For decades, researchers have searched for a memory architecture that could match or exceed DRAM’s performance without requiring constant refreshing. There’ve been a number of proposed technologies, including MRAM (in some cases), FeRAM, and phase change memories like Intel’s Optane. We’ve seen both NAND flash and Optane used as system memory in some specific cases, but typically only for workloads where providing a great deal of slower memory is more useful than a smaller pool of RAM with better access latencies and read/write speeds. What scientists want is a type of RAM that can accomplish both of these goals, offering DRAM-like speed and NAND or Optane-level non-volatility.
A group of UK scientists is basically claiming to have found one. UK III-V (named for the elements of the periodic table used in its construction), would supposedly use ~1 percent the power of current DRAM. It could serve as a replacement for both current non-volatile storage and DRAM itself, though the authors suggest it would currently be better utilized as a DRAM replacement, due to density considerations. NAND flash density is increasing rapidly courtesy of 3D stacking, and UK III-V hasn’t been implemented in a 3D stacked configuration.
According to the team, they could implement a DRAM replacement by using a NOR flash configuration. Unlike NAND flash, NOR flash is bit-addressable. In DRAM, the memory read process is destructive and removes the charge on an entire row when data is accessed. This doesn’t happen with UK III-V; the device can be written or erased without disturbing the data held in surrounding devices. This design, they predict, would perform at least equivalently to DRAM at a fraction the power
What the authors claim, in aggregate, is that they’ve developed a model for a III-V non-volatile RAM that operates at lower voltages than NAND, with better endurance and retention results. At the same time, these III-V semiconductors are capable of operating “virtually disturb-free at 10ns pulse durations, a similar speed to the volatile alternative, DRAM.” The three major features of the technology? It’s low-power, offers nondestructive reads, and is nonvolatile.
Right, But Will You Ever Be Able to Buy It?
Honest answer: I have no idea. The actual device hasn’t been fabricated yet, only simulated. The next step, presumably, would be demonstrating that the device works in practice as well as it does on paper. Even then, there’s no guarantee of any path to commercialization. I’ve been writing about advances in phase change memory, FeRAM, MRAM, and ReRAM for nearly eight years. It’s easy to look at this kind of timeline and dismiss the idea that we’ll ever bring a DRAM-replacement technology to market. The evolutionary cadence of product advances can obscure the fact that it often takes 15-20 years to take a new idea from first paper to commercial volume. OLEDs, EUV lithography, and FinFETs are all good examples of this trend. And new memory technologies absolutely have come to market in the recent past, including both NAND and Optane. Granted, Optane hasn’t completely proven itself in-market the way NAND has, but it’s also not nearly as old.
There are similarities between the difficulty of replacing DRAM and the trouble with finding new battery chemistries. In order to serve as a DRAM replacement, a new technology has to be able to hit better targets in terms of density, power consumption, cost, and performance than a highly optimized technology we’ve used for decades. We already have alternatives for every single individual characteristic of DRAM. SRAM is faster, Optane is higher density, MRAM uses less power, and NAND costs far less per gigabyte.
Similarly, we need battery technologies that hold more energy than Li-ion, are rechargeable, sustain original capacity over more charge cycles, charge more quickly, remain stable in a wide range of temperatures and operating conditions, and don’t explosively combine when breached in ways that make a Li-ion fire look like a Bic lighter. There’s a long road between theory and product. I will say that this team appears to think it’s solved more of the issues preventing a non-volatile DRAM replacement — but that, in turn, requires that it be easy to manufacture and cheap enough to interest the industry.
Top image credit: Getty Images
- Intel Confirms Its 22nm FinFET MRAM Is Production-Ready
- Spin Memory, ARM, Applied Materials Ink Joint MRAM Agreement
- Intel Releases Specs for Its Optane+QLC NAND H10 Memory