Talk:Memristor/Archive 2
This is an archive of past discussions about Memristor. Do not edit the contents of this page. If you wish to start a new discussion or revive an old one, please do so on the current talk page. |
Archive 1 | Archive 2 |
Current and Charge
There is a statement in the article "Of course, nonzero current implies instantaneously varying charge.". Would it be better to say time varying charge here instead?Arnob (talk) 17:42, 31 January 2009 (UTC)
- Maybe. The intent is, time varying charge at that point in time. If charge varies with anything, it must vary with time… maybe cut it down to simply "implies varying charge" or "implies that Q is changing"? Potatoswatter (talk) 20:00, 31 January 2009 (UTC)
- Charge could vary with position without varying with time (as in a nonuniform charge density) so the statement "if charge varies with anything, it must vary with time" is false. Blm19732008 (talk) 14:33, 1 February 2009 (UTC)
No, charge density can vary with position. That is different from charge varying with position. Potatoswatter (talk) 04:52, 4 February 2009 (UTC)
Of course the amount of charge can vary with position. For example in the basic model of atoms there are a different number of electrons associated with different shells as determined by quantum physics. Electrets also demonstrate charge variation with position. Your distinction between charge and charge density is ridiculous because there are many examples in electromagnetics textbooks of charge being written as a function of position (i.e. q(x,y,z)) such as in Gauss's law in integral form. —Preceding unsigned comment added by 129.174.74.102 (talk) 03:25, 6 February 2009 (UTC)
- As a general rule any physical quantity that can vary with time can also vary with position. In any case the original phrasing was confusing so I changed it as suggested by Arnob1. Blm19732008 (talk) 15:14, 6 February 2009 (UTC)
- To get from charge density to charge, you must integrate over a volume. This integration eliminates the position parameter. Electron wavefunctions vary probability with position, not charge. To say that a finite amount of charge exists at a single position would violate the uncertainty principle and would be nonsense. Just because a textbook uses q to represent charge density does not mean "the distinction is ridiculous." If the units are different, the concepts are different. It should be obvious that there is no such thing as position on a schematic diagram, and schematics are incapable of representing waves, which tie time and position together. Memristors (and the other 2-terminal components) exist only in schematic diagrams, where there is such a thing as a node or a terminal. Potatoswatter (talk) 16:27, 6 February 2009 (UTC)
- You are confusing yourself by thinking in terms of point charges and schematics. Circuit elements exist outside of schematics and are embodied by actual materials such as semiconductors. Semiconductors can be doped to produce a particular charge distribution in which different amounts of charge are located in different positions of the semiconductor providing a variation in the amount of charge with respect to position. In any case this argument is moot since I already changed the confusing section of the article.Blm19732008 (talk) 22:48, 6 February 2009 (UTC)
- Those "different amounts of charge" are still charge density, not eg Q(x,y,z) = 1.6*10^-6 Coulombs. Typically dopant density is measured in atoms per cc. Potatoswatter (talk) 23:46, 6 February 2009 (UTC)
- You'd be better off not using phrases like "confusing yourself" per WP:CIVIL. You are particularly bad at EE if you think there's ever a physically meaningful Coulomb-valued function of position, for what it's worth. Potatoswatter (talk) 23:49, 6 February 2009 (UTC)
- While I agree that charge density is used more often there are cases where it is more useful to talk about the difference of the total charge (not charge density) between two spaced apart regions of a semiconductor. One example is when determining capacitive effects. This may also have relevance to some aspects of memristive metal oxide systems since the total volume is fixed and it is the distribution of oxygen vacancies (expressed in terms of charge) between different discrete regions that determines the resistance state. To avoid confusion in the future when you state that something is being varied you should specify if it is time or position. Blm19732008 (talk) 19:59, 8 February 2009 (UTC)
Memristors have nothing to do with neurons
In only the narrowest and antiquated sense do memristors have anything to do with neurons. Having spent several years developing artificial neural network technology at Intel in the late 1980s and early 1990s, motivated by the claims that EPROMs were an electrical analog of synapses and therefore a mesh of them was neural, I learned a few things. One of them is that over the decade of enthusiasm beginning (in 1982) with Hopfield and Tank's seminal work, everyone in the (large) field learned that an artificial neural network is a mathematical object relevant to statistics, but not to neuroscience. Why then are neurons so prominent in this article and articles by others? No doubt it is because Stan Williams' and HP's documentation on memristors includes the topic, and maybe the memristor's theoretical inventor, Leon Chua, mentions them, too. In this sense, memristors' neuralness is a meme, but memes can mislead. I think these references should be removed, but I am new to the community and would like some input from other users.
How do I sign this? Dan —Preceding unsigned comment added by Mlvlvr (talk • contribs) 23:57, 15 January 2009 (UTC)
- Sign like this: ~~~~ . It sounds like you can add your expertise to the article, by drawing a distinction between a neuron (the complex physical object) and a neural network (the "learning graph"). Removing info is often considered inflammatory, but here "modeling the fire of neurons" looks certainly out of line, now that you point it out, and there is still another reference to "artificial neural networks. So you can probably delete the offending sentence and leave the properly qualified one. If you like add a disclaimer to that one, too. Potatoswatter (talk) 13:15, 16 January 2009 (UTC)
- In the paper "Memristive Devices and Systems" by Chua and Kang (1976) the model of memristive systems was shown to be applicable to the Hodgkin-Huxley model describing action potentials in neurons. Greg Snider at HP Labs is also working on designing a neuromorphic architecture based on memristors which was described at the 2008 Memristor and Memristive Systems Symposium at UC Berkeley. Blm19732008 (talk) 18:51, 16 January 2009 (UTC)
- Neurons are REALLY COMPLICATED. Being applicable to one very simple model does not equate to generally modeling. Potatoswatter (talk) 19:12, 16 January 2009 (UTC)
- I agree that an analysis of neurons can be "really complicated" as you state. Nevertheless the Hodgkin-Huxley model is a fundamental model related to neurons and the paper by Chua and Kang make the connection between memristors and this model. Thus the statement "Memristors have nothing to do with neurons" contradicts the evidence. Blm19732008 (talk) 19:20, 18 January 2009 (UTC)
- I didn't say that they were unrelated. Mlvlvr doesn't either, if you read his post. The assertion of "modeling" is misleading because simple models of large neural networks have been around for a long, long time. However the behavior of a memristor can be creatively interpreted, it won't bring AI to the next step because the current challenge is not related to the computing equipment. Proponents of exotic physical neuron analogues, particularly semiconductors, are mostly disingenuous. Therefore an alternative viewpoint is needed to stay NPOV, or the promotion should be toned down. Potatoswatter (talk) 21:52, 20 January 2009 (UTC)
- You seem to be providing a lot of opinion but not much fact. If "Mlvlvr" agrees that memristors are related to neurons than the choice of title for this section is very poor. Do you have any basis for your opinion that the next step in AI is not related to computing equipment? Pattern recognition seems like an important ingredient to AI and Greg Snider of HP Labs published an article in 2007 entitled "Self-organized computation with unreliable, memristive nanodevices" noting the potential for memristor circuit architectures for enhanced pattern recognition. Another article entitled "Memristive model of amoeba's learning" by Yu V. Pershin, S. La Fontaine, and M. Di Ventra identified memristive behavior in amoeba's learning and, as already pointed out, Chua and Kang's 1976 article note the connection between memristive systems and the Hodgkin-Huxley neuron model. At the 2008 Memristor and Memristive Systems symposium the connection between neurons and memristors was also discussed by Greg Snider. Thus there is verifiable evidence in the literature that there is a connection between neurons and memristors and, while it may be too early to tell how important this connection will turn out to be, the connection itself is what "Mlvlvr" originally questioned.Blm19732008 (talk) 22:36, 21 January 2009 (UTC)
