Paper Title: Using Machine Learning Techniques to Reduce SSD Costs

Paper Abstract: Previous research has shown that a machine learning based error recovery scheme can lead to great improvement in SSD endurance and throughput. The scheme also works very well under different failure modes and operating conditions. New research applies the technique to client level flash. The endurance and latency comparisons between enterprise level flash and client level show that designers can reduce costs by using client level flash without degrading performance significantly.

Paper Author: Cloud Zeng, Technical Supervisor, Phison Electronics

Author Bio: Cloud Zeng is an Associate Project Manager at the LiteOn/NVM Laboratory, where he works on error control coding, signal processing, and machine learning. He has focused on developing error handling schemes that can enhance the reliability and reduce the recovery latency of NAND flash memory. He reported the Error Recovery research based on Machine Learning at Flash Memory Summit from 2016~2018 and show great improvement in endurance and throughput. He earned an MSEE and BSEE (Honors Program) from the National Chiao Tung University (Taiwan).