3 ms·

Memristors. Thus far, they have not been used very well, mostly due to 'lock in' effects. But their power savings are fairly intense (as compared to current sol

Memristors. Thus far, they have not been used very well, mostly due to 'lock in' effects. But their power savings are fairly intense (as compared to current solutions, about a factor of ~10^3 - 10^4). With AI and ML stuff literally heating up, those power savings will become more and more attractive. Eventually, a sea change will occur and memristive based computing will become the dominant one. Personally, I think that the synapse-like characteristics of memristors aren't as fully exploited as they should be and new chip designs will greatly aide the transition to better and better AIs and ML devices on less power and smaller form factors. Think the modern image classification on a Tesla but powered by a watch battery, for months.

MiracleUser 7y ago
The wikipedia page seems to imply we still dont know if any memristors physically exist. Your comment implies it's in experimentation, so I am assuming wiki is out of date I would love to learn more about this topic, do you have any additional information / sources?
- Balgair 7y ago
  The Talk page on the wiki has some activity from this year, so it may be up to date - ish. There are a few companies that are selling memristors [1], but, as you say, they may not be fully truthful in describing their products. I'd take a look at some of the texts that are for sale [2] to dive deeper, provided you already know a fair bit of EE. If you do not, pick up this book [3] to get a good education. I may not be the best resource on this though. I'd email Dr. Chua directly and just ask him [4] for some resources, or just go through his publications list on it [5]. If you do email him, try for breaks in semesters, as Profs tend to be less busy then. Also, he has had 25 PhD students, so you can reach out to them as well. [1] https://knowm.org/product/ https://knowm.org/product/ [2] https://www.amazon.com/memristor/s?k=memristor https://www.amazon.com/memristor/s?k=memristor [3] https://www.amazon.com/s?k=the+art+of+electronics&crid=T6LE1KKXENK8&sprefix=the+art+of+ele%2Caps%2C177&ref=nb_sb_ss_i_1_14 https://www.amazon.com/s?k=the+art+of+electronics&crid=T6LE1... [4] chua@eecs.berkeley.edu [5] https://people.eecs.berkeley.edu/~chua/circuitrefs.html https://people.eecs.berkeley.edu/~chua/circuitrefs.html
  - MiracleUser 7y ago
    Amazing reply. Thank you. I see Dr Chua is dropping a new text in December. I'll see what I can learn by then and if it sticks I'll support with a purchase
    - p1esk 7y ago
      Don’t bother. I work at a lab where some of the largest and best memristor based crossbars have been developed, and I can tell you it’s almost all hype at this point. There are significant problems with memristive circuits, and very little advantage over something a lot more mature such as NOR flash crossbars (we work on both). There are many competing device technologies to use for analog computation, it’s far from clear which one is better.
      - MiracleUser 7y ago
        Ah bummer! I'll just keep an eye on it then. Anolog computation is where my interest is. I have a personal project where I have a subproblem of trying to use analog hardware to create a digital "impression" for processing. Terrible description, sorry

MiracleUser 7y ago

The wikipedia page seems to imply we still dont know if any memristors physically exist. Your comment implies it's in experimentation, so I am assuming wiki is out of date I would love to learn more about this topic, do you have any additional information / sources?

Balgair 7y ago
The Talk page on the wiki has some activity from this year, so it may be up to date - ish. There are a few companies that are selling memristors [1], but, as you say, they may not be fully truthful in describing their products. I'd take a look at some of the texts that are for sale [2] to dive deeper, provided you already know a fair bit of EE. If you do not, pick up this book [3] to get a good education. I may not be the best resource on this though. I'd email Dr. Chua directly and just ask him [4] for some resources, or just go through his publications list on it [5]. If you do email him, try for breaks in semesters, as Profs tend to be less busy then. Also, he has had 25 PhD students, so you can reach out to them as well. [1] https://knowm.org/product/ https://knowm.org/product/ [2] https://www.amazon.com/memristor/s?k=memristor https://www.amazon.com/memristor/s?k=memristor [3] https://www.amazon.com/s?k=the+art+of+electronics&crid=T6LE1KKXENK8&sprefix=the+art+of+ele%2Caps%2C177&ref=nb_sb_ss_i_1_14 https://www.amazon.com/s?k=the+art+of+electronics&crid=T6LE1... [4] chua@eecs.berkeley.edu [5] https://people.eecs.berkeley.edu/~chua/circuitrefs.html https://people.eecs.berkeley.edu/~chua/circuitrefs.html
- MiracleUser 7y ago
  Amazing reply. Thank you. I see Dr Chua is dropping a new text in December. I'll see what I can learn by then and if it sticks I'll support with a purchase
  - p1esk 7y ago
    Don’t bother. I work at a lab where some of the largest and best memristor based crossbars have been developed, and I can tell you it’s almost all hype at this point. There are significant problems with memristive circuits, and very little advantage over something a lot more mature such as NOR flash crossbars (we work on both). There are many competing device technologies to use for analog computation, it’s far from clear which one is better.
    - MiracleUser 7y ago
      Ah bummer! I'll just keep an eye on it then. Anolog computation is where my interest is. I have a personal project where I have a subproblem of trying to use analog hardware to create a digital "impression" for processing. Terrible description, sorry