- It is a poor title, and you really should read his post before writing a longer post yourself. Pattern recognition is nice but, as we're pointing out, is understood in statistical terms not biological terms. We don't know how neurons perform pattern recognition. Memristors implement neural nets, they do not model neurons. The papers are nice, but keep in mind that research tends to occur at the crossroads of buzzwords. Research on neurons and artificial neural nets have advanced separately to the point where there is a definite distinction. Also, note that amoebas do not have neurons!! Potatoswatter (talk) 01:06, 22 January 2009 (UTC)
- I did initially read the entire post by Mlvlvr and it is reflective of the title. The original posting expressed ignorance of the connection between memristors and the Hodgkin-Huxley neuron model which you still do not seem to comprehend based on your statement that memristors do not model neurons. Also unless you can back up your opinions with verifiable evidence please refrain from making blanket statements about neuroscience or artificial intelligence and do not refer to your opinions in the plural ("We"). You are not the representative for all the neuroscientists and computer scientists of the world. Amoebas do not have neurons per se but they have been found to exhibit spatiotemporal oscillatory behavior which has been studied for neurocomputing (do a search for "amoeba-based neurocomputing" in GoogleScholar.Blm19732008 (talk) 15:45, 23 January 2009 (UTC)
- "We," Mlvlr and I, are both saying that H-H was state of the art long long ago but neuroscience has moved on. I'm not going to do research for you. The wiki attitude that a few Google searches makes one an expert is unfortunate, but I don't expect to sway you. Mlvlr claims to have been a researcher in the field for years, and I studied it for about a year in high school, mostly reading edited anthologies of influential papers. I think you still should re-read his post again. Let me repeat: simple artificial neural networks, while useful, are not the best analogues of neurons. The word "neuron" should not be used alone to describe an element of an artificial neural network. Potatoswatter (talk) 19:06, 23 January 2009 (UTC)
- From mlvlr's posting it seems he (or she) was not aware of the Hodgkin-Huxley model or at least the connection of the model with memristors so the grouping of "we" does not seem appropriate. For someone who admittedly has only a year of high school study on the subject I would display a little more humility in your postings rather than try to come off as an expert. From a study of the literature there is actually a stronger connection between memristors and neurons than between memristors and neural networks. Neural networks are more closely connected to the crossbar architecture which Greg Snider of HP Labs was looking at to implement molecular neural networks a few years back based on rotaxane switches. Greg Snider's research has moved on to the examination of memristors for neuromorphic architectures which go beyond the limitations of neural network architectures. If you are seriously interested in memristors I would consider educating yourself a little more in the literature. UC Berkeley has a series of videos covering the Memristor Symposium on YouTube and I would strongly recommend watching them, particularly the video covering Snider's work on memristive neuromorphic circuit architectures.Blm19732008 (talk) 14:59, 25 January 2009 (UTC)
- No, he definitely is aware. Reading papers for a year earns expertise at any age, and if you had any methodical study yourself you'd know which papers are seminal. But you didn't study and you don't know. The point of describing my study that way is to not pose as an expert… you're really pretty rude… Have fun with YouTube. Potatoswatter (talk) 19:22, 25 January 2009 (UTC)
- Please avoid personal attacks since this is against Wikipedia policy. I am sorry if I appear rude to you but I am simply attempting to point out the facts related to the relationship between memristors and neurons which is the topic of this section. It seems bizarre to me that you are accusing me of lacking any methodical study when I am the only one who is backing up my position with verifiable evidence and you have only provided your unverified opinions (Incidently I am a graduate student and have actually published papers on this subject). Below is the abstract from Greg Snider of HP Labs describing his neuromorphic architecture in which memristive nanodevices serve as neural synapses available at
- "Neuromorphic hardware has long been hindered by the difficulty of implementing “synapses” efficiently—they simply require too much area. Dynamical nanodevices, each no larger than 30 nm X 30 nm, can supply a multiplicative transfer function and implement correlational learning laws. Using conventional CMOS for neurons, nanowires for axons and dendrites, and memristive nanodevices for synapses, I'll present architecture for implementing fundamental cortical circuits."
- Reading the latest research is not methodical study. "Neuromorphic hardware" cannot be understood if you have not read the first few chapters of a biology or psychology textbook on neurons. It looks to me like you're jumping into popular science without background. If you had background, you would not have assumed mlvlr was ignorant. Clear now? Potatoswatter (talk) 19:17, 27 January 2009 (UTC)
- Well if you think your admittedly high-school level knowledge is better than someone who has studied this on the graduate level and has published on the topic more power to you. There is a distinction between neural networks and neuromorphic systems. Memristors have not been applied to the former but have been applied to the latter. Both Mlvlr and your ignorance was established by your failure to note the connections between memristors and the Hodgkin-Huxley model or Snider's work.Blm19732008 (talk) 14:52, 28 January 2009 (UTC)
- Lol. So I suppose if you studied at the graduate level, you wrote a paper then? Potatoswatter (talk) 19:22, 28 January 2009 (UTC)
- Yes, Blm19732008 did. If word choice is to be trusted, then he wrote more than one. He stated it once implicitly in the paragraph to which you just responded ("Well if you think your admittedly high-school level knowledge is better than someone who has studied this on the graduate level and has published on the topic [...]" — emphasis added), and also previously ("Incidently I am a graduate student and have actually published papers on this subject"). Please do try to read what the other person is writing, and avoid bickering. The goal here is to try and improve the article. That said, I'm dubious of hyperbolic statements, e.g. that reading papers on a topic for a year earns one "expertise" — this seems to negate the role that research (especially the hands-on kind) plays in forming understanding of a complex subject that is nowhere near fully understood. I for one am very interested in reading about the latest research into the relationship between memristors, neurons, and artificial neural networks, whatever that might be. Sadly, this article has been tagged for updating and improvement, but the last debate on this discussion topic thread seems to have ended back in January 2009. I have no horse in this race, but if one of the other contributors to this thread who's done work in the field would care to contribute some sourced information to the article, I think that would be a great start. (I noticed one update from a 2011 IEEE Computer article regarding neural network applications, so it seems there's been some activity, but I'm really interested to see information included about memristors as compared to biological neurons.) 151.151.16.9 (talk) 02:16, 23 February 2011 (UTC)
- To go back to Mlvlvr's original assertion, "In only the narrowest and antiquated sense do memristors have anything to do with neurons", this is perfectly correct in that there's no titanium dioxide in wetware neurons or synapses. I could argue that it's not that "memristors are neurons" but that "wetware synapses behave like memristors" only I feel that's too philosophical for a cold February afternoon.
- But to answer Mlvlvr's question, "Why then are neurons so prominent in this article and articles by others?", it is because the lack of nano-scale nonvolatile analogue memory has been just about the single biggest impediment to building practical large-scale distributed representation (DR) systems in hardware. The issue is the ratio of neuronal units (neurons in wetware) to inter-unit connections (synapses) and their relative physical size. In wetware (which we must all admit does DR rather well), the synapes-to-neuron ratio is typically ≈ 1000:1 or more. Wetware synapses are tiny compared to wetware neurons so it all fits together well. However, in pre-memristor analogue CMOS hardware, the silicon area of each "synapse" has typically been as large - or often much larger - as the "neuron" making high density networks impractical.
