3D NAND flash memory has become a cornerstone in contemporary storage systems due to its cost-effectiveness and high performance. However, its reliability is a paramount concern, given that it is prone to several types of errors as it ages with program/erase (P/E) cycles. Although 3D NAND flash memory is a high-performance storage, its performance tends to decrease with the increase in retention time and P/E cycles, resulting in it not meeting requirements adequately. This thesis merges multiple facets of research to present innovative methodologies that significantly mitigate the challenges posed by the reliability and performance of 3D NAND flash memory, focusing on reliability metrics, refresh operations, and read retry methods.

In addressing reliability, this thesis proposes a novel metric called valid window which serves as an accurate and stable measure of reliability compared to the commonly used raw bit error rate (RBER). The valid window illuminates the size of error regions between two neighboring levels, allowing us to determine whether the data can be correctly read with further read retry. This new metric is instrumental in developing an effective method to reduce the number of read retry operations by optimally adjusting the program operations of 3D NAND flash memories. Experiments conducted on actual 3D NAND flash chips substantiate the efficacy of the introduced method.

To counteract the challenges posed by frequent data refresh operations in 3D NAND flash memory, this study offers an innovative refresh scheme designed to minimize the frequency of such operations. The scheme exploits the optimal RBER (ORBER), developed from extensive evaluations and analyses on a set of real 3D NAND flash chips, to determine the precise timing for refresh operations. This approach significantly curtails unnecessary refresh operations, especially in the late-life stage of the flash memory, thereby extending the lifetime of 3D NAND flash memory. Experimental results reveal that the proposed method can reduce 75% of the P/E cycles consumed by refresh operations on average and enhance the lifetime by 2.5X with marginal overhead.

To further enhance the read performance of flash-based storage devices, this thesis presents a meticulously crafted methodology for read-retry methods. The methodology involves the development of tailored Read Retry Tables (RRT) for each flash model. These tables, derived from extensive characterization and modeling of real flash chips, consider read granularity and error behaviors, unlike current pre-defined RRTs. A dynamic read retry procedure, followed by a proximity-search method, is introduced for fine-tuning the read offsets, achieving near-zero retries and significantly reducing the average number of read retries, especially for aged flash chips.

In summary, this research presents innovative solutions to enhance the reliability and performance of 3D NAND flash memory. A new reliability metric, valid window, outperforms traditional reliability measures, reducing read retry operations. An innovative refresh mechanism, using the optimal RBER (ORBER), minimizes refresh actions, extending the memory's lifespan. Dynamic read retry method, with tailored Read Retry Tables (RRT), further enhances read performance, particularly in aged chips. Together, these methodologies present a holistic solution to the reliability and performance challenges faced by 3D NAND flash memory.