Apr 20, 2010
In the early 1970's an electrical component was hypothesized by Leon Chua, who was working at the University of California at Berkeley as an electrical engineer. Chua was said to be working on a mathematically rigorous foundation for the science of electronics, and during the course of his work he concluded that a fundamental component was missing. A memristor is essentially a component which remembers how much current has passed through it for a duration of time (technically, there is a relationship between the integrals over time t between current and voltage). While that doesn't seem all that interesting it actually is unusual insofar as solid state electronics are concerned. Memristors are really only feasible when constructed on the nanoscale (i.e., billionths of of a meter in size) because they operate based upon the movements of individual atoms subjected to an electric current, and they also act very much like neurons do in organic brains. Memristors that 'fire' often are more likely to fire in the future; those don't are less likely to do so. In 2008 a research team at HP fabricated the first practical memristor in their lab and just two years later they're working on practical applications for them.
Memristors are unique in that they don't lose their state when the power is turned off, unlike RAM, which would make them useful for building faster solid-state data storage devices. However, what really has people interested is that they seem to operate using analog rather than digital principles. Digital computers represent everything in terms of strings of two states, 0 or 1 (on or off, high or low). Analog computers represent everything using a precisely calibrated electric current; outputs are displayed using oscilloscopes or voltage meters. The upshot is that they're more accurate than binary representations of decimal values (for certain definitions of and limits for accurate, as always), but they're also much slower than digital logic circuitry, which is why they fell out of use by the late 1960's save in academia. Memristors, however, are now smaller and operating faster than conventionally fabbed transistors. HP's test models are approximately 3nm in size and can switch state in about one nanosecond if their published results are accurate. More surprisingly, memristors have been able to construct circuits in the lab that exhibit some properties formerly limited to organic structures: those circuits were trained to recognize patterns of signals passed through them and could successfully predict the next signals in a series. While this probably won't help you play the latest and greatest FPS there are implications for the fields of data mining, intelligence analysis, signal processing, and network security as well as potential applications in weak and strong AI... definitely a technology to keep your eye on.