- Hence the excitement in the neuromorphic electronics community (of which I am a humble member) that we might finally see a device in the TiOx memristor that can store continuous-valued weights on tiny areas of silicon. Sure, memristors can also work wonders for digital (binary) storage density but it's the analogue storage that make me
droolthink they're cool. (A discussion on all sorts of problems building neural networks in digital hardware doesn't belong here - try the Von Neumann bottleneck for a start.) P.r.newman (talk) 15:27, 17 February 2012 (UTC)
Flux Capacitor
At time point 46:40 of this video http://www.youtube.com/watch?v=QFdDPzcZwbs Dr Chu describes a "Memory Capacitor" that is a capacitor that is a function of flux. So it can be accurately called a "flux capacitor" or more accurately a "flux-based capacitor". I saw on many web sites that people were trying to call the memristor a flux capacitor which is not at all correct. But I wanted to point out that he did invent a "flux capacitor", as the 5th possible element in basic circuits. —Preceding unsigned comment added by 24.214.120.227 (talk) 04:56, 23 January 2009 (UTC)
- Materials which could exhibit a memory capacitor effect to enable such a "flux capacitor" or "memacitor" (the term I prefer) are discussed in "Ultralarge capacitance-voltage hysteresis and charge retention characteristics in metal oxide semiconductor structure containing nanocrystals deposited by ion-beam-assisted electron beam deposition" (Applied Physics Letters, Vol. 78, No 7). I exchanged some E-mails with Prof. Chua about this topic after the symposium and the possibility exists for memristive system analogies to both capacitors (memacitors) and inductors (memductors) so it may possibly be said that there are six rather than four fundamental circuit elements. Alternatively, it could be more appropriate to say that there are still three basic circuit elements and that the memristor, memacitor, and memductor are more generalized versions of the three basic elements which exhibit the pinched hysteresis Lissoujous curves characteristic of memristive systems. Blm19732008 (talk) 16:08, 23 January 2009 (UTC)
this page is way too technical for laypersons
seriously, i have no idea what this is all supposed to mean and all i wanted to know was what a memresistor was :( 99.245.16.164 (talk) 18:23, 19 April 2009 (UTC)
- If you don't know what a resistor, inductor, and capacitor are, and you don't have a specific question, you might just be out of luck. In particular, we can't point to a physical thing and say "this is what a memristor looks like." A memristor is any device with memristance, and memristance is a mathematical concept. As the article says, it is similar to variable resistance (yet very different). Sorry. Potatoswatter (talk) 02:00, 20 April 2009 (UTC)
- It seems to me that the memristor is one kind of resistance change memory (RRAM). The memristor was first introduced by Dr. Chua in UC Berkley in 1971. The definition of the memristor (flux linkage and electrical charge) was given at that time. The memristance has its own meaning, it means that the equivalent resistance of the memristor is equal to the charge that has passed through the device. I think the most confusing part is distinguish the memristor with RRAM. RRAM shows nonvolatile attribute which the memristor also has. However, the intrinsic resistive switching mechanisms are different. There are lots of technical papers about the switching mechanisms, most conceivable proposals include: ionic transportation, thermal effect, electrochemical effect. The memristor fabricated by Strokov et al. was mainly based on ionic transportation effect. —Preceding unsigned comment added by 130.65.11.70 (talk) 01:35, 20 February 2011 (UTC)
- I must say, the first sentence of the "Theory" section (even though it still cannot be understood by laypersons) is significantly easier to understand than the introduction of the entire article! Surely the introductions should be far simpler. Here is the first sentence of the "theory" section:
- "The memristor is formally defined[4] as a two-terminal element in which the magnetic flux Φm between the terminals is a function of the amount of electric charge q that has passed through the device."
- Or are the two definitions different in some way? Regards, telewatho (talk) 17:50, 11 April 2010 (UTC)
Inaccessible Article
This is an appallingly inaccessible article. The Memristor is a simple device, and needs a simple English explanation. "If you can't explain it to a six-year old, you don't understand it yourself". Albert Einstein.
I would suggest some simple mechanical analogies: A geared motor driven by the current, which drives a potentiometer? A negative coefficient thermistor which is heated/cooled by a Peltier module? A simple electronic circuit which accurately simulates a Memrister? Some simple electronic oscillator circuits using a Memrister? Gutta Percha (talk)
- I second you completely. Really, the memristor is extremely simple and intuitive device but its theory is too complex and abstract. This is a good example of how a theory makes the simple things complex... Thank you for the great Einstein's thought; let's try to put it into practice. A geared motor driving a rheostat is a very good electromechanical example. If you apply input voltage across or current through the motor and take the rheostat resistance as an output, you get a 4-terminal integrating memistor. It is even better than the 3-terminal Widrow's memistor:) since the output is completely isolated from the input. If you connect the motor and the rheostat in series, you will get a 2-terminal integrating memristor. Regards, Circuit dreamer (talk, contribs, email) 11:29, 8 March 2011 (UTC)
I think a better analogy would be as R. Stanley Williams put it, consider a memristor as a pipe whose diameter changes as water flows through it. Getting narrower and therefore harder to traverse as the water flows in one direction, and easier and easier as it flows in the other. --TheRandom — Preceding unsigned comment added by 144.173.5.197 (talk) 14:42, 12 December 2011 (UTC)
Formal Definition of Memristor
I think there is a slight misunderstanding of the memristor definition.
In the article the memristor is formally defined as: "The memristor is formally defined[4] as a two-terminal element in which the magnetic flux Φm between the terminals is a function of the amount of electric charge q that has passed through the device"
However the link between electric charge and magnetic flux is only a more specific case of the definition given by Chua. In fact his definition was more general. More specifically, his definition first defines how the memristor's voltage depends on current and a "state variable". The "state variable" in this case is a quantity that measures a physical property of a device. The second part expresses how the changing state variable depends on the charge flowing through the device.
This means that a magnetic interaction is not necessary for memristance and linking electric charge and magnetic flux is only one way to satsify the definition. —Preceding unsigned comment added by Mvd1221 (talk • contribs) 17:02, 23 April 2009 (UTC)
- Read the section " Magnetic flux in a passive device" to see why "magnetic flux linkage" is an integral of voltage, unrelated to any "magnetic interaction." Potatoswatter (talk) 19:29, 23 April 2009 (UTC)
- Also, what is the difference between this definition and the definition here: [1] ?
- Can a simple (non-technical) definition be added to the start of the article? Regards, telewatho (talk) 17:54, 11 April 2010 (UTC)
- The definitions agree perfectly. Can you give an example of what you would like to see? Potatoswatter (talk) 08:03, 13 April 2010 (UTC)
- It is my understanding that a memristor is an electronically variable analogue device which retains its value when no power is applied. The definition(s) of the electric charge and magnetic flux devices are memistors, but I believe that they are one type of memistor, not the definition itself.
- In the Widrow / Hoff patent, #3,222,654, devices in figures 1 through 15, the value of resistance is read to determine what value was stored. Figure 19 is a memory element in which the value of capacitance is electrically varied and stored. Figure 20 uses an alternating magnetic field from one coil to another mounted in a parallel plane to measure the stored value. Figure 21 discloses a device that stores the variable as a function of mutual coupling between two orthogonal coils.
- Perhaps these devices would not come under the definition of a "memory resistor", yet they, and the titanium device are all devices which retain an analogue value representing an integral of the signal applied to it over its entire lifetime.
- The shortest way to describe the difference between the titanium device and 'all others' is that all the others (I think) use an electrolyte to aid in the change to the device. As I have interpreted the HP titanium device, there is no electrolyte involved.
- Unless there is a term to describe memory resistors separately from memory capacitors, separately from memory inductors, then the definition of memistor at the beginning of this wikipedia page on the subject should start out with a more general definition of a device that stores an analogue value (excluding potentiometers). Then proceed to describe the various analogue memory devices of differing technologies.
- If the term memistor is to have a broad definition, then I would expect it to include mercury coulombmeters. The only mercury coulombmeters I have seen were read optically, I am unaware of any being read by any other method. These devices remember the total NET couloumbs which have passed through them.
- This question might have an answer in whatever document was the first to use the term memistor or memristor. Does anyone know when & where one of those terms was used?
