Chinese fabless semiconductor startup Numemory is developing storage-class memory (SCM) chips using Optane-like phase change memory and crossbar technology.
5 × endurance compared to (presumably SLC) NAND sounds around the ballpark of Optane’s advertised performance, though I don’t know if 10 × read/write speeds can also be attributed to Optane. Optane was never that great at sequential I/O.
They claim great latency as well, but provided neither latency nor IOPS figures which is sus.
But they’ve seem to have also put in a lot of work in R&D and patents, so that can’t be that far off from showing something off within the next couple years.
From what I’ve heard, SK hynix is much further along with the technology, they’ve graduated from PCM to SSM for their cross-point portfolio.
The fact that numemory is fabless makes me think they are not in a position to ever scale the memory density to become profitable; there are materials required to make cross-point memory viable that I doubt a contract foundry would ever develop themselves.
I’m wondering if they’ve figured out how to create an Optane-like product without the Optane-like special fabrication. My understanding of the Optane failure was Intel’s production volume never reaching critical mass and they got tired of subsidizing it for years.
I reckon a considerably different approach will be needed this time around. Starting with not repeating Intel’s DUMB marketecture mistakes such as chipset vendor lock-in, whack-a-doodle PCIe connections, gimped products that only use 1 or 2 PCIe lanes (H10, M10, 800P I’m looking at you)…
The product is about memory, replacing or supplementing DRAM, so being basically indestructible is necessary. You don’t want your RAM to wear down and bits starting to fail.
If you sell every PMEM with a loss and there is no larger market paying a hefty premium while the competition (NAND) gets better every year, eroding most advantages 3D-XPoint had, it’s a desperate situation.
We now have 60TB NAND devices with 200+PBW values and vastly better latencies than 5 years ago as well as CXL memory devices. The market isn’t really developing in Optane’s favor.
I’m all for new kinds of memory. Especially if it can merge memory and storage and finally get rid of the latter all together.
I doubt people want to pay the price. DRAM is the fastest we have and NAND is pretty fast, faster than most software and even other hardware can handle…and dirt cheap in comparison.
fabless startup…I wish them luck to find the key to economically viable manufacturing. We need options. And we need technologies rocking the boat.