Array-Based Memory May Put a Terabyte On a Chip
Lucas123 writes "A new type of flash memory, called array-based memory, could offer a terabyte of data on a single chip within the next decade by bypassing current NAND memory technology, which is limited by the miniaturization capability of lithography. According to the Computerworld story, start-up Nanochip Inc. is being backed by Intel and others, and over 11 years has made research breakthroughs that will enable it to deliver working prototypes to potential manufacturing partners next year. And by 2010, the first chips are expected to reach 100GB capacity."
We only need 50 GB, and in a form factor about a third of a deck of cards, and cheaper than a BRD. And it doesn't even have to be rewritable. Then, the tyranny of spinning disk media will finally be temporarily ameliorated, by USB high definition video players.
Imagine: not having to worry about your media obsoleting because the interface is so cheap and useful that it is guaranteed to be on every computing device long after it has been surpassed by superior buses.
Imagine a robust format that doesn't skip or scratch, even if you keep it in your pocket with your keys.
Imagine a built-in crypto chip ensuring strong DRM by essencially creating an encrypted ssh tunnel straight to your video display device, using a different key every time for the actual data.
(ok, the last one's maybe not so great, but there's no reason why anything with a usb connect can't have the crypto built in, so you'd still have your portability. If there's still a problem, then it's better to enforce the rules as perfectly as possible. People don't usually object to rules that don't affect them, witness the capricious speed laws for your example there)
Can you be Even More Awesome?!
This technology is essentially what is used in atomic force microscopes, and was being investigated by IBM for data storage under the name "Millipede". It basically involves a huge array of cantilevers that have very sharp tips on them (typical tip size is 20 nm but smaller is possible). The tips are used to read and write dots on the surface.
So yes, this system has moving parts. The tips have to scan across the surface, and the cantilevers are basically springs that bend up and down as the tips move over the surface. This definitely has some wear issues to consider, but it's nothing like the large-scale and high-speed movements of a hard drive (where a >2" disk is rotating at >7,000 rpm). Instead, the tips are moving laterally by micrometers at most (the huge array is what allows a large surface to be probed), and the cantilevers are springing up-and-down by only nanometers. The movement in an AFM is controlled using piezoelectric deformation of quartz actuators. This small-scale movement is very robust and reproducible. Quartz oscillators can vibrate/move thousands of times a second continuously for years without much problem (think of oscillators used for clocks, etc.). Moreover this technology has been used in commercial AFMs for years, so it's well-understood.
The thousands of tips are probably all actuated together by a single piezo-motor. They move in unison which would actually allow for high-speed reading/writing (since thousands of bits are read/written at once). You're right that each tip is in principle a point of failure. However, with the right error-correction algorithm, the device could be built so that even if a few tips break, no data is lost.
I agree that the access time isn't going to be as fast as modern RAM, but it could very well be faster than modern hard-drives. I think this is intended as permanent storage, not volatile memory.