AlanDewey (talk) 22:28, 26 May 2010 (UTC)
- And now for some more potential confusion:
- See http://www.stormingmedia.us/48/4883/0488356.html Horace Tharp Mann working for the National Security Agency and TRW makes reference to "The Persistor" in a 1966 report on associative memory. I have not been able to find any more about the Persistor, but the name sure sounds relevant to memistor. I found this as a result of investigating H.T. Mann in reference to Frank Rosenblatt's Perceptron AlanDewey (talk) 17:14, 27 May 2010 (UTC)
Nice research (especially the NSA "persistor"). Widrow's memistor (1960) was a 3-terminal electrochemical device and he did not develop a mathematical model for it. Chua's memristor (1971) was based on a mathematical model for a fourth non-linear 2-terminal passive circuit element but did not represent a real device at the time. Chua and Kang extended the memristor concept to cover a larger class of memory circuits in 1976 and this is what HP's memristor basically represents. Several example of materials having memristive characteristics have been known for over 50 years in the form of both electrochemical and solid state thin film materials. I have some online articles which provides some additional info. Google "Widrow memistor" or go to the following link: http://knol.google.com/k/memistors-memristors-and-the-rise-of-strong-artificial-intelligence# —Preceding unsigned comment added by 98.218.144.3 (talk) 04:29, 5 June 2010 (UTC)
- I read the article which was referred to by 98.218.144.3. I found it to be extremely helpful on this subject. AlanDewey (talk) 14:33, 7 June 2010 (UTC)
I sent an inquiry by email to Dr. Widrow, requesting a clarification on memistor versus memRistor. This is his reply. The first paragraph is about the memistor, the second is about the memRistor.
The memistor is a three electrode device. It is really a liquid state device. It uses electroplating to change resistance between two of its electrodes. Electroplating is caused by current flowing between the third electrode and the other two. The amount of electroplated metal depends on the integral of the current. The equivalent circuit of the memistor is that of a transistor with a built-in integrator.
The memristor, as I understand it, is a two element solid state device whose resistance is changed by current through the device. HP is planning to use it for digital non-volitile memory, and possibly to use it for neural networks. The memistor was developed for neural networks and was used to implement them. Other applications were for non-volitile analog memory. Commercial applications existed but did not develop into big markets. AlanDewey (talk) 20:44, 9 June 2010 (UTC)
Needs work
Coming to this page for the first time to find out what a memristor is, I feel I really couldn't get the whole picture without coming to the discussion page, and thence to the history pages. There are some complaints people already voiced which I think are valid:
- There needs to be a simpler, clearer description in words of what a memristor is, does and what it is used for. I don't see anything here that is logically so complex it can't be rendered in plain english. These sentences in the opening paragraph are most awkward, "There is no such thing as a generic memristor. Instead, each device implements a particular function. A linear time-invariant memristor is simply a conventional resistor." What is anyone meant to make of these unexplained facts if they just came here?
- The article should more explicity explain why it is called a memristor (this is only addressed in passing, and a most unsatisfactory reason is given).
- It became apparent reading the history that there are some problems/controversies with the concept and the application of memristors that I would certainly like to have known about without trawling through article changes.
In general, this is certainly not written for the layman, nor someone new to the concept. I hope someone with knowledge on the subject could take a stab at improving the clarity and the flow of information within the article. David 218.143.30.1 (talk) 02:32, 9 July 2009 (UTC)
- Part of the problem is that a particular kind of memristor has gotten a lot of popular coverage, so you expect this page to be simpler or entirely different. Typically when a laboratory invents a new device, they name it something unique, ie "micro-memristor" or "memristive memory," with an associated acronym. However, HP labs instead chose to claim they had "found the missing memristor" when in reality Chua had from the start illustrated the idea with everyday examples.
- Perhaps there should be a separate page for HP's celebrated "invention." (And, as the timeline shows, a few predecessors.) But there isn't an agreed name besides "HP's memristor." I propose "solid electrolyte memristor," maybe it'll stick.
- On the face of it, you can't begin to understand what a memristor is without knowing about current, voltage, and integrals (maybe throw electrons into the mix), because these are the symbols used to define the word. And a lay audience by definition knows about none of those things. Potatoswatter (talk) 14:25, 9 July 2009 (UTC)
Stan Williams (of HPLabs) has a short Youtube video (available in the External Links section) which does a very good job for a layperson introduction. I also wrote a brief introduction to memristors in a knol entitled "Programmable Electronics using Memristor Electronics" which is also linked in the External Links section and which is intended be accessible to the layperson. Blm19732008 (talk) 02:24, 10 July 2009 (UTC)
- The only YouTube external link is [2], which is not short at all. The knol link (Programmable Electronics using Memristor Crossbars) doesn't attempt to define the memristor at all aside from as a device with a Lissajous I-V graph, and does not address a lay audience as it assumes knowledge of signal processing. Note that hysteresis is not a property of a pure memristor, but defines when the device leaves the memristive regime. Potatoswatter (talk) 17:30, 10 July 2009 (UTC) Woops, I confused hysteresis and saturation. Potatoswatter (talk) 19:07, 10 July 2009 (UTC)
The Youtube video I was thinking of is entitled "6-minute memristor guide" which I thought was on the links page but is easy to find in any case. I was referring to the first section of the knol which uses the zero-crossing I-V curve which defines memristors (according to the 1976 paper of Chua and Kang) and which may be understood without a knowledge of calculus or differential equations. Incidently the equations used in this wikipedia article are not really the best. See my knol "An Introduction to Memimpedance and Memadmittance Systems Analysis" for a more complete analysis which integrates the memcapacitive with the memristor concept to form a more general mathematical foundation to explain the memory effects found in thin films due to oxygen vacancy drift and filament formation. —Preceding unsigned comment added by Blm19732008 (talk • contribs) 01:45, 13 July 2009 (UTC)
- I don't think an IV curve or a Lissajous figure are more accessible to the average person than an integral. More people take calculus in high school than ever see an IV plot or hear a definition of hysteresis. Even undergrads have trouble interpreting that, particularly when the concepts are put together with arrows, although it is more "visual." The assumption of sinusoidal input is particularly obscure.
- The equations in this article were cherry picked from Chua's original paper. Your knol would not be a suitable reference. Note that there is still no article at mempacitor. Potatoswatter (talk) 03:55, 13 July 2009 (UTC)
This is taken from the Wired article in External Links:
"Indeed, Chua’s original idea was that the resistance of a memristor would depend upon how much charge has gone through the device. In other words, you can flow the charge in one direction and the resistance will increase. If you push the charge in the opposite direction it will decrease. Put simply, the resistance of the devices at any point in time is a function of history of the device –- or how much charge went through it either forwards or backwards."
Wouldn't something to this effect be a much better introduction? I don't have the knowledge to say if this is accurate, but something like this, or drawing analogies to analogue computing, or a comparison with transistors would go a long way to making this article more readable. David 218.143.30.1 (talk) 01:59, 13 July 2009 (UTC)
- It is a good, lay, description of a memristor. One reason the lead isn't more like that right now is that the article was bombarded by cranks in its early development, who insisted the situation was more complicated than that and that they had some other math. Please feel free to add that, if you can make it not copyvio. Of course, it's not a substitute for math. Potatoswatter (talk) 03:55, 13 July 2009 (UTC)
- See Wikipedia:LEAD for advice and examples on writing the lead section. --Bcjordan (talk) 08:19, 13 July 2009 (UTC)
Ideal Memristor
The article seems to go to lengths to expound that there is no such thing as an "ideal memristor" (basically because there seems to be no consensus what the ideal, most logical, M(q(t)) would be barring constant (in which case it's a resistor).)
If that's the actual state of things, fine, but if you actually want to make the concept of a memristor accessible, an example is merited, and that means choosing an example M() so that a nice illustration or animation can be produced, even if no claims are made that it represents an "ideal" device. So the question to those who've been digging into the literature is: what would that expedient M() be?
The insistence on using "magnetic flux" and accompanying many of those references with explanation that no, it's more a general idea than "magnetic flux" is distracting. Would it be possible to get away with just "flux" -- is that term used alone in literature enough to justify doing that? In that case, the clarification could be restricted just to the discussion of comparisons with an inductor.
Another tact towards showing the relationship in a more layman-friendly format might be to try it out in the frequency/impedence domain and see if it is prettier.
