The problem with PCM is that it isn't a drop in replacement for other memory types and it needs the system people to optimise for it. See for example: http://www.cs.rochester.edu/~ipek/cacm10.pdf
It seems systems designers are not yet ready to do so.
Good points everyone. In the end, potential dispruptive technologies like PCM/PRAM need to show substantial improvement that matters over today's technology. The jury is still out and that's why there's not much prodution today. Can PCM or any of the other contending "universal memories" be commercially viable for cost-efficient, volume production? Will they exceed the benefits of today’s technology by a degree large enough to justify customer and industry investment in rearchitecting? I think we are a long way away.
Actually, Numonyx 1Gbit 45nm device was supposed to ship in 2009. Then in early 2010. Then in late 2010. Now, never.
Samsung's PRAM was found in just one fake, planted, non-commercial handset (I mean one unit!) - since destroyed. Yes, the initial specs for Samsung GT-E2550 called for PRAM, but it turned out PRAM simply uses to much power and drains the battery, so the PRAM was quickly replaced back with NOR. No Samsung phone currently in production uses any PRAM. No other phones or any other commercial products use any PCM/PRAM either.
PCM/PRAM sucks. It is horribly overpriced, too slow and power-hungry in write, unreliable, with poor density, and it does not scale. It is the longest-running Techno-Ponzi scheme, but it is now coming to an end. Finally.
I am finding the statement that systems guys are slow in adopting PRAM for unknown reasons a little suspicious. I think if the advantage of PRAM was obvious everyone and his grandmother would be implementing those memories...Kris
Blog Doing Math in FPGAs Tom Burke 2 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...