Hardware-Accelerated Memory Read-Level Selection Under Charge Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional memory systems face challenges in accurately reading data due to charge loss in memory cells, leading to increased read errors and inefficiencies in determining optimal read levels, particularly in higher density memory devices, which results in higher computational demands and energy consumption.
Innovation Solution
The method involves determining optimal read levels using a limited subset of memory cells with known reference data, employing a hardware accelerator to sweep read levels and calculate error counts, thereby bypassing iterative read levels and error correction codes to reduce computational load and energy consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional iterative read levels and error correction codes are used to determine optimal read levels, then read accuracy can be maintained, but computational load and energy consumption increase significantly
Solution Approach 1:
The patent extracts the read level determination process from the main data read path by using a separate calibration region containing reference data. This allows optimal read levels to be determined independently using a hardware accelerator, avoiding the need to process actual user data through complex iterative algorithms and error correction codes, thereby reducing energy consumption while maintaining read accuracy.
Solution Approach 2:
The patent performs read level calibration in advance by storing known reference data in a dedicated calibration region. The hardware accelerator uses this pre-prepared reference data to determine optimal read levels before actual data reading operations, eliminating the need for computationally intensive iterative processes during normal read operations.
2Reliability
If iterative read levels and error correction codes are employed to compensate for charge loss, then read errors can be corrected, but computational complexity increases
Solution Approach 1:
The patent introduces a hardware accelerator as an intermediary component specifically designed to perform read level determination. This dedicated hardware unit handles the computational complexity of analyzing voltage distributions and determining optimal read levels, separating this complex task from the main memory controller and reducing overall system computational burden while maintaining reliability.
Solution Approach 2:
The patent uses reference data that is a known copy or representation of expected memory cell states stored in a calibration region. By comparing actual reads against this known reference, the system can determine optimal read levels without needing to perform complex error correction on actual user data, simplifying the computational process while maintaining read accuracy.
3Measurement precision
If a large number of memory cells are scanned to determine optimal read levels, then accuracy improves, but time consumption increases
Solution Approach 1:
The patent segments the memory device into a calibration region containing reference data and a main data region. The read level determination process is performed separately on the calibration region using a hardware accelerator, which can quickly analyze the known reference patterns. This segmentation allows accurate read level determination without scanning the entire memory array, significantly reducing time consumption while maintaining accuracy.
Data Source
AI summary
Methods, systems, and apparatuses include retrieving reference data for a memory device. A read margin is scanned by determining sense data and pre-sense data for read levels of the read margin by applying a threshold voltage at each read level to cells of the memory device. A total error count is calculated by the memory device for each read level using the sense data, the pre-sense data, and the reference data. An optimal read level is selected for the read margin in response to comparing the total error count for each read level and determining a read level that has a lowest total error count.