Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.forth > #13492

Re: Phase Change Memory

Show all headers | View raw

Paul Rubin wrote:
> FRAM is great but I get the impression that it is expensive and low
> density compared to conventional ram or flash.  Its first deployment
> is
> in those TI microcontroloers containing just 16K or so of FRAM.  This
> PCM stuff is being apparently being made in devices containing 128 meg
> out of the gate.

Megabits, that's 16MB (more than good enough for many embedded 
applications, not nearly sufficient for Android or similar).  The 
maximum SDRAM chip capacity you can get now is 4Gb, i.e. 512MB (a factor 
of 32).  However, Micron's PCM is 90nm, and the feature size of 4Gb 
SDRAMs is 25nm (factor 13 for area); I don't know how the chip size 
relates (the PCM package is pretty small) - the cell is a 4F² design 
(smaller than a DRAM cell), as the phase change material is actually 
just implemented as via and scales down nicely for higher densities (and 
it is possible to stack several layers on one chip, like n vias for 2n 
metal layers).

Well, let's see how they make progress.  They have demonstrated a 1Gb 
45nm chip on ISSCC 2010 (that would be 128 meg), but this still hasn't 
become a product yet.  That's the latest paper I found, which is from 
last year:

http://www.epcos.org/library/papers/pdf_2011/Oral-Papers/S7-01.pdf

The read performance with 266MB/s is already slower than SDRAM (though 
not really bad), but the write performance with just 9MB/s is way too 
slow to replace SDRAM.

> I do worry about data retention at high temperatures.  The blurb says
> that the heat of soldering a chip to a circuit board erases the data.
> I'd like to think heating the pins doesn't cause the actual chip die
> to get all that hot.

This is a BGA, probably with the balls going straight to the die ("chip 
scale package"), so yes, the chip actually might get that hot.  The TSOP 
should not be that sensitive.

> So I worry about high ambient temperatures erasing
> the data.

They specifiy data retention of 10 years at 70°C, comparable to Flash.  
Their highest spec'd product is -40..85°C, so not for automotive or 
military applications.

Like Ti's FRAM, this is a pilot product.  It has good density for the 
process it is made in, it's a real product, and they can learn from it.  
With currently about 100 million writes per cell, it is not good enough 
to replace SDRAM for all possible purposes, so it's more a drop-in 
replacement for NOR flash.

-- 
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://bernd-paysan.de/

Back to comp.lang.forth | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

Phase Change Memory Jason Damisch <jasondamisch@yahoo.com> - 2012-07-02 08:39 -0700
  Re: Phase Change Memory "Rod Pemberton" <do_not_have@notemailnot.cmm> - 2012-07-02 19:45 -0400
    Re: Phase Change Memory Mark Wills <markrobertwills@yahoo.co.uk> - 2012-07-03 03:29 -0700
      Re: Phase Change Memory Paul Rubin <no.email@nospam.invalid> - 2012-07-03 10:47 -0700
        Re: Phase Change Memory Bernd Paysan <bernd.paysan@gmx.de> - 2012-07-04 00:38 +0200
        Re: Phase Change Memory Albert van der Horst <albert@spenarnc.xs4all.nl> - 2012-07-04 02:06 +0000
          Re: Phase Change Memory Paul Rubin <no.email@nospam.invalid> - 2012-07-04 10:04 -0700
        Re: Phase Change Memory "Rod Pemberton" <do_not_have@notemailnot.cmm> - 2012-07-04 05:05 -0400
          Re: Phase Change Memory rickman <gnuarm@gmail.com> - 2012-07-04 12:33 -0700
            Re: Phase Change Memory "Rod Pemberton" <do_not_have@notemailnot.cmm> - 2012-07-04 21:48 -0400
              Re: Phase Change Memory rickman <gnuarm@gmail.com> - 2012-07-05 09:57 -0700

csiph-web