Samsung claims key-value Z-SSD will be fastest flash ever

Among a blizzard of news from Samsung's Tech Data, El Reg has spotted smaller processor nodes, FPGAs added to SSDs, stacked and cubed memory, quad-level cell flash and object-storing SSDs on the way.

7LPP

Sammy has started fabbing 7LPP – 7nm extreme ultra-violet (EUV) low power plus lithography – wafers at its S3 Fab in Hwaseong, Korea, and claims to now have a clear path of 3nm processes.

7LPP replaces the prior 10nm FinFET process – it uses 13.5nm wavelength light to work on the silicon wafers. The chaebol said conventional argon fluoride (ArF) immersion tech uses 193nm wavelengths and requires multi-patterning mask sets.

Up to four ArF masks can be needed to complete a layer on the wafer whereas EUV only needs a single mask, saving time and cost, and delivering devices with increased performance, lower power needs and a smaller area.

Compared to 10nm FinFET, the 7LPP process delivers up to a 40 per cent increase in area efficiency with 20 per cent higher performance or up to 50 per cent lower power consumption, Samsung said.

SmartSSD

This SSD, currently being prototyped, has an added Xylinx Zynq FPGA accelerator for faster data processing and the ability to bypass server CPU limits, bringing compute to storage in other words.

Sammy claimed SmartSSDs will have higher processing performance, more virtual machines (VM), scalable performance, better de-duplication and compression, lower power usage and require fewer CPUs per system.

Wells Fargo senior analyst Aaron Rakers suggested the integration of the Xilinx FPGAs could provide a 2.8-3.3x increase in performance positioning SmartSSDs for AI, video, and storage-centric applications.

Other companies bringing compute to storage include NGD and ScaleFlux, which also uses Xylinx FPGAs.

Quad-level cell SSDs

Sammy said it will introduce quad-level cell (QLC, 4bits/cell) SSDs, offering 33 per cent more storage per cell than TLC (3bits/cell) SSDs. It claimed this will consolidate storage footprints and improve the total cost of ownership (TCO).

Unlike the consumer focus of Micron's Crucial, Western Digital's SanDisk business unit, and Intel's 660p QLC drives Sammy's will be for enterprise and data centre use.

Stacked RDIMM and cubed memory

Sammy has a 256GB 3DS (3-dimensional stacking) RDIMM, based on 10nm-class, 16-gigabit DDR4 DRAM that will double current maximum server DRAM capacity up to 16TB, and deliver higher performance and lower power consumption.

Its High Bandwidth Memory (HBM) 2 Aquabolt device has 8GB capacity and, Samsung claimed, provides the fastest data transmission speed and highest performance of any DRAM-based memory product on the market today at 307GB/s per HBM cube.

Key Value and Z-SSDs

Samsung's key-value (KV) SSD is also on the way, and the firm has reiterated its Z-SSD story.

KV-SSD is claimed to eliminates block storage inefficiency, reducing latency. Samsung said it allows server performance to scale even as CPU architectures max out, also providing a competitive TCO, improvements to write amplification factor (WAF), and scalability.

This reminds us of previous and failed KV devices, such as Seagate's failed Kinetic drives, and so far unproven devices such as Igneous's Arm-powered drives, plus similar devices developed by OpenIO.

A KV flash drive will be faster than a KV disk drive, but what actual customer problems would they solve?

Samsung claimed its Z-SSD will be the fastest flash memory ever introduced, with dual-port high availability, ultra-low latency and a U.2 form factor. Z-SSD will feature a PCIe Gen 4 interface with a 12GB/s sequential read rate, which is way faster than today's typical NVMe SSDs and their 2GB/s or so sequential read bandwidth.

The company will offer QLC-SSD and Z-SSD in a tiered storage package that should have a QLC capacity store front-ended with a fast Z-SSD tier. It’s still not explaining how its Z-SSD technology works though. ®