Probe Memory Packs 138 Terabytes per Square Inch
spectrum.ieee.org
spectrum.ieee.org
New and cool: fully automated, and works well above room temperature (unlike more traditional cryogenic designs).
Expectedly impractical: very-very slow write speed, probably pretty slow read speed, because everything is mechanical. Only a couple dozen bytes actually written in the experiment.
With this density, you can just duplicate (or more) the data and still have an useful capacity. Parity based methods would do better, I guess.
That would be triplication. ;-) Duplication is only single error detection.
“Sometimes hydrogen atoms can be removed from incorrect locations.”
the memory would have cycles, a read 0 phase, any bit that needed to be flipped to one would be picked up, the array shifts to the left, any probe holding an hydrogen would then write. then a read 1 phase, which prepares for the zero writes.
Obviously this depends critically on arrays of probes, which might not be possible. If it is, there's no reason to think this can't be massively parallel.
vvvvvvvvvvv
'',,','''''
^^^^^^^^^^^
far far easier said than done i'm sure. :D
Hardrives are roughly 1Tb/in^2 in areal density right now. SSD's are a little better.
This one using hydrogen atoms is ~100x better.
Hydrogen atoms are pretty damn small. Aside from improvements with 2D packing factor, is this pretty much the limit?
It gives the maximum amount of data in space before collapsing into a black hole.
I find this useful in discussing solar power: there is an absolute maximum of 1300 watts per square meter; most people don't know this and implicitly assume substantially higher is achievable.
For the sorts of scales human beings are interested in, of course, the Bekenstein bound is a long long way from being relevant, and what matters for us in principle is volume rather than area. (There are practical difficulties in using "substantially" 3-dimensional regions for storage, but they have nothing to do with the danger of the storage device turning into a black hole.)
reference: https://physics.stackexchange.com/a/2283
In other words, we live in a hologram.
https://en.wikipedia.org/wiki/There%27s_Plenty_of_Room_at_th...
I am pretty sure 1Tb is 1 Terabit/in^2. Here it is 138 Terabytes / in^2, or 1104 Terabit/in^2, hence the 1000 times difference. Not 100
Western Digital MAMR allows up to 4 Terabit/in^2, so the difference is now slightly closer to ~250x. Then there is Bit pattern media, which should get us to 10 Terabit/in^2.
But it doesn't look like HDD manufacture are in any hurry, they will try to milk the market for as long as possible. The Total HDD market will fall to 50% of its peak volume sometimes next year or in 2020.
In terms of business isn't exactly good.
I would expect that the market will drive hard drive manufacturers to effectively zero profit.
This is a good thing for everybody.
So yeah the roughly 650Tbps seems usable in such a situation.
However, looked at another way, it has also never been and never will be true, if one defines "digital network" broadly enough to include any connection between the storage device and, say, a CPU.
Sneakernet or "a station wagon full of tapes" merely increased the bandwidth of one segment of that end-to-end connection. What's usually forgotten is that it doesn't do anything about the bandwidth of, for example, the tape drive (if as many as one are freely available for long enough).
In the more modern world, it can be easy to forget just how huge hard disks are, compared to how much they can transfer. A 12TB drive that can do 120MB/s would need 100k seconds (almost 28 hours) to transfer its entirety.
The situation is particularly severe with "spinning rust" but SSDs seem to be headed in that direction, as densities increase faster than interface speeds (even NVMe).
Tapes suffer from that big time. They've grown in storage space, comparable to hard drives, but since it's normal to have libraries with many tapes per drive, there's severe practical utility limits even for backups.