Finally, while it is certainly the case that memristors of interest operate with M(q(t)) never negative, the case for maintaining this restriction when discussing a hypothetical "ideal" device is not laid out convincingly, if it even exists. Just because a device would inject power into a circuit does not disqualify it from conceptual meaningfulness. If some of the literature describes hypothetical power generating memristors (or even attempts to simulate them with power sources) then that restriction can be pushed into the but-in-the-real-world section, further cleaning up the theory area. If there's a solid theoretical basis for why memristive theory cannot tolerate power generation, it needs to be said.
(140.232.0.70 (talk) 18:59, 16 December 2010 (UTC))
HP's 2008 memristor paper in Nature used M(q)=ROFF(1-mvRON/D^2 q) where ROFF, RON are the off and on resistances, mv is the ionic mobility of oxygen vacancies, D is the film thickness and q is charge.
Regarding the more general idea beyond magnetic flux this is usually associated with Chua and Kang's 1976 paper on memristive systems which used a dynamic systems definition rather then the "fourth fundamental circuit element" memristor definition. In reality most real materials exhibiting memristive effects obey the 1976 dynamic system definition but have nothing to do with the 1971 magnetic flux definition. My opinion is that HP has used the 1971 definition more for marketing purposes then for any real technical merit.
Regarding negative memristance, one of the proven mathematical properties of memristive systems is that they are passive and do not produce energy. Thus if there are devices having some properties similar to a memristor but which produce (or amplify) energy it would not be formally describable as a memristor or a memristive system. This is one reason why it is unlikely for memristors to replace transistors since they are incapable of performing signal amplification on their own.Blm19732008 (talk) 02:57, 18 December 2010 (UTC)
Unclear
I'll add my voice to the people saying that this article is very unclear. The original start of the article ("theory") defined a memristor in terms of magnetic flux. Memristors aren't magnetic devices! There's no reason to define them in terms of magnetic flux; I can't see how this makes anything clear.
I don't think that the magnetic flux section is well written, actually; magnetic flux is undefined for a two-terminal device (terminals are one dimensional). I don't even understand the statement "The magnetic flux]Φm between the terminals is a function of the amount of electric charge q that has passed through the device." As a general thing, the magnetic flux on a memristor will be zero if no charge is flowing through it, independent of how much charge has passed through it. The article also mentions the charge on the device; it's a little unclear here, since memristors are uncharged. This needs to be clarified "the integral of current that has flowed through the device," or "the total charge which has flowed through the device." I would personally delete this "theory" section completely, and rewrite it without ever mentioning magnetic flux. However, since apparently somebody thought it was appropriate to start here, so even though I don't know why they though it was appropriate, I am hesitant to delete their work.
There also seems to be some confusion between the simple memristor (i.e., the memristance equivalent of a resistor) and more complicated nonlinear memristors (i.e., the equivalent of nonlinear resistors) in which the memristance is a more complicated function of the history of the device. Chua moves very quickly to generalize the idea to nonlinear cases, but it is, I think, useful to start by just defining the simple memristor. (Chua apparently considers a diode to be just a nonlinear form of a resistor. ) So I tried to make this more explicit.
In any case, I have now rewritten an introduction section in order to (I hope!) make it more clear. Geoffrey.landis (talk) 15:40, 3 September 2010 (UTC)
Capacity
100 gigabits (12.5 GB) per cm² maybe seemed amazing some years ago, but today I can purchase a 32 GB smart card using conventional transistor technology (I am assuming that is the case anyway) The chip inside is probably smaller than 1cm² as well.
Is this capacity theory understated, or has standard transistor technology surpassed it? What is the practical application of this? Am I comparing apples to oranges?
Mrrealtime (talk) 16:56, 22 December 2009 (UTC)
- Read this news article. Memristers can apparently go smaller than transistors. And could actually replace them, allowing for memory and processing to be done simultaneously.
- Everyone else read the article because it's apparently new information. --trlkly 11:27, 9 April 2010 (UTC)
Resistors really are general memristors
Since real resistors change with state variables such as length, area, and temperature, resistors provide a readily-available example of a more general class of memristors than the article suggests. The example can be made very general: temperature depends on heat dissipation and even heat capacity of resistors, so that change in charge results in non-trivial change in flux (i.e. area under the voltage*time curve). Ywaz (talk) 10:56, 26 January 2010 (UTC)
- Memristance isn't variation with any variable, it's variation with Q(t), where reversing the direction of current reverses the change in Q. Potatoswatter (talk) 01:18, 27 January 2010 (UTC)
Resistors and memristors are both special cases of the more general state variable equations defining memristive systems described in a 1976 paper by Chua and Kang "Memristive Devices and Systems." These equations include a first equation (1) which is a generalization of Ohm's law which relates voltage (v) and current (i) except that the resistance R is dependent on a state variable w. The second equation (2) defines the rate of change of the state variable with respect to time (dw/dt).
(1) v = R(w) i
(2) dw/dt = f(w,i)
In the case of a resistor f(w,i)=0 and the memristance function R(w) is a constant.
In the case of a memristor f(w,i) = i and the memristance function is dependent on charge (the time integral of current).
The case described by Ywaz in which the state variable w is based on temperature was also treated in the Chua/Kang paper and was used in the analysis of a thermistor.Blm19732008 (talk) 01:02, 4 February 2010 (UTC)
add more early history of the memistor ?
I would like to consider adding this information on the history of development of the memistor. I am reading Dr. Buck's notebooks for this. I am an electronic engineer, not a chemist; for that reason I would like a chemist to review this for accuracy before we consider adding this.
1958 Interest in Frank Rosenblatt's Perceptron led Dudley Allen Buck to experiment, unsuccessfully, with tin dendrites as a means of using electrical current to form electrical connections. This work was quickly followed by an artificial synapse made from cuprous sulfide, Cu2S, as an electrolyte, with copper iodide, CuI, as a cathode and graphite as the anode in a quartz envelope.
I understand the controversy over whether a memistor is a neuron, but Dudley Buck clearly labeled the device Synapse 1 in his notebook. From his notebook, 1 July 1958: "Cuprous Sulfide appears to be a good candidate for such a self-oranizing system component. By plating copper out of Cu2S, one can change the composition from Cu1.996S to Cu1.93S ..... CuBr or CuI is a good cathode for accepting the copper ions. Pt or graphite forms a suitable anode. Electronic conduction is primarily by holes."
I am very new here on wikipedia. I am lost trying to add the citation; Dudley A. Buck 1 July 1958, M.I.T. Computation Book - 3/28/58 - death, page 7 (If deserving of a separate citation, the tin-dendrite work was June 25 through 29, pages 3 to 5.)
AlanDewey (talk) 17:29, 26 May 2010 (UTC)
- I will retract my suggestion to add this information on the early history. It has been pointed out to me that my suggestion is considered "original research" because it is not cited anywhere. (I learned something new! ). We may revisit this someday in the future. AlanDewey (talk) 20:35, 9 June 2010 (UTC)
ReRam - Update
Article needs updated, ReRam coming... two reuters articles [1] [2] and tons of other coverage; HP teamed up with Hynix to manufacture R3ap3R.inc (talk)
Flux
I did one more pass through the article, separating the part of the "theory" section explaining memristance in terms of magnetic flux into a separate subsection, "Flux Forumulation of Memristance", which now follows (instead of preceding) the discussion of the I/V relationship. I'm still not sure that this material is necessary to the article: the first part of the Theory section now says little more than various ways of restating Ohm's law for the case when resistance is a function of the charge that has passed through the device, and the second part of the Theory section is, at best, unclear, and tends to be misleading (since memristors are not magnetic devices). I didn't delete it, however, since I'm still hoping that somebody who understands the value of this particular formalism will come by and clarify the discussion. Geoffrey.landis (talk) 20:35, 4 September 2010 (UTC)
- Perhaps you should read Chua's 1971 paper on the memristor before making all of these changes. In that paper the memristor is defined as a non-linear functional mapping between magnetic flux and charge based on a symmetry relationship with non-linear resistors, capacitors, and inductors. I can sympathize with you that the definition used by HP deviates from this and is not a magnetic device but I do not agree that it is a good idea to ignore Chua's definition since he was the originator of the idea. —Preceding unsigned comment added by 98.218.144.3 (talk) 18:25, 5 September 2010 (UTC)
- I have an online article at http://knol.google.com/k/memistors-memristors-and-the-rise-of-strong-artificial-intelligence# which also might shed some light on Chua's original argument of the memristor as a "4th fundamental circuit element" linking magnetic-flux linkage and charge. —Preceding unsigned comment added by 98.218.144.3 (talk) 22:13, 5 September 2010 (UTC)
- Thanks for your suggestion. I have once again switched the order of the theory section. I do understand that you want to start with the definItion you list as originating from Chua, but this is starting at a far too detailed level, not an introductory place to start beginners. Please note, however, that I did not delete any content. The material that you want to keep in the article still there, just rearranged.
