Storage Systems with Non-Volatile Memory Devices
NAND Flash memory has become an increasingly important component as a nonvolatile storage media because of its small size, shock resistance, and low power consumption. The NVRAM has non-volatile and byte addressable characteristics, so it can be used as both of main memory and main storage with low power and high speed features. Although the ultimate objective of NVRAM is to replace all the current memory devices to NVRAM with its gorgeous technology, it will not completely replace DRAM and flash memory for the next decade since the current state of the art technology for NVRAM has limitation such as high latency, short lifetime, and endurance problem in comparison with DRAM, and low density in comparison with flash memory. Thus, the study of hybrid memory architecture that uses all the memory devices including NVRAM, DRAM, and NAND flash memory is required. In this research, NVRAM based low- power and high performance operating system technology was investigated.
The advances of flash memory process geometry scaledown has lead dramatic capacity increase, however, as a result, the endurance of flash memory is severely degraded resulting in a excessive increase of data errors. To recover the errors occurred during runtime, well known error correction codes, and additional parity schemes like RAID technique are multiplexed and integrated into flash memory controller for embedded storage systems, which results in excessive computational overhead. In this research, we investigate efficient and system level error recovery scheme, with the combined scheme such as compression, deduplication, and error correction codes with BCH and LDPC.
High Performance Computing and Interconnect Network
We consider the development of the storage system architecture for high performance IO operation of NVMe-based PCIe SSD storage systems. The NVMe interface is emerging technology for storage interface which gives new standard between host and flash drives to support high speed IO request processing. The NVMe interfaced storage systems will substitute current interface in many areas in the near future. Also, the NVMe-based storage systems are suitable for big data systems to support very-fast data storing and retrieval of numerous IO requests. That is the reason why we consider to pursuit new research theme with NVMe interface. Specifically, I consider three parts of storage system design. We research about the coprocessor issues for the higher NVMe IO queue parallelism. Second, Operating Systems internal block layer will be studied to be designed for software-level IO queue to promote hardware IO queue. Lastly, we study about the DRAM management issue for flash-based storage device, such as cache architecture, page IO architecture, zero copy issues.
HPC(High Performance Computing) system can be setup of cluster system that connect with interconnect network for several computing nodes. Those kinds of interconnect network is important role for HPC. To develop its own high performance computing system not using well-known interconnect such as infiniband or ethernet switch, we can consider low power and high performance PCIe based interconnect technology. For the PCIe interconnect technologies, PCIe switch-based NTB interconnect technology couldnot be deployed due to the reason for chip vendors. So, we need to consider PCIe NTB based switchless interconnect network technology to replace or overcome the lack of PCIe NTB switch technology. We plan to do research and development for PCIe NTB based switchless interconnect network, with the utilization of multi-port PCIe NTB host adapter. Based on the PCIe NTB interconnect network, we do research how to support parallel programming model susch as OpenSHMEM programming interface to support the PGAS mechanism for PCIe NTB interconnect networks.
AI, Embedded and IoT Systems
The Internet of Things(IoT) is the extension of internet connectivity into physical devices and everyday objects. Embedded with electronics, Internet connectivity, and other forms of hardware (such as sensors), these devices can communicate and interact with others over the Internet, and they can be remotely monitored and controlled(Wikipedia). The IoT system is composed of four components including Services, Platform, Network, and Devices. Sensors and devices are connected to Platform via Network through IP-based network. The IoT Platform provides several functions such as discovery, device management, gathering information, user management, and so on. With those functions, IoT services can be implemented and user can deploy useful IoT services through mobile and web based applications. In this reasearch, we study about the developing IoT devices and service applications with RaspBerryPi or Arduino embedded devices and lots of sensors. Also, we investigate to setup efficient IoT System management and network protocol system with Embedded operating system and network protocol stacks such as MQTT or COAP IoT network protocol. Applications include IoT automobile system, intelligent indoor surveillance system, and so on.
In addition, computing technology that directly computes algorithms such as AI in IoT terminals, such as recent edge computing, is developing. In edge computing, the embedded system of a terminal requires a hardware accelerator capable of complex operations such as AI. We study the structure and implementation method of Deep Learning Accelerator that can operate efficient recognition intelligence algorithm such as CNN in Embedded Processor.
In-Storage Deep Learning Accelerator for Embedded Edge Device,
Funded by National Research Foundation(NRF), 2021.06. - 2024.02.
Virtual SoC Platform for Digital NFT Development,
Funded by Electronics and Telecommunications Research Institute(ETRI), 2022.08. - 2022.11.
Development on the PCIe Non-Transparent Bridge-based Storage System Operation and Management,
Funded by National Research Foundation(NRF), 2019.06. - 2022.02.
Development for Virtual Platform Architecture of RISCV-based Deep Learning Accelerator,
Funded by Genesys Logic, 2020.03-2021.10.
Development of Switchless Network Technology with Optical Communication-based System Bus,
Funded by KISTI, 2019.03. - 2019.10.
Development on NVMe-based Storage System for High Speed IO,
Funded by National Research Foundation(NRF), 2016.06. - 2019.05.
Development of NTB(Non-Transparent Bridge)-based Interconnect Communication Technology,
Funded by KISTI, 2018.05. - 2018.10.
Global Multi-modal Knowledge Research Center,
Funded by National Research Foundation(NRF), 2015.08. - 2017.02.
Design of Mobile Storage Hub System,
Funded by LG Electronics, 2013.11. - 2014.05.
Development on Android Mobile Application for SNS Sytem,
Funded by Yoosoo Soft. Co., LTD., 2014.01-2015.12.
Developing NVMe Linux Device Driver,
Funded by BStech Co., LTD., 2013.07. - 2013.12.
Development on HTML5-based Application and Portal System,
Funded by Yoosoo Soft. Co., LTD., 2012.09-2013.12
Development of FTL for Next Generation NAND in Mobile Platform,
Funded by Small and Medium Business Administration(SME), Korea, 2011.06. - 2012.05.
Development of Next Generation NVRAM based Low Power and High Performance Operating System,
Funded by National Research Foundation(NRF), 2010.09. - 2015.08.
Hybrid Storage Architecture with Next generation Memory for Embedded Systems,
Funded by Samsung Electronics, 2006. - 2007.
Embedded NAND Flash File System Design,
Funded by Samsung Electronics, 2005.
Ubiquitous Fashionable Computer,
Funded by Ministry of Information and Communication(MIC), 2005. - 2008.
Low Power and Real Time Operating Systems for Embedded Systems,
Funded by Ministry of Information and Communication(MIC), 2003. - 2004.
A Parallel Internet Server System,
Funded by Ministry of Science and Technology(MST), 2001. - 2005.