Hard Drives Have A Future

As flash-based solid state disks have entered the mainstream, users and analysts alike have been contemplating the end of the hard disk era. Even my compatriot George Crump wondered last week when SSDs would dominate the data center. While flash has displaced spinning disks in some applications like MP3 players where small size and shock resistance are at a premium over capacity, I don't see the disk drives we all know and love going away any time soon.

Most arguments for the imminent demise of the spinning disks cite the exponential increases in solid state device density described by Moore's law which states that device density in integrated circuits doubles very two years. They then argue that while flash may cost 20-70 times as much as capacity oriented disks on a $/GB basis that Moore's law will drive the price of SSDs down so fast they'll be just 2-5 times the cost of spinning disks and at that cost no one will buy disk drives.  

The main problem with that argument is that it ignores the fact that hard drives have followed a similar path to higher and higher densities, known as Kryder;s law after former Seagate CTO Mark Kryder. Kryder's law states that disk bit density doubles annually. Even if flash SSD costs fall by a factor of three every two years while the cost of hard drive space continues to fall only by a factor of two, flash won't hit the 5x the cost target until 2018 or so.

Then there are the issues that may limit how small NAND flash geometries can be reduced. Chips in general are have reached geometries small enough that in addition to pushing the  limitations of lithography, manufacturers also have to consider quantum effects as they reduce the size of transistors and traces.

Flash, in no small part because of the high voltages needed for block erasure, is also susceptible to stress-induced leakage current that can cause cells to lose their charge--and your data--as nearby cells are erased or even written.  Smaller cell geometries means cells are closer together and therefore more vulnerable to this effect.Of course, hard disks have their own technology hurdles but developing technology like bit pattern media and heat assisted magnetic recording should get us through this decade and well into the next so we'll see 60-80TB drives around 2020.

The future's not so bright for performance-oriented drives.  I've already reached the stage where I don't understand why anyone buys 10K drives. 15K drives are the same price and  SSDs are already much cheaper than fast disks on a $/IOP basis.

It appears to me that the future is flash, or some exotic non-volatile memory technology using electron spin or glass phase changes, and trash (capacity oriented drives).  We just have to continue developing automated tiering, huge cache and other technologies to use solid state memory for cheap IOPS and spinning disks for sequential use patterns and capacity.