Memristance Is !Futile
April 16, 2010 7:40 AM Subscribe
What's a simple mental model of a memristor?
Memristors are new enough that the only information you can really find talks about substrates and binding energies and blah blah blah. I'm looking for much more practical information, like how many leads does it have?
This video makes it look like a memristor would only have 2 leads and would function kind of like a potentiometer that adjusts itself.
But I think you'd be erasing your setting every time you read it if that were really true. So at minimum, there must be 3 leads. Pretty much exactly like a transistor, only you don't have to power the base to make current flow. It is only necessary that you once powered the base. Is that right?
I'm planning to build an actual, physical, working model of a memristor and I don't want it to be hilariously wrong.
Memristors are new enough that the only information you can really find talks about substrates and binding energies and blah blah blah. I'm looking for much more practical information, like how many leads does it have?
This video makes it look like a memristor would only have 2 leads and would function kind of like a potentiometer that adjusts itself.
But I think you'd be erasing your setting every time you read it if that were really true. So at minimum, there must be 3 leads. Pretty much exactly like a transistor, only you don't have to power the base to make current flow. It is only necessary that you once powered the base. Is that right?
I'm planning to build an actual, physical, working model of a memristor and I don't want it to be hilariously wrong.
How would you measure the stored resistance of a memristor without affecting its resistance?
posted by schmod at 8:19 AM on April 16, 2010
posted by schmod at 8:19 AM on April 16, 2010
Not an EE, but the mental model I've had is similar to clogged arteries: picture the memristor as a tube and the current passing through a tube as containing some kind of substance that will -- if left unchecked -- coat the surface of the tube over time, thereby thinning the effective diameter of the tube (and consequently increasing its resistance).
If I understand memristors correctly this model breaks in that reversing the direction of current flow ought to progressively de-clog the tube, but that's hard to see happening. Maybe you have like leaky minecarts that always drive facing the same way, so when they go in one direction they leak dirt (thinning the tub) and in the other direction they scoop it up (widening the tube).
In terms of 2-vs-3 leads: from what I have been able to surmise the way it works with 2 leads is that the memristor element has to be part of a larger circuit. As current goes through the memristor (in one of the directions) the resistance increases.
You then "read" the resistance value by sending through a tiny pulse of at a known voltage, and measuring the amount of current that makes it through; the pulse is small enough that it only makes a very small change in the memristor's current resistance.
This is an interesting question, and one I'm interested in as well. Hopefully someone who really knows this will step in and supply an awesome answer correcting any misconceptions.
posted by hoople at 8:20 AM on April 16, 2010
If I understand memristors correctly this model breaks in that reversing the direction of current flow ought to progressively de-clog the tube, but that's hard to see happening. Maybe you have like leaky minecarts that always drive facing the same way, so when they go in one direction they leak dirt (thinning the tub) and in the other direction they scoop it up (widening the tube).
In terms of 2-vs-3 leads: from what I have been able to surmise the way it works with 2 leads is that the memristor element has to be part of a larger circuit. As current goes through the memristor (in one of the directions) the resistance increases.
You then "read" the resistance value by sending through a tiny pulse of at a known voltage, and measuring the amount of current that makes it through; the pulse is small enough that it only makes a very small change in the memristor's current resistance.
This is an interesting question, and one I'm interested in as well. Hopefully someone who really knows this will step in and supply an awesome answer correcting any misconceptions.
posted by hoople at 8:20 AM on April 16, 2010
Response by poster: ...this model breaks in that reversing the direction of current flow ought to progressively de-clog the tube, but that's hard to see happening.
I've just been thinking of a potentiometer where instead of a person twiddling the knob the applied voltage itself is doing it. Then it works both directions. But I was thinking of there being 3 leads. If there are only two, the potentiometer model still works but makes it seem useless...
You then "read" the resistance value by sending through a tiny pulse of at a known voltage, and measuring the amount of current that makes it through; the pulse is small enough that it only makes a very small change in the memristor's current resistance.
