The last major transition in hard disk drive
was a breeze; the next one is turning into a nightmare.
Looking back, the transition to
perpendicular magnetic recording
(PMR) HDD technology was just so straightforward. Flip the recording elements upright and have them arranged in the recording medium like upright circular rows of needles with each tip being a collection of magnetic grains. It's given us several generations of development, with the number of grains reducing each time and the needles getting closer together. We now have 2 TB, 4-platter, 3.5-inch drives and 750 GB, 2-platter, 2.5-inch drives and more to come. But PMR HDD technology is heading toward a dead end of unreliability.
The problem is that as the PMR bits (the needles) get smaller and packed closer together, they become more susceptible to apparently random bit flipping, reversals of their magnetic charge, due to influence from neighbouring bits or temperature fluctuations. You can do a fair amount to combat this with error checking and correcting (ECC) but, as the areal density of a disk platter trends toward 1 Tb per square inch, the problem worsens to the point of insolvability.
Yet we still need more disk capacity. Flash can't replace disk, and we can't deal with phenomenal increases in data storage requirements by simply adding more spindles, even if they do hold deduplicated data in thinly provisioned arrays. So the HDD industry has been looking into alternatives. Time is short; the highest areal density used on 2.5-inch drives today is 541 Gb per square inch. Two more generations should take that to 800 Gbit per square inch or beyond, say, by 2013 or so, and then PMR technology comes to a shuddering halt.
Four alternatives have been identified. One, called
(BPM), involves surrounding the bits at the 1 Tb per square inch areal density level and beyond, with insulating rings to prevent bit flipping. Making this stuff is theoretically possible, and it involves using fantastically complex lithographic machines that don't exist yet to etch the recording medium with billions of nanoscale pits. The medium itself is then self-assembled from a chemical stew, using the etched surface as a base, and the resulting platter is used as a master platter to imprint production disk drive platters with a nanoscale set of features.
Suppliers like Intevac, Molecular Imprints and Obducat are speculatively developing machine tools to do this, and some HDD manufacturers are building research prototype systems to produce the platters and heads. Toshiba has had some success in building platters, producing one with a theoretical 2.5 Tb per square inch areal density. However, head technology has not advanced as fast and heads can only detect tracks, not read and write the bits in the tracks.
Seagate has been researching and developing
heat-assisted magnetic recording
(HAMR) HDD technology, in which the recording medium is deliberately made much more resistant to random bit flipping. The read/write head has an added laser to heat each bit for a few picoseconds, overcoming its resistance to magnetic charge change and enabling the write elements to write data to the bit, which becomes resistant to further change as it cools. Imagine the precision needed by the read/write head as it locates bits, heats them and writes to them in picoseconds while the disk surface moves beneath it.
Read/write heads are often likened to jumbo jets flying inches above the ground. HAMR heads need the agility of a shark, as well as needle-point precision. The head electronics, as with BPM, have to read or write a signal to/from a bit in a vanishingly short amount of time and the techniques involved require skills at the Ph.D. level and beyond to research, devise and test.
A third technique is
discrete track media
(DTM) HDD technology, in which disk tracks are laid out much closer together. Again, the head servo controls need currently unavailable levels of accuracy, landing the jumbo jet on a pin head so to speak. DTM is thought to be a possibility after BPM.
A fourth technique is to cram more tracks on a disk, partially overlapping them, like tiles on a roof, and rely on digital signal processing techniques to read the data from the partially obscured tracks. This is called shingled writing, or
shingled magnetic recording
(SMR), and is being investigated by Hitachi Data Systems and, apparently, Western Digital. It could be used with current PMR recording technology and effectively give it another generational increase in areal density, delaying the time at which investment in BPM or HAMR will be needed. It is like adding more runways to an airport by partially overlapping them with existing runways, but still having planes land and take off safely.
The sums of money involved in developing HAMR or BPM disk drive production technology and facilities are enormous. There are only five HDD manufacturers: Hitachi GST, Samsung, Seagate, Toshiba and Western Digital. Their collective demand for next-gen test and production machinery is finite. If a hard disk drive tool supplier can sell to all five, its addressable market is larger than if it can sell only to a subset of the five.
Suppose Seagate pursues HAMR HHD technology and everyone else goes for BPM. The costs of the HAMR production and test tools will have to be recovered just by sales to Seagate, so costs will be much higher than if the kit were sold to all five manufacturers. Collectively, the entire HDD supply chain is facing a monumentally expensive transition to next-generation technologies and is beginning to doubt its ability to afford it -- if competing technologies are chosen by different suppliers.
, the industry's trade body, has set up a
Storage Technology Alliance
(STA), open to all IDEMA members, with Hitachi GST, Seagate and Western Digital as founder members. The STA aims to produce a roadmap for the HDD industry and its supply-chain members, so that the industry's collective development efforts are harnessed and go in the same direction, and university-type research projects are funded that support that direction.
The function of the STA is to take the financial sting out of the next-gen recording technology transition nightmare by deciding on a common roadmap. It's a multi-year effort, and the HDD supply-chain members need to see the bigger picture and play nice. Whether they will all do that remains to be seen; it could be like forming a coalition government in Italy, only worse. Que sera sera.
Chris Mellor is storage editor of The Register.