmetic Device Magnetoresistive Elements Write Error Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current arithmetic devices face challenges in optimizing computational processing for AI deep learning, particularly in reducing power consumption while maintaining error tolerance, especially in neural networks like convolutional neural networks.
Innovation Solution
The device incorporates a control circuit that sets conditions for write operations in magnetoresistive effect elements based on information related to write errors, using a combination of first and second computational circuits with magnetoresistive effect elements on separate conducting layers, and executes computational processing using signals from these circuits to control voltage and current settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If write operations are performed with high voltage/current to ensure data accuracy, then computational accuracy is improved, but power consumption increases
Solution Approach 1:
The patent dynamically adjusts write operation parameters (voltage, current, pulse width) based on the specific computational context, data patterns, and error tolerance requirements. This allows optimization of power consumption while maintaining sufficient computational accuracy for each operation.
Solution Approach 2:
The system performs write operations with just sufficient strength to achieve the required accuracy level, avoiding excessive voltage/current application. For operations where high precision is not critical, reduced-strength write operations are used to save power.
2Use of energy by moving object
If error tolerance is increased to reduce computational requirements, then power consumption is reduced, but computational precision deteriorates
Solution Approach 1:
The system dynamically adapts error tolerance levels based on the computational task requirements, data importance, and current power constraints. Error tolerance is not fixed but adjusted in real-time to balance power consumption and precision needs.
Solution Approach 2:
Different error tolerance levels are applied to different computational operations or data elements based on their importance. Critical computations maintain high precision with low error tolerance, while less critical operations can tolerate higher errors for power savings.
3Productivity
If multiple computational circuits are used to improve processing capability, then computational performance is improved, but device complexity increases
Solution Approach 1:
Multiple computational circuits are merged into an integrated architecture where they share common resources such as control logic, memory interfaces, and power management units. This reduces overall device complexity while maintaining enhanced computational performance through parallel processing.
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 power consumption while maintaining computational accuracy by optimizing write operations based on error tolerance, enhancing the overall performance of arithmetic devices in neural network processing.
Implementation Method 1
Arithmetic device having magnetoresistive effect elements
Data Source
AI summary
According to one embodiment, an arithmetic device includes a first computational circuit including a first string, the first string having a first magnetoresistive effect element on a first conducting layer; a second computational circuit including a second strings, the second string having second magnetoresistive effect element on a second conducting layer; a third computational circuit executing computational processing using a first signal from the first computational circuit and a second signal from the second computational circuit; and a control circuit. The control circuit sets a condition on write operations with respect to at least one of the first and second magnetoresistive effect elements, based on information related to write error in at least one of the first and second magnetoresistive effect elements.