I wondered if you could send a pulse through in alternating directions. First one way to read it, then the other way to undo the effect of the read. But that's waaaay more complicated and defeats the power and space savings that memristors offer.
Your idea sounds dangerous at first, but actually. The article I found mentions how memristors reminded Chua of synapses. And recently there was a RadioLab about how remembering things alters the memories. So yeah, basically it sounds like those are the two options: Every time you read the memory you either weaken it OR have to rewrite it.
posted by DU at 8:33 AM on April 16, 2010
I've just been thinking of a potentiometer where instead of a person twiddling the knob the applied voltage itself is doing it. Then it works both directions. But I was thinking of there being 3 leads. If there are only two, the potentiometer model still works but makes it seem useless...
You then "read" the resistance value by sending through a tiny pulse of at a known voltage, and measuring the amount of current that makes it through; the pulse is small enough that it only makes a very small change in the memristor's current resistance.
I wondered if you could send a pulse through in alternating directions. First one way to read it, then the other way to undo the effect of the read. But that's waaaay more complicated and defeats the power and space savings that memristors offer.
Your idea sounds dangerous at first, but actually. The article I found mentions how memristors reminded Chua of synapses. And recently there was a RadioLab about how remembering things alters the memories. So yeah, basically it sounds like those are the two options: Every time you read the memory you either weaken it OR have to rewrite it.
posted by DU at 8:33 AM on April 16, 2010
Response by poster: From later in the article:
posted by DU at 8:48 AM on April 16, 2010
The resistance of these devices stayed constant whether we turned off the voltage or just read their states (interrogating them with a voltage so small it left the resistance unchanged).So I guess there's some minimum turn-on voltage.
posted by DU at 8:48 AM on April 16, 2010
Seems so.
Even if there wasn't it doesn't seem that dangerous for storing digital data: define an arbitrary cutoff resistance (above: 1, below: 0), and the only danger would be that the state of the memristor is poised right at that threshold, so read a 0 but your test voltage pushes it into the 1 regime. If the rest of the design ensures that memristors are kept far away from the threshold at all times the danger of a read triggering a flip seems easily avoided.
posted by hoople at 9:04 AM on April 16, 2010
Even if there wasn't it doesn't seem that dangerous for storing digital data: define an arbitrary cutoff resistance (above: 1, below: 0), and the only danger would be that the state of the memristor is poised right at that threshold, so read a 0 but your test voltage pushes it into the 1 regime. If the rest of the design ensures that memristors are kept far away from the threshold at all times the danger of a read triggering a flip seems easily avoided.
posted by hoople at 9:04 AM on April 16, 2010
Regular dynamic RAM already has to be refreshed every time it's read (if I'm remembering correctly) so it doesn't seem like adding the same circuitry to memristor memory would be too burdensome.
Also, as others have alluded to, a memristor's resistance is proportional to the net current through it, so it could be read using alternating current. I don't see any reason why this would be particularly problematic; it's not like you'd need extra control circuitry at every single memory cell.
posted by teraflop at 10:10 AM on April 16, 2010
Also, as others have alluded to, a memristor's resistance is proportional to the net current through it, so it could be read using alternating current. I don't see any reason why this would be particularly problematic; it's not like you'd need extra control circuitry at every single memory cell.
posted by teraflop at 10:10 AM on April 16, 2010
It sounds like the hysteresis in the impedance of a memristor acts like the hysteresis in the magnetization of a ferromagnet. If you want to change the magnetization of a chunk of metal, you need to apply a strong field, and the changed magnetization persists after the strong field is gone. If you only apply a weak field the change in the magnetization is negligible, but your ferromagnet will still try to twist and align with the probe field, which is something you can measure.
posted by fantabulous timewaster at 10:43 AM on April 17, 2010 [1 favorite]
posted by fantabulous timewaster at 10:43 AM on April 17, 2010 [1 favorite]
This thread is closed to new comments.
posted by DU at 8:14 AM on April 16, 2010