Jun 30, 2014
Computer memory chips are manufactured identically to any other kind of integrated circuit. Wafers of ultra-pure silicon are selectively doped, masked with layer after layer of circuit diagrams, etched.. you get the picture. The extreme sensitivity of the process is one of the reasons behind the cost of microprocessors and memory these days. What if, however, there was a less touchy and expensive process? A research team lead by Der-Hsien Lien, a graduate student at the National University of Taiwan in Taipei figured out how to print memory circuitry on paper with an inkjet printer. The team fabricated a form of memory called resistive RAM, which is comprised of resistor/transistor pairs and is non-volatile (when the power is cut, the data doesn't evaporate). First, a piece of paper was screen printed with carbon paste which served as both a substrate and a common electrode. Then the paper was run through an inkjet printer which used a hacked cartridge containing a solution of titanium oxide nanoparticles instead of ink to print a layer of circuitry on top of the carbon layer. Then minute dots of silver were deposited strategically atop the titanium oxide layer (exactly how isn't clear) to form the final layer. The sheet of RAM was then subjected to various functional tests to demonstrate that it did, in fact function in the way rRAM is supposed to. The tests were successful, even after the paper was folded up and retested. Lien's team reports that they were able to fab memory cells 50 micrometers in size, smaller than the end of a human hair but as visible as a speck of dust floating in the air. The printer they used could print a single megabyte of rRAM on an A4 sheet of paper, but a printer with better resolution could hypothetically achieve data densities approaching one gigabyte of storage per page. More elabrate circuitry could probably increase that theoretical information density even farther; time will tell.
A couple of years back an obscure electrical component called a memristor became a topic of research in the field of computer engineering. Memristors work a little like a resistor and a little like a capacitor and hold a charge (representing a bit of data) for extremely long periods of time even after the power's been disconnected. A couple of years ago I wrote an article about HP initiating a research program to make memristors a practical part of computer architecture. Four years later John Sontag (VP, Director of Systems Research at HP Labs) announced that they're developing The Machine, an experimental computing platform that uses memristor-based RAM and novel optical interfaces to the silicon instead of metallic interconnects. The Machine is supposed to use much less power than comparable server-class machines and will boast many terabytes of storage online. They're working on a functional architecture for The Machine (what kind of data goes where and when) and Sontag says they're going to work with the Linux community to figure out how to use it in a practical manner. Very large in-memory data stores are already pretty common today so this doesn't sound too daunting a task, and using RAM as a file system is actually a very old but effective technique. It seems that one of the potential advantages of memristor-based memory is that it could potentially eliminate an entire class of programming problems, namely, determining which parts of a data set need to be cached because they're used heavily and which aren't and can be evicted to free up space. HP plans on offering memristor-based memory modules somewhen around 2016 and The Machine will hopefully be sold as a product three years later. Again, time will tell but they've come amazingly far in so short a period of time.