Memory Controller Dynamic Offset Voltage ML Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory devices face challenges in ensuring the reliability and efficiency of read operations due to errors in data retrieval, which are not adequately corrected by conventional error correction methods, leading to suboptimal read voltage levels and increased latency.
Innovation Solution
A memory controller system that includes a buffer memory to store a table for static offset voltages and a machine learning model to generate dynamic offset voltages based on cell count information from on-chip valley search operations, allowing for optimized read voltage adjustments and table updates to improve read operation reliability and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional error correction methods are used, then error correction is attempted, but read operation reliability is insufficient and latency increases
Solution Approach 1:
The system performs on-chip valley search (OVS) operations before the actual read operation to detect threshold voltage distribution and determine optimal read voltage levels. By preparing the optimal read voltage in advance based on OVS detection information, the system avoids multiple retry reads and reduces latency while improving reliability.
Solution Approach 2:
The system uses machine learning models to process OVS detection information and generate dynamic offset voltages based on cell count information. This feedback mechanism continuously optimizes read voltage levels based on actual memory cell states, improving read reliability while minimizing the need for error correction retries that would increase latency.
2Measurement precision
If static offset voltage is used, then table storage is simple, but read voltage accuracy is insufficient
Solution Approach 1:
The system transitions from static offset voltage tables to dynamic offset voltage generation using machine learning models. The machine learning model dynamically adjusts offset voltages based on real-time cell count information from OVS operations, significantly improving read voltage accuracy while the complexity is managed through efficient model implementation.
Solution Approach 2:
The system changes the parameter representation from fixed static offset values to dynamic offset voltages generated by machine learning models based on cell count information. This parameter transformation enables adaptive read voltage adjustment that accurately reflects actual memory cell threshold voltage distributions.
3Reliability
If multiple read operations are performed, then error correction opportunity increases, but operation speed decreases
Solution Approach 1:
By performing OVS operations and determining optimal read voltages before the actual read operation, the system ensures that the first read attempt uses the most appropriate voltage level. This preliminary preparation maximizes the success rate of the initial read, reducing the need for multiple retry operations and maintaining high read speed while preserving error correction capability.
Solution Approach 2:
The system replaces the mechanical approach of sequential retry reads with a smarter approach using machine learning-based voltage optimization. Instead of simply retrying reads with default voltage levels, the system uses ML models to predict and apply optimal voltages, reducing the frequency of retries and improving overall read throughput while maintaining error correction effectiveness.
Data Source
AI summary
A memory device may determine cell count information from a threshold voltage distribution of memory cells and may determine a detection case based on the cell count information when an error in read data, received from the memory device performing a read operation is not corrected. A memory controller may control the memory device to execute a read operation using a development time determined in consideration of an offset voltage of a read voltage corresponding to the detection case. When an error in the read data is successfully corrected, the memory controller may update a table, stored in the memory controller, using a dynamic offset voltage obtained by inputting the cell count information to a machine learning model.