T&L Publications

MrAl, Given how little we really know about these things, your motor/pot analogy is as good as any. However the output is determined by the input at all times making it more like just a non linear function. To be a memristor (by my definition) it needs some hysteresis. Maybe if you added some slop to the gear box or a band where the motor would turn but the pot was still stationary

Going back to the device described in the IEEE article. I like to think in terms of a curve trace. Imagine an X,Y graph where X is voltage and Y is current (Curve trace plot) Starting at the origin such a device would trace out a resistive characteristic going in the +, + direction (upper right quadrant) at some point the slope of this resistor would start to change (lower in this case) when the sweep voltage is reversed, the trace would proceed back to the origin at the new lower resistance slope. Operating the device below the current/voltage where it started changing would not alter the resistance. Now going in the -, - direction the slope would be for a low resistance until it reached the threshold again and then the characteristic would bend back to the higher resistance and stay that way as the voltage were reduced back to 0V again. Naturally there would be limits to how much the resistance could change. It appears they want to use this device in and out of saturation like a switch rather than as a non linear transfer function (which I described as near the point the curve bends) but that may yield interesting applications too.

Theoretically it may be possible to make one of these starting with a FET but the channel would need to trap the minority carriers (that allow conduction) and keep them there while the power was off then reject those carriers when an opposite gate voltage were applied. In practical terms it would be very hard to find a material that allows carrier mobility without having massive leakage when it was off. Remember charge carriers are holes and electrons which can move within a crystal lattice while a memristor requires that the atoms or ions become mobile. Flash memory is similar but requires several or more transistors and a very special capacitor like structure to replicate this function.

haklesup wrote:MrAl, Given how little we really know about these things, your motor/pot analogy is as good as any. However the output is determined by the input at all times making it more like just a non linear function. To be a memristor (by my definition) it needs some hysteresis. Maybe if you added some slop to the gear box or a band where the motor would turn but the pot was still stationary

The output is determined by the input history, which is different than a simple non linear circuit which depends
solely on the instantaneous input level.
Can you tell me where you got the idea that a memristor *has* to have hysteresis or it isnt really a memristor?
A memristor either has hysteresis or doesnt have hysteresis, but it's only an ideal memristor if it doesnt have hysteresis, right?

Theoretically it may be possible to make one of these starting with a FET but the channel would need to trap the minority carriers (that allow conduction) and keep them there while the power was off then reject those carriers when an opposite gate voltage were applied. In practical terms it would be very hard to find a material that allows carrier mobility without having massive leakage when it was off. Remember charge carriers are holes and electrons which can move within a crystal lattice while a memristor requires that the atoms or ions become mobile. Flash memory is similar but requires several or more transistors and a very special capacitor like structure to replicate this function.

I guess i havent yet limited myself to trying to come up with a device based on using only one already existing device (except
by accident with the overdriven thermistor example). I welcome using almost any combinations of existing circuit elements just
for the purpose of simulating a device. I do think it's also interesting that you are trying to simulate one using a *single* device
that already exists, and i wouldnt mind doing that either, but i dont think we should limit our scope of understanding about
what a memristor can be, in general, by forcing ourselves to do it with a single element until such time that we have access
to a chip manufacturing plant where we can design in almost any feature we feel like on any day of the week
However, an FET with a capacitor in some kind of circuit? The FET would provide the resistance part (as you wanted to do im
sure) and the capacitor would provide the charge history collection...just a thought...since you thought of the FET alone maybe
you could take the FET/Cap further and see what happens.

In both cases, I was mostly just musing about something we know little about and have little data to go on. My assumptions may have been unnecessary.

A FET kept biased by a capacitor is very similar to several non volatile Memory structures in the form of a floating gate structure
http://aplawrence.com/Makwana/nonvolmem.html

The novelty of the memristor is that it promises to do a similar trick with a simpler structure.

Hello again,


Well ok, good luck with finding one.

Before I put this to bed, two more references

http://en.wikipedia.org/wiki/Memristor
Wiki does a decent job at generalizing the definition, providing some math and summarizing the references as usual

In particular the first eternal link leads to a FAQ by the HP team
http://www.hpl.hp.com/news/2008/apr-jun ... r_faq.html

Hi Hackle,

I'll have to read that later today hopefully.

The only thing i have to disagree with is that "(the memristor) is something we know little about and have little data to go on".
I think we know a lot about it and have quite a bit of data from an electrical standpoint.

The other thing that bothers me a little is when i have read that the memristor would make a good
memory element that would replace transistors for each cell, and i hope that didnt mean "binary"
cell. That would be like saying that a 10k potentiometer makes either a good 10k resistor or
a good 0 ohms resistor, which leaves an almost limitless number of possible states unused.
The memristor could mean that computer memories could be made from 'analog' cells, which could
have a huge number of states. The simplest to picture would probably be where one memristor
takes the place of one *byte* of memory, having 256 (or more) states.

If the memristor works out, it could mean to analog signal processing what LSI did for digital.
Analog signal processing would broaden in itself too.

My only question now is can this new device (at least in theory) tell us something new about
the universe that we didnt know or didnt realize before this? After all, this device will yield
some very unique transfer functions.

Another view-
http://science.videosift.com/video/The- ... Hard-Drive
Fascinating discussion folks!

There was a short piece in NV a few months back.

Here's the link.

http://www.nutsvolts.com/index.php?/blo ... damentals/

Larry Lemieux