UltraRAM Breakthrough Could Replace Your PC's RAM And Storage In One Subsystem
by
Zak Killian
—
Wednesday, January 12, 2022, 04:48 PM EDT
Modern computers have so many layers of storage. You have the registers where the processor stores its work, then a few levels of progressively slower and larger caches, your main system memory, and finally the ponderous primary storage of the system. All of this exists because primary storage is so slow that, without this hierarchy of RAM, the CPU would spend practically all of its time idle, simply waiting on data.
The fantasy scenario is a computer with primary storage so fast that there's no need for separate RAM and fixed storage. You can already do something like this by booting from removable storage, but the problem is that anything you don't save to the removable storage device is gone when you shut the machine down and it loses the contents of RAM. That's because the medium that we use for RAM, known as DRAM, is very fast, but also volatile. When it loses power, it loses data almost immediately.
What if we had a memory technology that was totally non-volatile, faster than DRAM, more durable than flash memory, and consumed a hundred times less power than either? It sounds like the stuff of science fiction, and frankly, the name that its creators chose isn't helping that impression: UltraRAM. Yes, indeed; despite the outlandish claims and silly name, UltraRAM is in fact real, and also really promising.
Created by engineers at Lancaster University in the UK, UltraRAM is a new storage medium that makes use of quantum tunneling effects. If you want the gritty details on the technology, you can read the research paper for yourself, but the layman's version is that it's a new type of storage that is fundamentally dissimilar from DRAM or flash memory. It's not an evolution of existing tech, but a whole new idea.
Diagram illustrating the process of fabricating UltraRAM on top of Silicon substrate.
The technology is still in fairly early stages. It's not exactly brand new, either; the scientists who came up with it first published their ideas in June 2019. Their initial successes fabricating UltraRAM cells on exotic Gallium-Arsenide substrate led them to attempt to develop a form of UltraRAM for silicon, and that's what the latest announcement is about, as the team has succeeded in doing so—albeit at a very large feature size.
When originally conceived, it was described as a new storage medium for Internet of Things devices, and the primary design goal was to create non-volatile memory with extremely low energy requirements. That's because some kinds of IoT devices may not have ready access to plentiful power. It's possible to work around this while using traditional DRAM and flash memory, but it requires complicated circuits and expensive components like supercapacitors.
Test data indicating consistent high performance even after many program/erase cycles.
Well, it seems like the group succeeded in that goal. Lancaster's scientists say that a finalized version of UltraRAM should have switching energy requirements "100 and 1000 times lower than DRAM and flash, respectively." That alone would be cause for interest in the technology, but the bold claims don't stop there. The researchers also say that UltraRAM should be able to retain its data without power "in excess of 1000 years," and that the medium shows no degradation over a million program-erase cycles.
So, UltraRAM is another durable non-volatile storage medium like 3D XPoint, right? Sure, it could probably be used that way, and may even come to market in that form first. Arguably the most exciting part of UltraRAM, though—at least for enthusiasts—is that the developers say that the switching performance of a mass-production version could be faster than DRAM.
Test data showing a plateau in charge loss after around 20,000 cycles.
To be clear, that remains to be confirmed. Extant UltraRAM prototypes are fabricated on a very large 20-micron feature size, and as a result the operating speed is pretty slow comparatively. That "faster than DRAM" expectations (as well as all of the other remarks on UltraRAM's capabilities) is based on extrapolation from observed values at that feature size, and assumes "ideal capacitive scaling." As we all know, the real world is rarely ideal.
Forgive us for being somewhat skeptical, but we're at least cautiously optimistic for the future of UltraRAM. Even if the performance promises don't quite pan out, we're still looking at an extremely low-power and durable non-volatile storage medium, which is certainly welcome given the high price of 3D XPoint SSDs (which probably aren't getting any cheaper anytime soon.) Hopefully Lancaster's researchers can get this tech to market sooner rather than later.
