UltraRAM takes big step toward production after years of development

Shawn Knight

Posts: 15,818   +202
Staff member
Forward-looking: It's been years in the making but UltraRAM is finally showing meaningful progress towards commercial production. Earlier this summer, IQE plc – a leading supplier of advanced wafer products – completed a year-long project to develop a scalable epitaxy process for the promising memory technology. CEO Jutta Meier described the step as a milestone towards industrial production.

Profiled on TechSpot more than three years ago, UltraRAM aims to deliver the best features of RAM and NAND in a single package – without the shortcomings. The tech promises non-volatile DRAM-like speeds, exceptional endurance, and improved energy efficiency. A prototype tested at the time reportedly offered 1,000 years of data retention and more than 10 million program / erase cycles – 100 times better than flash. Some have dubbed it "universal memory."

Blending RAM and flash into a single product is not a new idea. Big-name companies including Intel and Samsung have taken a stab at similar tech in recent years but their efforts have not had much of an impact on the market thus far. Intel bailed on its Optane platform in 2022, although Samsung recently renewed its Z-NAND product to fill the gap.

Back in 2020, Kioxia and Western Digital joined forces to create an Optane competitor known as XL-FLASH. This past May, a custom PCIe 5.0 SSD based on that tech was shown at Computex boasting an impressive 3.5 million random read IOPS. The N3X SSD from InnoGrit affords sustained performance of up to 14 GB/s sequential reads and up to 12 GB/s sequential writes. Random write IOPS were clocked at 700,000 while read latency checked in at just 13 microseconds.

UltraRAM maker Quinas is now working to get pilot production up and running with foundries and strategic collaborators. There is still no guarantee the tech will be a success – it has to get through production first – but it is encouraging that manufacturers are not giving up on the idea of combining the two types of memory.

Permalink to story:

 
This stuff is going to probably be outrageously expensive if it ever even leaves the lab at all. I'm not yet convinced this company is going to survive given it's an academic project that has solely relied on grant money to remain in existence thus far. If the UK gov bails them out again enough to get a fab partner, then maybe this will have a chance of making it to market.
 
Erm, I'm out of my depth here, but "read latency checked in at just 13 microseconds"...
13 microseconds is 13,000 nanoseconds. Typical DDR5 latency is about 15ns.

This assumes my math is correct.

This is not a wise assumption.
 
It was about time to hear tech news like that.

Although I'm afraid it will take some time until this new tech will come to the consumers.
 
It was about time to hear tech news like that.

Although I'm afraid it will take some time until this new tech will come to the consumers.

If at all. How many potential standards have fallen by the wayside over the years? Unless the new standard/technology can show much better performance with little price increase it'll be dead in the water. Either that or relegated to niche' uses like AI server farms.
 
Back