- Nevertheless, a significant problem remains: the section discussing magnetic flux, as written, is incomprehensible. It needs to be clarified by somebody who understands the physics well enough to explain it clearly. Magnetic flux is defined across a two dimensional surface (typically a loop, if you're using Faraday's law of induction, which you are) and it is not now clear over what surface you are defining flux. I don't know what the phrase "flux between the two terminals" means. The definition doesn't seem to have any meaning for a resistor, the case M(q)=constant. I put the "clarification needed" tag back on; but this is only the tip of the problems with the explanation; it needs to be thought out with the objective "how can this be clearly explained to a beginner?" Geoffrey.landis (talk) 20:31, 6 September 2010 (UTC)
- Geoffrey, I just rewrote that section. Also, it appears the text had barely changed at all since I wrote it a couple years ago. You might as well address any issues to me. As I've already said on this page, someone who doesn't understand resistors, capacitors and inductors, and without knowledge of calculus (that is, a beginner) has no hope of understanding memristors, and would probably skip the theory section anyway. I don't think attempting to adjust this section to cater to the lowest common denominator will clarify or improve it for anyone.
- Looking at the first two sentences of the section as now written, it says "The memristance can be written as (equation). It can be inferred from this that memristance is simply charge-dependent resistance." ***No, it can't.*** It is now extremely confusing and off-putting right at the outset, because you moved the introduction and the section now begins at the middle. Potatoswatter (talk) 02:42, 7 September 2010 (UTC)
- OK, I'll address coments to you: the section you wrote needs a lot of work. It is not clear.Geoffrey.landis (talk) 03:43, 7 September 2010 (UTC)
- Can you be any more specific? Do you agree about my reasoning why your edit is even less clear? It's hard to do this all by myself. There's a lot of nonconstructive criticism here, and nobody else has edited it for two whole years. Potatoswatter (talk) 05:37, 7 September 2010 (UTC)
- To be more specific myself, I see now that you removed the lead sentence
Why do you feel this is inappropriate? Indeed the term "magnetic flux", which is really only appropriate to inductors, is defined later, but that is the definition. Is it really right to override and reformulate Chua's definition? Or would it be better to add a note to the effect of, "This term is defined below"? Potatoswatter (talk)The memristor is formally defined[3] as a two-terminal element in which the magnetic flux Φm between the terminals is a function of the amount of electric charge q that has passed through the device.
- To be more specific myself, I see now that you removed the lead sentence
Superb-- the rewrite you just did makes is a lot clearer! Yes: if you're using the term "magnetic flux" to mean something other than magnetic flux, you must indeed define it the first time that you use it (which you now do) and state that it "does not represent a magnetic field here" (which was never explicitly stated earlier). What the earlier draft had done was to link to the Wikipedia article magnetic flux, which had been completely misleading.
This draft is much clearer. I cleaned up the text a little; added a definition of Φm that follows more immediately after its first use, and some other minor changes, mostly trying to remove some of the implication that Φm represents the flux of physical magnetic fields, and instead make it clear that it's a generalization of the concept from inductors to general circuit elements. There's still a little inconsistency over whether capital Q or small q is used to represent charge-- the theory section starts with Q but then when it gets to memristor characteristics, it uses q. This may be ok, though, so I left it as is.
This is good work; thanks! Geoffrey.landis (talk) 14:18, 7 September 2010 (UTC)
Typo in Definition of Q
Third sentence in the Theory section: "where Q is defined by Q = dI/dt"
Actually, I = dQ/dt therefore Q = integral of I dt.
Negative resistance
In 1962, Hickmott observed a negative resistance in a thin anodic oxide film. The result shows that a current through this anodic film increases when the voltage across the film decreases. This negative resistance phenomenon may also relevant to the memristor —Preceding unsigned comment added by 130.65.11.70 (talk) 01:41, 20 February 2011 (UTC)
HP's memristor may be propaganda
I have been a researcher in memory resistance materials since 2006. When I first heard about HP's memristor I believed they had actually developed something new and I was enthusiastic about their device. I presented a short paper at a conference in Boston based on applications of the memristor in signal processing and I was invited to speak at the 1st Memristor and Memristive Systems Symposium at UC Berkeley in 2008. I was somewhat surprised that the model they were using actually diverged from the original charge vs. magnetic flux linkage definition of Chua and that Stan Williams group was using a more generic memristive systems model. At the time this did not bother me too much and I was simply glad to have been invited to participate.
I knew that there were several earlier examples of memory resitance materials and I started the timeline of this wikipedia page to provide some context with respect to HP's development. A few months after the Memristor Symposium I received an e-mail from one of the main figures associated with the memristor suggesting that any type of generic memristive system should now be referred to as a memristor. I had a little problem with this and noted to this person that I believed it would cause confusion with other forms of memory resistance materials being developed for Resistive RAM by other companies but I was not in a position to do anything.
Over time I began to notice an increase in scientific publications describing different types of materials as "memristors" without providing a proper dynamic systems model in accordance with the definition of memristive systems developed by Chua and Kang. This bothered me a lot because without such a model how were electrical engineers going to build systems based on these devices. Meanwhile papers were being published by circuit theorists based on the original charge vs. magnetic flux linkage concept of Chua even though this does not describe real memory resistive effects due to non-linear ionic mobility among other considerations. At this time it seemed to me that many of the material scientists who published on the "memristor" only had 2nd-hand knowledge of the theory while the circuit theorists did not understand the material considerations.
Last summer I was invited to speak at a memristor session in the 2010 IEEE International Symposium on Circuits and Systems (ISCAS) and I attempted to raise issues such as whether it was correct to say that the memristor was a missing 4th fundamental circuit element found by HP when similar work was done in the 1960's by Argall "Switching Phenomena in Titanium Oxide Thin Films" and HP's memristor does not actually follow the original charge vs magnetic flux definition of Chua. A few weeks after ISCAS a representative from IEEE e-mailed my advisor at George Mason University and told him that it was bad for my reputation to give this presentation.
As time goes on I am seeing a greater number of articles written by researchers who appear to me are just trying to ride the coattails of the memristor meme without any understanding of the dynamic framework required in order for a system to be considered memristive. The original definition of memristor is now so bastardized that some people seem to be applying it to any dynamic system with memory.
