RRAM False Failure Prediction With Controlled Repair Voltage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current RRAM chip life prediction models lack accuracy, and existing repair methods for false failures exacerbate aging and reduce service life without effective prevention.
Innovation Solution
A method using a nine-dimensional data model trained with LSTM replicas to predict false failures in RRAM chips, incorporating forming and resetting voltages, resistances, and fluctuations, allowing for timely repair with controlled voltage excitation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical modeling of individual device is used for service life prediction, then the model can capture physical quantity variations and failure mechanisms inside memory unit, but the prediction accuracy for chip-level service life is insufficient due to ignoring process fluctuation variations across devices
Solution Approach 1:
The patent merges the physical model (capturing intrinsic failure mechanisms) with a data-driven model (capturing process fluctuations across devices). This combination allows the system to leverage both the physical understanding of failure modes and the statistical variations observed in actual chip operation, thereby achieving accurate chip-level service life prediction while maintaining manageable model complexity through modular architecture.
Solution Approach 2:
The patent introduces a data-driven component as an intermediary that bridges the gap between individual device physical modeling and chip-level prediction. This intermediary layer processes operational data from multiple devices and adjusts the physical model predictions to account for process fluctuations, enabling accurate chip-level service life prediction without requiring complete redesign of the underlying physical model.
2Reliability
If remediation is performed after memory unit failure occurs by simply increasing voltage and storage window, then the failure can be addressed, but the excessively high voltage accelerates aging and reduces service life of memory units
Solution Approach 1:
The patent performs preliminary classification of failure modes before actual failure occurs by analyzing operational data and identifying early signs of degradation. By classifying memory units as true failure or false failure candidates based on patterns in voltage, resistance, and operational history, the system can apply targeted remediation strategies in advance, preventing actual failure while avoiding the excessive voltage that would accelerate aging.
Solution Approach 2:
The patent changes the approach from uniform high-voltage remediation to parameter-adaptive remediation. Based on the classified failure mode, the system adjusts voltage levels and remediation parameters appropriately - applying lower, controlled voltage for false failure cases where conductive filament recovery is needed, and avoiding aggressive voltage increases that would accelerate aging and reduce service life.
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 method achieves 86.75% prediction accuracy, reducing error rates and extending RRAM chip life by preventing false failures without accelerating aging.
Implementation Method 1
Training a model M by the constructed training dataset, the model M comprises K LSTM replica models {M[0], M[1], . . . , M[K−1]} with the same parameters
Implementation Method 2
a resetting voltage greater than the average voltage is applied, to 'resurrect' it according to the physical mechanism of its failure, so that its conductive filaments can recover normal growth and fracture
Data Source
AI summary
The present application relates to a service life prediction and repair method for a resistive random access memory (RRAM) chip. The trained model is obtained by training the model using training dataset which is constructed comprises input data X, and bicategory labeling of true failure or false failure, process the measured data of the memory unit to be predicted into nine dimensional input data then input into the trained model, obtain the prediction results of the memory unit of the last few erase-write periods, and finally determine whether or not it is a false failure memory unit that actually needs to be repaired, then performing repair operation on the false failure memory unit that actually needs to be repaired.