Anyway, while some of the articles are a bit dated (like all the industry pundits trying to forecast when the recession will be over--surprise they were mostly all wrong), the technology articles are especially interesting, with the added bonus that any promising technology I read about I can pop over to Google and do a two year followup on.
The latest technology was not only interesting, but it seemed more like something a blacksmith would come up with than a EE. Chalcogenide RAM is actually one part heat treated metal, one part spot welder, and one part volt-meter; and if they ever get it to work, its going to be some pretty good stuff.
So you start with a mix of germanium, antimony, tellurium & tin; which bakes up into a kind of spongy layer. Then you start heat treating it like you'd do steel.
As any good blacksmith knows, iron and carbon mixed together (steel) can do some weird things depending on the temperature. You can end up with all kinds of different arrangements of the carbon and iron atoms, and the resulting structures can have different properties. You can have austenite, cementite, pearlite, martensite, bainite, and all kinds of various sizes of ferrite (the iron crystal). Most of the time you're trying to either work out stress cracks (annealing) to regain strength and the ability to absorb shocks, or trading endurance for strength and hardening tools (heat treating). Since you can't have both at one time, there's always some give and take when working the metal.
The chalcogenide family has always been interesting to material engineers, because as you temper it and anneal it, you get other interesting physical property changes. Rewritable CD-RW disks make use of cholcogenides that change how reflective they are depending on whether their internal structure is crystalline or amorphous. In the case of cholcogenide RAM, the scientists actually found an alloy that changes its resistance. In the case of tin doped Ge2Sb2Te5, the resistivity dropped from 50kohms to 4kohms, which is pretty easy to tell apart.
The next challenge is building the on-chip spot welder. In the CD-RW disks you could use an external laser to melt the thing, but here the chip itself has to heat up a little bit of material to more than a thousand degrees and hopefully rather quick, as a memory with a write time measured in seconds (vs nano seconds) would not be that useful these days. Luckily the spot isn't very large, and material isn't very thick; but you're certainly not going to run the programming charge across aluminum, gold, or copper traces, as they'd probably melt as well. So here our old friend Tungsten arrives to save the day. Simply build the base pad out of Tungsten, and just for fun (since chip manufactures always love the challenge of dealing with strange new exotic alloys) make the top grid out of TiW (titanium tungsten, for those of you that don't have the periodic table memorized), and you're all set.
You don't even have to worry about large programming voltages like the early days of flash memory (or for those old enough: 15V & 24V EEPROMs). While EEPROMs and Flash work by zapping a charge across a thin insulator (thus requiring large voltages), melting stuff actually works great with low voltages. In fact, if you take the basic formulas for voltage ( V = I * R ) and power ( P = I * V ) and combine stick them together, you get P = I * I * R which means for any given resistance, to blow the maximum amount of energy we just want to crank up the current. That's why a spot welder actually runs at about half a volt. By taking line power at 240V (or higher for big machines), and turning it into 0.5V you increase the current from 25A to 12,000A which will dump a lot of energy (which turns to heat) into steel even though steel is an ok conductor. Its that current squared term that just goes ballistic.
So what's so great about chalcogenide memory? Its how the state is stored: by changing the crystal structure of the material, vs a electron charge or magnetic state. Basically its really hard to accidentally change the state. C-RAM would be immune to electric fields, alpha particles, magnetic fields or just about anything else that causes soft errors in modern storage devices. In fact an early use will probably be space applications that need immunity to all the high energy particles that hit equipment out beyond our protective atmosphere.
So what's happened in the last two years? There's been a few tidbits in the press since then, including one strategic report put out this last February by memory strategies showing at least seven groups working on it. (Two of them are even in my state: University of Arizona & Arizona State University. Too bad NAU didn't join in, I could just pop over after the holidays and see what's cooking.) Time will tell if C-RAM becomes the next big thing, or goes the way of bubble memory.