Wednesday, October 03, 2007

Digital storage technology

In 1983 personal computers (there were no laptop computers in those days) usually did not come with hard disks as standard equipment, but the disks could be bought as an add-on. I bought a 10 MB (megabyte) disk for about $200 back then.

In 1994 hard disks were standard equipment, but not too reliable. I had to buy a replacement disk for the computer I had then, when it was only a year old. The disk was 1000 MB = 1 GB (gigabyte), and still cost about $200. Today, for less than $200 one can buy a 500 GB disk, while larger disks (750 GB or 1000 GB = 1 TB (terabyte)) are available for about $300.

So there is nearly an improvement of 5 orders of magnitude in 24 years. Since 105 is about 216.6, that is 16.6 doublings in 24 years in disk capacity per dollar. Which is one doubling in about 1.5 years, or 18 months. And that's not even allowing for inflation – which would make the doubling time even less in terms of inflation-adjusted dollars. Since 1.61.5 is about 2, a doubling every 18 months (1.5 years), is roughly the same as an increase of 60% per year, while a doubling in 2 years is an increase of about 40% per year (since √2 ≅ 1.41).

Oddly enough, that's pretty close to what Moore's Law states for the doubling time for the number of transistors that can be fabricated in one integrated circuit chip. This number is usually figured to be between 18 and 24 months, but closer to the latter. (For Gordon Moore's latest thoughts on the "Law", see here, here, or here.)

However, I say "oddly", because the technology of hard disks is quite different, and largely independent of, the technology of semiconductor integrated circuits. This is important, because even Moore himself sees the end of his "Law" within 10 years, or maybe 15 years at most. If it lasted 10 more years, we would expect a factor of about 32 increase in the density of transistors on a chip. But that would imply the linear size of each transistor should decrease by a factor of 5 or 6. So the size of one transistor would have to shrink to just about 10 nm. That's only a few dozen silicon atoms, which implies big problems in quantum effects and fabrication difficulty of the chips.

The demise of Moore's Law presents the same problem for the density of digital storage that can be implemented in semiconductor chips. This includes "flash" memory, now used extensively in digital cameras, MP3 players, and similar gadgets.

The ultimate limits of data storage density using either silicon chips or magnetic media occur for the same reason – quantum effects that arise when the size of circuit elements or magnetic storage domains approach the size of a few dozen atoms, somewhere below 10 nm. But because of the difference in technology between silicon chips and the magnetic media used in hard disks, the limits need not be reached at the same time.

Data bits in magnetic media are not stored in silicon-based transistors, but in grains of magnetic material (originally iron oxide but more recently a cobalt-platinum alloy). With current technology it takes more than one grain to store a single bit – from 50 to 100 are actually required for acceptable reliability and absence of "noise". So without a change in technology, the only way to achieve higher density is through reduced grain size.

Unfortunately, grain sizes are already around 8 nm, and any further reduction with cobalt-platinum grains would not work, because the quantum effects in that material would permit the direction of magnetization to "flip" (i. e. reverse randomly) at room temperature.

What to do? Fortunately, serious problems are still a few years away, because one major change in technology is already going on – the switch from "horizontal" to "vertical" recording. Prior to the first shipment of disks using the latter technology in 2005, the 50 to 100 grains needed to store each bit were laid out horizontally (though not end-to-end) on the disk surface. But in 2005 disks employing "perpendicular recording" technology became commercially available from Toshiba. (See here.)

Perpendicular recording entails orienting the magnetic cell consisting of individual grains so that its long axis is perpendicular to the surface instead of parallel to it. The change was accomplished by redesigning the "head" that writes each bit by establishing a particular direction of magnetization. Successful implementation of perpendicular recording has not been either quick or easy, as the idea has been under investigation since the mid 70s.

The estimated maximum density of data on a disk using the older longitudinal recording technology is 100-200 gigabits per square inch, while the estimated maximum for perpendicular recording is 1000 gigabits (1 terabit) per square inch. So there is a potential increase in the maximum density by a factor of 5 to 10. But the full potential won't be realized immediately. The disk introduced by Toshiba in 2005 had a density of "only" 133 gigabits/in2. In 2006 Toshiba raised the density to 178.8 gigabits/in2 (see here).

Some experts estimate that it will be possible to keep improving densities with perpendicular recording at a rate of 50% per year. At that rate (which may be optimistic), the limit (1 terabit/in2) would be hit in 2010-2011. By comparison, the only real competitor at this time for practical, bulk, random-access, non-volatile, read/write digital data storage is silicon-based flash memory. As noted before, such technology has doubled in density about every 2 years, or 40% per year. Although flash memory is currently more expensive per bit than magnetic disk and has other disadvantages, as well as some advantages, the two types of memory are currently competitors. So there will be pressure on magnetic media to increase in density no less than 40% per year, and at that rate magnetic media would still reach its limits around 2011-2012.

Flash memory, on the other hand, might keep improving in density a little longer, until Moore's Law itself hits the wall somewhere around 2017. So there's a lot of pressure on magnetic media vendors to come up with new technology that is more than just incremental refinement, simply to stay in the race.

Is this possible or likely? Yes, as a matter of fact it is. But we'll save discussion of possible new technologies for magnetic media for another time, and also look at future data storage technologies that are silicon based, as well as other possible technologies – of which there are quite a few.


Industry divided over future of hard drives

Tags: ,


Links to this post:

Create a Link


Post a Comment

<< Home