Market researchers big up NRAM
Silicon to carbon change could boost Nantero's business
A market-research report suggests carbon nanotube storage developer Nantero could be on the verge of a breakout after years of disappointment and struggle.
The trend-report-pusher suggests NRAM revenues could be $75m in 2018, $120m in 2019 and $850m in 2023, a 62.5 per cent CAGR.
We've covered the technology in a little detail in the Fujitsu licensing article and explain there how its basis is the movement of the ends of carbon nanotubes to connect or break current-carrying connections and so provide binary ones or zeros.
So far, NRAM has been positioned as a potential candidate resident of the DRAM-NAND gap, being faster than NAND, up to DRAM in speed, and with virtually unlimited endurance. So it is competing with Intel/Micron's XPoint, PCM, STT-RAM, and other candidate tech.
But the report makes it clear that NRAM can be faster than DRAM, as well as using less power, which would lead to longer battery life in mobile devices. Unlike DRAM, it does not need refresh power sent through it to maintain its state.
The report notes:
NRAM read and write consumes lower energy in comparison to flash or DRAM. Due to which NRAM is expected to have much longer battery life. In addition, it is expected to be much faster to write than flash and DRAM.
However its speed is gated by the interface used, DDR4 currently, and associated protocols. Since NRAM is non-volatile, it does not need, as the report emphasis, the "refresh" part of the DDR4 interface.
How fast is carbon nanotube memory?
Ann Jacoby, client solutions director at BCC Research, tells us: "[The] media (ie, carbon) has an intrinsic response time down to picoseconds. One needs to have an interface just as fast in order to take advantage of that speed. The first wave of products will use the standard DDR4 interface – so that's the DRAM speed. Some early application engineers are working to optimize NRAM for the specific tasks on their wish lists. These will be faster than DRAM."
A picosecond, a trillionth of a second, is 1/1,000th of a nanosecond. If we say DRAM access latency is in the order of 14ns then that would mean 14,000ps. To describe NRAM as having picosecond speed needs it to be lower than 1000ps. We haven't seen actual numbers and so cautiously assume that we are looking at speeds of 800ps – 999ps. It could be much lower, but we need to see numbers.
The report goes on to say NRAM can be made in logic factories as well as at mass-production semi-conductor fabs, and that, therefore, NRAM dies could be customised, unlike DRAM. It means NRAM customers, such as mobile phone manufacturers, could order NRAM chips designed specifically for their phones.
A downside is capacity. The report quotes Mark Webb from MKW Ventures Consulting as saying: "Initial reported plans for 2018 are [a] 256 Mbit range of density, which is significantly lower than we expect from RRAM and 3D Xpoint in 2018. As such, it would appear to be competing in the NOR, FRAM, MRAM, embedded memory space for the next few years. RRAM, NAND, Xpoint will fight it out for the 128 Gbit and higher market."
Productisation with this kind of tech is a long, long and rocky road. The BCC Research report suggests that NRAM could be about to make it, even though the initial market destination is going to be a niche.
The market researchers want $2,650 for a single user licence for the report. ®