This video single-handedly got me a lot more interested in electronics. What I study right now in the equivalent of high school includes a lot of electricity and circuit theory, both in physics and in pure electronic classes. I've had a bit of a hard time keeping up with those classes, because frankly they've been quite boring, but videos like this one made me interested from the beginning. This and your other videos remind me a lot of jan misali, probably because of the white text on a black background. Both of you are very good at teaching, and this specific style of video is great for learning, at least for me. This made me subscribe. Keep it up!
maybe the problem is that we allow the resistor to rewrite the resistance, and with enough redundancy and permanent write ability (disallow the erasure) we can create a learning mechanism for AI which allows for a unavoidable memory system that inherently includes a comparison model...that is given enough opportunity for the AI system to judge the pathways and determine the value of each probability for success based on those "memsistors" to determine success vs failure
Memristors definitely seem like they'd be useful for machine learning, in fact one of the reasons they're so promising is that they can combine processing and memory to act more like the brain. It might be more economical to keep the erase function so that a single chip could train more than one model. I'm not an ML expert though.
Interesting idea, I'd guess the limiting factor here is signal quality. It's much easier to quantize a signal like this with 2 levels (or 3 levels if we account for metastability) than for the N levels that would probably be needed for the comparison model. There's a lot of room for error here since we're working at distances of a few nanometers. I also think the speed at which the ADC/DAC used for read/write operations work would be a limiting factor. This is not to mention the other circuit design complications to allow for sufficient programmability of the connections between the memristors. At the end it might just be a lot faster and more realistic to do this at a higher abstraction level. Maybe a different implementation of the memristor would allow for this, the idea of analog memory is very interesting.
Well, knowm ships memristors in DIP6 Package. It has the same curve characteristic, besides digital also analog state retention tungsten doped devices. It's something . . .
error at 02:04 device cannot remember the amount of voltage that passed through it over time, because voltage cannot "pass" through a device. did you mean something else? this is confusing
It's been a while since I made this video so I don't remember the details off the top of my head, but you are right that voltage doesn't pass through circuit components. There's probably a better phrase to use - if I think of one (or if you have one to suggest) I can put up a correction flag.
@@SignoreGalilei i suggest to say that they remember the current that passed through them over time. its also not 100% accurate w.r.t memristor mechanics, but it makes much more sense from electronics stand point
Underrated explanation! The content and explanation are clear. Thanks for your help!
This video single-handedly got me a lot more interested in electronics. What I study right now in the equivalent of high school includes a lot of electricity and circuit theory, both in physics and in pure electronic classes. I've had a bit of a hard time keeping up with those classes, because frankly they've been quite boring, but videos like this one made me interested from the beginning.
This and your other videos remind me a lot of jan misali, probably because of the white text on a black background. Both of you are very good at teaching, and this specific style of video is great for learning, at least for me. This made me subscribe. Keep it up!
Thanks so much! It means a lot to know that I can get people interested in a topic.
3:33 - we apply voltage! not applying current.
I love you man, you raised my grade by at least a point, big love
maybe the problem is that we allow the resistor to rewrite the resistance, and with enough redundancy and permanent write ability (disallow the erasure) we can create a learning mechanism for AI which allows for a unavoidable memory system that inherently includes a comparison model...that is given enough opportunity for the AI system to judge the pathways and determine the value of each probability for success based on those "memsistors" to determine success vs failure
Memristors definitely seem like they'd be useful for machine learning, in fact one of the reasons they're so promising is that they can combine processing and memory to act more like the brain. It might be more economical to keep the erase function so that a single chip could train more than one model. I'm not an ML expert though.
Interesting idea, I'd guess the limiting factor here is signal quality. It's much easier to quantize a signal like this with 2 levels (or 3 levels if we account for metastability) than for the N levels that would probably be needed for the comparison model. There's a lot of room for error here since we're working at distances of a few nanometers. I also think the speed at which the ADC/DAC used for read/write operations work would be a limiting factor. This is not to mention the other circuit design complications to allow for sufficient programmability of the connections between the memristors. At the end it might just be a lot faster and more realistic to do this at a higher abstraction level. Maybe a different implementation of the memristor would allow for this, the idea of analog memory is very interesting.
Cool Bro Straight forward explanation
Well, knowm ships memristors in DIP6 Package. It has the same curve characteristic, besides digital also analog state retention tungsten doped devices. It's something . . .
error at 02:04
device cannot remember the amount of voltage that passed through it over time, because voltage cannot "pass" through a device. did you mean something else? this is confusing
It's been a while since I made this video so I don't remember the details off the top of my head, but you are right that voltage doesn't pass through circuit components. There's probably a better phrase to use - if I think of one (or if you have one to suggest) I can put up a correction flag.
@@SignoreGalilei i suggest to say that they remember the current that passed through them over time. its also not 100% accurate w.r.t memristor mechanics, but it makes much more sense from electronics stand point
top notch!
I wonder how well this would work for persistent storage.
That's one of the main areas of research right now, as far as I'm aware
Yes and i m currently reaserching for implementation in storage devices @SignoreGalilei
I wish i could get practical help
Awesome
Thanks!
Thnk You!