Hard disks are getting close to the limit of perpendicular recording and will have to adopt HAMR or BPM. At least the disk vendors are used to big technology changes; AFAIK flash has been using the same floating gate cell all along.
This is linkbait with some technical jargon thrown in. The companies behind SSD technology - Samsung, Micron, IBM, Intel, etc are not sitting on their duffs waiting for the limits of their design to hit.
People have been claiming the end of DRAMs for years because the size of the storage cell (capacitor) does not scale well with the bitlines, etc.
There has been a bottleneck though in the speed of processors because of simple power consumption - that's when we switched to multicore processors. However, i think the overall computational power has been keeping track with moores law.
I've heard this all before, only about hard drive density, quite a number of times. They were always wrong, even as they are wrong now about flash memory.
The consensus today is that traditional semiconductor scaling is going to get iffy below 22nm for most devices. Flash is going to hit that limit first, because it's on the leading edge with component densities doubling every 12 months or so. No one knows what's going to happen at that point, but it seems like the most attractive option is 3D integration - building up, rather than out. All of today's chips consist of a single plane of transistors, connected by layers of metal wiring. It's possible to build multiple layers of transistors, but there are serious fabrication issues that haven't been solved yet.
This is about the fourth time I've lived thru one of those consensus end-of-the-road forecasts for semiconductor feature sizes. Each time hey have gone past with maybe a years delay. Of course, lithography canny be extended forever, though it seems hard to predict the end of new techniques.
As you point out,3disa way to go, and it has it's own problems, but there seems to be no hard physical limit to the number of layers one can add, presuming that defect rates are low enough.
I'm really just Wanting two more generations of ssd, so that a 500 gb costs enough to not feel guilty buying it. From then on, it's gravy.
I don't know where you got that from, do you have any sources? The 2009 edition ITRS concluded that there's no significant remaining roadblocks for 22nm processes. They listed some challenges for 16nm but they do expect to overcome all these obstacles in time to keep Moore's law going for at least another 5 years. No manufacturer that I know of have said that 22nm is going to be a limit. In fact Taiwanese researchers have already produced a 16nm SRAM chip in Dec 2009:
I said it will get difficult below 22nm, not that 22nm was itself a limit. Moore's law isn't going to end at a discrete point. What will happen is that each generation will become progressively more difficult to put into production, until alternatives like 3d integration are more practical. We're already seeing some of that with 45nm and 32nm, but sub-22nm is going to be very challenging.
12 comments
[ 3.2 ms ] story [ 31.3 ms ] threadPeople have been claiming the end of DRAMs for years because the size of the storage cell (capacitor) does not scale well with the bitlines, etc.
There has been a bottleneck though in the speed of processors because of simple power consumption - that's when we switched to multicore processors. However, i think the overall computational power has been keeping track with moores law.
As you point out,3disa way to go, and it has it's own problems, but there seems to be no hard physical limit to the number of layers one can add, presuming that defect rates are low enough.
I'm really just Wanting two more generations of ssd, so that a 500 gb costs enough to not feel guilty buying it. From then on, it's gravy.
http://www.taiwantoday.tw/ct.asp?xitem=87144&CtNode=416
My source is also the ITRS.