Tikalon Header

Analog Memory

July 6, 2011

Physicists enjoy the contemplation of thought experiments, which were known in my generation by their German name, Gedankenexperiment. Their popularity might lie in the fact that no physical labor is involved; or, that theorists can perform experiments without setting fire to the laboratory. In defense of the later idea, I reference the Pauli effect, the mysterious failure of laboratory equipment when theorists are nearby.

There's an interesting thought experiment that I read many years ago. It was apparently first published by
Martin Gardner in 1982,[1] so my source must have been derivative of his publication. I wrote about Gardner in a previous article (Martin Gardner, May 31, 2010). I haven't read Gardner's version, so I'll paraphrase the one that I read.

An
extraterrestrial studied Earth for many years, and he composes a report to bring home with him. Alien technology is wonderful (viz., Clarke's Third Law that "Any sufficiently advanced technology is indistinguishable from magic"), so this is how he does it. He converts his electronic report, in whatever character set aliens use, into one long binary number by concatenating all characters. He changes this long integer into a binary fraction by placing a decimal point in front of it. Taking this number to be a fraction of the length of a rod used for such purposes, he places a mark at the exact point that specifies this number.

Practical considerations aside, all that's needed to retrieve the report is to read the position of the mark and work backwards to the character string. Alas, we Earthlings can think of many reasons why this can't be done, but it might only be that we're just not smart enough to pull it off. This would be an ideal sort of memory, an analog memory.

I discussed
phase-change memory (PCM, or PCME to distinguish it from pulse-code modulation) in a previous article (Nano Memory, September 30, 2010). In such devices, a memory cell changes electrical resistance in response to a write command. Digital data are stored by calling the high resistance states a 1, and the low resistance states a 0. PCM has an advantage that its cell construction is very simple, and it can be built in a crossbar switch architecture.

Electrical resistance, of course, is an
analog quantity. With a little effort, it's possible to store discrete resistance values in a memory cell instead of just low and high resistance states. This is what researchers at IBM's Zurich laboratory, Yorktown Heights laboratory, and a facility in Burlington, Vermont, have done in their demonstration of the feasibility of this multi-level cell state PCM memory.[2-3] A paper on this work was presented at the 3rd IEEE International Memory Workshop, which was held May 22-25, 2011 in Monterey, California.[4]

IBM multi-level cell state PCM memory chip.Multi-level cell state PCM memory chip, as demonstrated by researchers at IBM's Zurich laboratory.

(IBM Photo by Michael Lowry)

Phase-change memory is a non-volatile memory technology that has some advantages over
flash memory. PCM has read/write times that are a hundred times faster than flash, and it's also stable over millions of read/write cycles.[2] Flash can usually endure only about 10,000 cycles, and flash must be altered in a "page" mode in which thousands of memory cells must be rewritten when the data in just a single cell needs to be modified.[3]

The IBM research team was able to write four distinct resistance states in their PCM memory cells using an iterative write cycle in which
voltage pulses are applied and the cell resistance is measured in a control loop to attain the desired resistance value.[2] This method slows the write process to a 10 μsec cycle, but this is still a hundred times faster than flash.[3] Although there is inevitable drift involved over an operating temperature range, the relative order of resistance values does not change.[2]

The IBM demonstration chip, shown in the above photograph, contains 200,000 phase-change memory cells in an array fabricated using a 90-
nm process. The chip has been operating for five months, so this seems to be a viable process. 200,000 cells means 800,000 bits, or nearly a megabit. IBM expects the technology to hit the market in 2016.[2-3]

References:

  1. Martin Gardner, "Aha! Gotcha: Paradoxes to Puzzle and Delight," W.H. Freeman & Company, April, 1982, 164 pages (via Amazon).
  2. Ari Entin and Chris Sciacca, "Made in IBM Labs: IBM Scientists Demonstrate Computer Memory Breakthrough," IBM Press Release, June 30, 2011
  3. Paul Livingtone, "IBM jumps big hurdle in phase-change memory development," R&D Magazine, June 30, 2011.
  4. N. Papandreou, H. Pozidis, T. Mittelholzer, G.F. Close, M. Breitwisch, C. Lam and E. Eleftheriou, "Drift-tolerant Multilevel Phase-Change Memory," 3rd IEEE International Memory Workshop, Monterey, California, May 25, 2011.