'I was out in Korea just a couple of weeks ago': SK Hynix partner has a cunning plan to simplify SSDs and cut memory use by 99% amidst RAM crisis

• FTL-free SSD concept streams data sequentially with device-assigned addresses
• StreamFast claims thousand-fold reduction in memory needs for large SSDs
• Design targets cooler drives, lower power use, and simpler high-capacity storage

Hammerspace founder David Flynn is pitching a new SSD architecture that strips out the Flash Translation Layer and its controller DRAM, replacing both with a file-system-centric design called StreamFast.

According to Blocks & Files, the concept is being developed under a new StreamFast business working with the Open Flash Platform group, while Hammerspace itself will continue focusing on its system-level software.

Flynn says the current SSD model burns memory and power because controllers rely on an FTL stored in DRAM to track data locations.

The DRAM crunch

“It takes one byte of RAM for every kilobyte of flash on the SSD,” Flynn told Blocks & Files. “Think about that. If you’re going to have a petabyte flash on an SSD, that means you have to have a terabyte of DRAM with it.”

He links that overhead to the wider DRAM crisis, where manufacturers are moving capacity toward high-bandwidth memory for GPUs from companies like Nvidia and AMD.

The proposal is to remove the FTL entirely and let the file system interact with flash directly.

“We need to get rid of the block abstraction and move to something which is more native to flash,” Flynn says.

StreamFast instead uses device-assigned sequential addresses. The SSD writes incoming data streams one after another, then returns those addresses to the host file system.

“The magic is that the device assigns sequential addresses to arbitrary strings of data that are streamed to the device,” Flynn said.

Because writes are sequential, the host can replay the stream after a failure, rather than track every address in memory.

“With the StreamFast file system, it’s a byte of RAM for every megabyte of flash,” he said.

That’s a thousand-to-one improvement over the usual ratio. By Flynn’s own math, a 1PB SSD would need about 1GB of host memory instead of 1TB inside the drive.

Removing the FTL also cuts write amplification and reduces heat, since controller DRAM often forms the thermal hotspot.

“This simplifies the construction of the SSD to the point where it’s much more reliable,” Flynn said.

The company is working with partners across the flash ecosystem, although when Blocks & Files mentioned SK Hynix, Flynn was cagey.

“Can’t talk about specifics of our partnerships yet, but stay tuned. And I was out in Korea just a couple of weeks ago,” he said.

Flynn also claims the cooler, simpler drives could fit power-limited environments, including sealed or even orbital data centers.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!

And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.