Slashdot Mirror
World's Thinnest Flash Memory Cell Unveiled
qorkfiend writes "Measuring a scant 20 nanometers across, Infineon AG's new nonvolatile flash memory cell could lead to 32 gigabit flash chips within the next few years. The cell contains a unique structure with a fin for the transistor to avoid nano-scale physical effects and uses 90% less electrons than today's memory to store data."
They say 32 gigabit, not gigabyte. So if you divide 32 by 8, that makes for 4 gigabytes. At least, that's the way I understand bit-to-byte conversion.
You can lead a horse to water, but you can't make it dissolve.
The vulnerability can be compensated by error correction. Eventually we will reach an equilibrium where the error correction hardware is so big that we might as well use bigger memory cells, but right now it's the case that too many electrons are used.
If there is 90% less current, there is a 99% saving in Power, not 19%. The new current is 0.1I, so the new power is (0.1 I)^2 R = 0.01 I^2R
I'm not sure that 90% less electrons immediately leads to 90% less current, though. Everything else being equal, this is true, but perhaps other factors have changed as well.
Does this mean less resiliency/redundancy in the chip?
Yes - this is the primary reason that Intel is moving to OUM after the 45nm node (slide #32). Do note that this is still years off. OUM is rad-hard.
Also note that the research which is poured into XY-addressable OUM/chalcogenide memory can be potentially useful for the seek and scan memory that is also mentioned in that Intel presentation. My guess is that they'll come out with at leaset one variation or possibly both. The chalgogenide material is the same stuff used for RW optical media - you can change the phase via the application of energy (electrical, optical or otherwise). The change in phase causes many of the properties of the material to change, delineating unique, detectable states.
The probe storage is similar to a CD-RW but, instead of spinning the media below a single optical read/write mechanism, they are moving the media beneath thousands of atomic resolution probes that read/write with electrical energy. It is quite the technology.
HP says to expect it by 2006. Wow!
More
For "raw flash" a filesystem designed with wear distribution in mind is JFFS2.
And yeah, I concur with tmpfs for /tmp. I'd make it default for all distros.
The Raven
For everyone who doesn't know, four or so Infineon execs were just convicted of price fixing (the age-old German custom). Their arschen are going to jail! Finally the capitalists get punished for their asocial mischief :)
That's actually not that hard a problem to solve. Physically, it can't be done because of the way flash memory devices work, but provided you have enough spare memory (which is what growing capacities give you), you could certainly do some fancy work in software to rewrite a given memory bank only up to N times, and then move on to the next memory bank. You'll slowly lose storage capacity over time, but if you have a 32 Gbit device, you could create a 4 Gbit device that lasted 8 times as long, or for about 80,000 writes. Note that hard drive magnetic media also has a limited number of writes, although it's generally quite large. Still, this can cause bad sectors over time, which is why modern hard drives with their mind-boggingly capacities have spare sectors and remapping hardware to hide the problem.