Meanwhile there are several companies developing competing Resistive RAM to HP's potential product. In many cases these companies have been developing these materials since before HP came on the scene. As this wikipedia page has evolved it seems to me that there is less reliance on the original definition of the memristor and the term memristor is gradually being turned into a generic PR term so that HP can take control of the Resistive RAM market without consideration of whether it is truly justified. This is to the benefit of HP and to the disadvantage of HP's competitors and should be avoided. Blm19732008 (talk) 06:34, 14 March 2011 (UTC)
- The purpose of this page is to discuss improvements to the article. To put this analysis into the article we are obliged to reference it to published sources. Are there published papers you can point us to which make a clear distinction between charge/magnetic flux definition of memristor and other memory devices? Are there any that clearly state that the HP device does not conform to Chua's definition? Are there any that conclude that HP is using the memristor name as a marketing gimic? Can Chua's definition be said to be the mainstream definition - from what you have said above this appears to be rapidly becoming not the case. SpinningSpark 18:10, 14 March 2011 (UTC)
The original 1971 article of Chua "The missing memristor" defines the memristor as a 4th fundamental circuit element based on a relationship between charge and magnetic flux linkage (which can be equated to the time integral of voltage). That article along with the 2008 memristor article from HP should be the primary references for this wikipedia entry and the definition in the introduction section of the article should not deviate from these original references or attempt to generalize the definition beyond this. There are already separate wikipedia entrys for RRAM, CBRAM, and phase change memory which are also 2-terminal passive resistive memory but which are not claimed to be memristors by the companies developing the technology. My main concern is that this page should maintain the charge vs. magnetic flux linkage definition in the introduction in order to avoid "memristor" becoming a generic term which would create an artificial advantage for HP relative to competitors working on RRAM, CBRAM, and phase change memory. It is possible that one day in the future "memristor" might become a generic term but this should be based on market acceptance not because of a wikipedia article. It is premature to assume that this will be the case since no RRAM is available on the market yet. Blm19732008 (talk) 04:07, 15 March 2011 (UTC)
- My friend, I really have no disagreement with your sentiment and I do understand the difference to Chua's original description. The problem here is that it is not Wikipedia's business to protect the purity of Chua's definition nor to campaign againt HP misusing the name in its marketing nor to prevent memristor becoming a generic term. Wikipedia articles are encyclopedia articles and merely summarise what is in the sources. So we need published sources discussing this issue before we can make the points you want to make. SpinningSpark 08:01, 15 March 2011 (UTC)
What memristors, memcapacitors and meminductors are
A classification with respect to energy processing
I would like to show with the table below that the memristor, memcapacitor and meminductor are nothing else than kinds of resistors, capacitors and inductors (time-dependent nonlinear passive elements). The rows of the table represent the three possible electrical properties and the columns represent the three variations of these properties. The obtained passive elements (generic or specific) are placed in the according table cells. Please, discuss. Circuit dreamer (talk, contribs, email) 22:40, 16 March 2011 (UTC)
The table you present is representative of models developed by Yuriy Pershin and Max Di Ventra who are my fellow researchers in mem-electronic theory. However, perhaps better categories for the third column may be mem-resistive systems, mem-capacitive systems, and mem-inductive systems to address more generic dynamic systems. For example, there are dynamic circuit elements such as the thermistor which was identified in Chua and Kang's 1976 paper as a type of mem-resistive system in which the memory effect is volatile. The transient thermistor model plays a role in phase change memory resistor models when coupled to crystallization rate equations. There are also other mem-resistive systems models under development for different types of memory resistors such as ion doped chalcogenides which diverge significantly from HP's TiO2 memristor model and should not be lumped in with HP's memristor. There are also "mem-transistor" dynamic circuit models which I developed for application to modeling of the Widrow-Hoff memistor and synaptic floating gate transistors which I presented at ICECS 2010. This presentation is available at http://www.slideshare.net/blaisemouttet/memtransistor-systems .
I think your idea for the table is good but it might be better for a new wikipedia article on "mem-electronics" covering the wide variety of "mem" models that currently exist and are likely to emerge in the next few years. One reference discussing a variety of mem-systems is available at http://arxiv.org/abs/1011.3053 Blm19732008 (talk) 05:27, 18 March 2011 (UTC)
- This table doesn't make any sense and is just more imaginative taxonomy. Capacitors and inductors are non-linear. A varactor is not a fundamental component the same way as resistors, inductors and capacitors are. The memristor has a theoretical basis but the other elements of the table are simply made up. One reference is not sufficient. The earlier version of this article was more balanced and should be restored. Zen-in (talk) 17:24, 19 March 2011 (UTC)
- Actually, Chua himself co-authored a paper (which is already cited in the article) which categorises memristors, memcapacitors and meminductors. By the way, the descripion of memresistors in this paper seems to be at odds with his description in the "fourth circuit element" paper, and more along the lines being proposed by Circuit dreamer. There would seem to be two distinct uses of the term memristor which we should make more clear than it currently is in the article. Perhaps there should even be two articles. SpinningSpark 17:44, 19 March 2011 (UTC)
- So one paper is all that is necessary to completely re-write a wikipedia article? I don't agree with Chua's assertion; although I do sympathize with his need to publish. He is considering a second order effect, hysteresis, to be a memory property. So what if some mem-resistors are found to have a second order hysteresis effect? This would mean there are mem-mem-resistors. Maybe there are mem-mem-inductors and mem-mem-capacitors as well. It is all very specious. The article should be restored to its earlier state, possibly with these speculative ideas added. Zen-in (talk) 02:37, 20 March 2011 (UTC)
- The significance of this is that it is Chua that first proposed the Q/Φ constitutive relation memristor, yet in this paper memristor becomes a V/I constitutive relation - ie, simply a variant resistor (one with hysterisis). It is also significant for our article that the HP device is of the V/I variety and not the "fourth element" of Chua's original formulation, HP's marketing hype notwithstanding. SpinningSpark 10:20, 20 March 2011 (UTC)
- So one paper is all that is necessary to completely re-write a wikipedia article? I don't agree with Chua's assertion; although I do sympathize with his need to publish. He is considering a second order effect, hysteresis, to be a memory property. So what if some mem-resistors are found to have a second order hysteresis effect? This would mean there are mem-mem-resistors. Maybe there are mem-mem-inductors and mem-mem-capacitors as well. It is all very specious. The article should be restored to its earlier state, possibly with these speculative ideas added. Zen-in (talk) 02:37, 20 March 2011 (UTC)
A classification with respect to time behavior
The classification above is in respect to energy processing. It divides the passive electrical elements into three groups - dissipating energy (resistors), storing electric (potential or pressure-like) energy (capacitors) and storing magnetic (kinetic or flow-like) energy (inductors). From this viewpoint, memristors, memcapacitors and meminductors are only varieties of the three basic passive elements; so, they are placed in an additional column of the table (memristor is a kind of a resistor as it dissipates energy; memcapacitor is a kind of a capacitor as it stores electric energy; meminductor is a kind of an inductor as it stores magnetic energy). In this way, I have refuted Chua's assertion that the memristor is fourth, the memcapacitor - fifth and the meminductor - sixth circuit element.
Now, I would like to give hopes to Prof. Chua:) that it is still possible to present the "mem" elements as separate items if only we classify passive elements with respect not only to energy but to their behavior through time as well. From this viewpoint, the memristor is neither true resistor (as its resistance depends on time) nor true capacitor or inductor (as it dissipates instead to store energy). It resembles resistors (with respect to energy processing) and capacitors and inductors (with respect to time behavior); it may be defined as a time-dependent resistor (resistive capacitor or resistive inductor). So, maybe it is better to separate the memristor as a particular element staying between resistors and capacitors (inductors) and to say that it is a mixed element combining the energy properties of resistors and time properties of reactive elements? Then passive elements will be four (resistor, memristor, capacitor and inductor) and even five (resistor, memristor acting as a capacitor, memristor acting as an inductor, capacitor and inductor)? Or, we have to place them into three groups in respect to energy and time: dissipating time-independent elements (resistors), dissipating time-dependent elements ("capacitive" and "inductive" memresistors) and storing elements (capacitors and inductors)? Please, discuss. Circuit dreamer (talk, contribs, email) 00:21, 20 March 2011 (UTC)
- This is completely back-to-front. It is not for Wikipedia to "refute Chua's assertion". The Wikipedia article should be reporting Chua's assertions. If there exist published papers disputing Chua's work, then they can be mentioned too, but no amount of argument on this talk page, however convincing, can result in a change to the article. If what you are proposing is based on sources, then please name them, otherwise please stop pushing for this, it will just cause a lot of grief for no gain. SpinningSpark 00:47, 20 March 2011 (UTC)
- Well, I agree with you about Wikipedia aims and purposes and I will not insist on considering this topic any more. It seems this is just a question of taxonomy and it is not so important for the root of the matter. My aim was only to say something more about the nature of the memristor in a more intriguing way. Circuit dreamer (talk, contribs, email) 06:49, 20 March 2011 (UTC)
Empathy can help understanding memristor
I have the ambition to grasp the basic ideas behind "mem" elements by intuition and to explain them to people because the situation is very interesting and indicative. From one side, there are extremely simple and intuitive "mem" elements and circuits; from the other side, there is an extremely complex theory that cannot explain them in a human friendly manner and even misleads and prevents understanding. All the debates on this talk page (including the archives) confirm this contradiction.
