KAIST Open up the memory era of Tera Bite(TB)

(From left) KAIST Electrical Engineering Professor Myoung-soo Jung Dr. Ji-ae Jang Mi-ryeong Kwon Ph.D. Dong-hyun Kook [Photo provided = KAIST]

Non-volatile memory (Memory that can have data even when a power outage by adding flash memory and a super capacitor to the existing DRAM before referred to as NVDIMM) and ultra-low-latency SSD (Semiconductor storage an SSD that has a very low latency by improving the existing SSD) are integrated as one memory a technology of memory that has improved performance and storage greatly better than future permanent memory (memory with data retention) which is combined into a memory and led by only a few global companies was developed by our researchers.

KAIST Electrical Engineering Professor Myoung-soo Jeongs research team (Computer Architecture and Operating System Lab) succeeded in developing Memory-over-Storage (MoS) technology that combines non-volatile memory and ultra-low-latency SSD into one memory space. It was released on the 16th.

This technology developed by Professor Jeongs team provides four times above terabyte (TB=1024GB) of storage capability compared to Intel Optane while providing volatile memory (DRAM) and it can fulfill similar user-level data processing speeds.

Existing NVDIMM has the advantage that the CPU can directly access the non-volatile memory without the help of the operating system. However the problem is that NVDIMM cannot process large amounts of data because it utilizes DRAM as it is and cannot increase the battery size indefinitely. Alternatives to solve this problem include Intel Optane DC PMM and Intel Memory Drive Technology. However these technologies require the help of the operating system whenever accessing the non-volatile memory so the reading/writing speed is 50% lower than that of the NVDIMM.

The MoS technology proposed by Professor Jeongs team uses an ultra-low-latency SSD as the main memory and uses the NVDIMM as a cache memory (A technique that copies data which are stored in a slow memory into a fast memory so that frequently used data can be quickly accessed). As a result the limitations of future permanent memory technologies have been completely improved by allowing users to use a large amount of SSD storage space as memory and at the same time obtaining performance similar to that of using NVDIMM exclusively.

MoS technology is known as the memory controller hub (Generally known as North Bridge hardware that helps the CPU access high-bandwidth devices such as memory or graphics processing units hereinafter referred to as MCH) inside the motherboard or CPU. Handles all memory requests. User requests are generally processed in the NVDIMM cache memory but data that is not stored in the NVDIMM must be read from the ultra-low-latency SSD. In the existing technologies the operating system handles these SSD readings while the developed MoS technology directly handles the SSD input and output inside the MCH so the input/output overhead (additionally requested time) of the operating system (OS) that occurs when accessing the ultra-low latency SSD while allowing the large capacity of the SSD to be used like regular memory.

The MoS technology developed this time by Professor Jeong succeeded a 110% increase in data reading/writing speed with a 45% reduction in energy consumption compared to a software-based memory drive or Optane permanent memory technology. As a result it is expected to be able to replace the existing memory/future permanent memory used in data centers and supercomputers which require a large amount of memory and are sensitive to system failures due to power outages.

Website: : http://www.wip-news.com/news/articleView.html?idxno=5857