https://www.theregister.com/2022/08/02/kioxia_everspin_persistent_memory/

Last week Intel killed Optane. Today, Kioxia and Everspin announced comparable tech

19

Rumors of storage-class memory's demise may have been premature

Simon Sharwood

Tue 2 Aug 2022 // 07:57 UTC

Last week Intel killed its Optane storage-class memory product, because it just couldn't sell enough of it to make a difference to the bottom line.

But this week, two Intel rivals revealed advances in their own Optane-like tech.

Japan's Kioxia teased the second generation of its XL-FLASH Storage Class Memory.

In an announcement made ahead of the flash memory summit, which opens today, Kioxia said the update "achieves significant reduction in bit cost as a result of the addition of new multi-level cell functionality with 2-bit per cell, in addition to the single-level cell of the existing model."

Capacity of 256 gigabits will be possible once the memory is stacked – as is possible in two, four, or eight layers.

"In the future, it may also be possible to apply the product using Compute Express Link," (CXL) states Kioxia's announcement. That is a reasonably big deal, because CXL looks a lot like the future of pooled memory and composable architectures. If Kioxia can get persistent storage-class memory working with CXL, Intel's failed attempt to give the world a new tier of memory may yet have a chance of succeeding.

Upstart Everspin has also announced an advance in persistent memory – namely engineering samples of its newest spin-transfer torque magnetoresistive random access memory (STT-MRAM) that supports the Expanded SPI (xSPI) standard protocol to reach capacity of 128 Megabits.

"The combination of increased density with up to 233MB/sec full read and write bandwidth means that system designers now have the option of merging code and data memory on the same device, reducing cost, power, and area," Everspin claims, adding that it reckons the tech "will provide FPGA system designers with extremely fast configuration, instant-on boot capability, and rapid updates of critical application parameters such as weighting tables in AI applications."

• Why the end of Optane is bad news for all IT
• Samsung and Intel bosses discuss silicon cooperation
• Why Intel killed its Optane memory business

Why are Kioxia and Everspin pressing ahead with this stuff, given Optane failed?

For starters, both companies are memory specialists so they don't have much choice but to innovate in their chosen field.

Intel never fully bought into either the memory or storage businesses. Now that it's quit both, Kioxia and Everspin don't have to compete with Chipzilla and its enormous partner community … which of course failed to turn persistent storage-class memory into a thriving market.

Everspin's focus on FPGAs, however, means it's pursuing a niche where this class of memory will be easier to integrate than in the wider PC and server stacks. And Kioxia's hint of CXL adoption means it's tied its persistent efforts to a standard that plenty of players are behind, whereas Intel went it alone.

These two are not alone in believing the market is ready for storage class memory. Israeli/Australian outfit Weebit Nano recently tapped out demo chips for its own product and built them into a working memory module. Weebit Nano's chair is David Perlmutter, who once served as Intel's chief product officer. ®