I was emotionally connected with and fascinated by the elegant simplicity of Widrow's memistor idea in the late 60's when I (a schoolboy) wanted to make such an "integrating transistor" by bare pencil graphite. I remember also how in the middle 80's I tried to present to my students in the laboratory the time behavior of a capacitor and inductor by moving uniformly the slider of a current- and voltage-driven rheostat (I even planned to make a computer drive a digitally-controlled resistor). Then I did not know that we together (I and the humble rheostat:) actually emulated the Chua's memristor; "we" acted as a memristor:) I came to know about it a few days ago when Blm19732008 directed my attention to Pershin & Ventra's "mem-emulator" (this article is interesting and because it makes a connection with no less mystic gyrators and capacitance multipliers but let's discuss this topic later).
Now, 25 years later, I intend to repeat these empathy experiments (identifying with investigated objects with the purpose to understand them) in the laboratory to show to my students what a memristor actually is. The idea is the same as this one in 80's and as the Pershin & Ventra's emulator. First, we will supply a variable linear resistor (a rheostat) with constant current and will move uniformly the slider from zero to maximum resistance to obtain a memristor acting as a "resistive capacitor". Then, we will supply the variable resistor with constant voltage and will move irregularly the slider from maximum to zero resistance to obtain a memristor acting as a "resistive inductor". I intend to realize this computerized experiment by connecting the variable resistor in the place of the resistors R1 and R2 of an op-amp inverting amplifier so that the rest of the circuit acts as a perfect current-to-voltage and voltage-to-current converter. To obtain the Pershin & Ventra's emulator, we have to make the computer drive a digitally-controlled resistor R1 or R2. Circuit dreamer (talk, contribs, email) 13:06, 20 March 2011 (UTC)
Memristor is a conditionally nonlinear device
Memristor is not simply a nonlinear device as this article and many other sources claim; instead, it is a conditionally nonlinear device. For example, in the case of an integrating memristor, if we wiggle rapidly the input quantity (AC input signal), the memristor does not change its resistance; so it behaves as a linear (ordinary, ohmic, constant, steady) resistor. If we change slowly the input quantity, the memristor changes its resistance; so it behaves as a nonlinear resistor. Let's see this phenomenon in a few simple and clear memristor examples.
Electrothermal. Imagine we connect a constant current source to a thermistor with PTC fixed to a massive radiator (or simply a lamp with a thick filament, a solid heater, etc.) The constant current begins flowing through the thermistor; as a result, its temperature TR, resistance R and voltage drop VR begin increasing slowly. So this arrangement acts as an integrator with current input and resistive output, as a "resistive capacitor". If we change slowly enough the input current, then TR, R and VR will change as well. If we change rapidly the current, then TR, R and VR will not change at all.
Electromechanical. The example above (a geared motor driving a rheostat connected in series with it) is even more expressive. If we change slowly enough the input current, the motor will manage to move the rheostat slider and the resistance will change depending on the current and time; the "memristor" will act as an electromechanical integrator (a "resistive capacitor"). If we change sharply (or wiggle) the current, the motor will not react to our intervention at all. The resistance will stay constant and the "memristor" will be just a humble ohmic resistor. Circuit dreamer (talk, contribs, email) 06:51, 22 March 2011 (UTC)
Conclusion. For quick input changes the integrating memristor behaves as a linear (ohmic) resistor; for slow input changes it behaves as a nonlinear resistor. Shortly, an integrating memristor behaves as an inert nonlinear resistor. Please, discuss. Circuit dreamer (talk, contribs, email) 22:16, 21 March 2011 (UTC)
Memristor can emulate basic electrical elements
(revealing the connection between memristor, gyrator and multiplier)
It is worth to show the remarkable property of memristor to mimic (emulate, simulate) basic passive electrical elements (a resistor, capacitor and inductor) since it is convenient, for some reasons, to replace them by their circuit equivalents (a gyrator, multiplier, memristor...) It is important to say, in the very beginning, that these circuits emulate only particular properties (e.g., time behavior) of the genuine elements. Let's first see what these properties are.
Genuine elements. The general property of passive electrical elements is taking (consuming) energy from the input excitation source; resistors dissipate this energy while capacitors and inductors store (accumulate, "steal") it. But how do they do it?
Let's for concreteness imagine that the considered passive element is connected in series to the exciting voltage source. What does it do in this case? It subtracts voltage from the input voltage: the resistor creates opposing voltage drop across itself while the capacitor and the inductor create opposing voltage. Resistors do this by throwing out (dissipating) energy while capacitors and inductors do it by taking energy from the excitation source, accumulating it into itself and setting it against the input source. In the first case there is a voltage drop while in the second case there is a voltage. This means that we can emulate these elements by other elements having the same but opposing voltage (having contrary to the excitation voltage polarity when travelling along the loop). So, we can emulate these elements by replacing them with some other elements producing the same voltage.
Emulating by varying voltage source. First, we may replace them with varying voltage sources and this is the most natural way of making emulated capacitors and inductors (as they behave as varying through time voltage sources). Op-amp gyrator, multiplying, memcapacitive (fig. 1b) and meminductive (fig. 1c) circuits do it in this way. In these circuits, the op-amp output voltage represents the voltage across the according capacitor or inductor (see more about the topic in Demystifying gyrator circuits).
Emulating by varying resistance. But a memristor can do the same by replacing these voltages by equivalent voltage drops across dynamic time-dependent resistors. Transistor gyrator and multiplying circuits do it in a similar way.
Conclusion: To emulate passive elements, we may replace elements behaving as resistors with sources and elements behaving as sources - with resistors (besides the more natural "sources - with sources" and "resistors - with resistors").
Thus we have made a connection between apparently different memristor, gyrator and multiplier. It is even more interesting to make a connection with the true (absolute) negative resistor. What is a true negative resistor? What does it do? Here, the ordinary ohmic resistor is replaced again with a voltage source (exactly as in the case of gyrators and multipliers) but it is with reversed polarity. It has the same polarity as the input voltage source so that it adds the same energy that the equivalent "positive" resistor would consume. Please, discuss. Circuit dreamer (talk, contribs, email) 10:41, 24 March 2011 (UTC)
Demystifying the circumstances about memristor rising
I have read with great pleasure and interest the talk (especially archive 1) and have got a good intuitive notion about the essence of the memristor idea. It is extremely interesting for me to unriddle the mystery of the basic Chua's postulates about memristance not giving people peace as many as forty years. First, I will show what the problems with this formulation are; then, I will give some recommendations how to solve them. I would like to share my insights with you in a bit nontraditional emphatic manner by tracing back the evolution of Chua's idea and trying to guess how he was thinking when propounding his memristor theory. Of course, only the very Chua can say if the hypothetic story below is true. For concreteness, let's consider a current- (charge-) driven memristor although it can be voltage- (flux-) driven as well. Also, let's consider an integrating memristor although it can be differentiating or some else as well.
In 1826, Ohm formulated his law (V = R.I) establishing a linear relation between voltage and current if only the resistance stayed constant. In the case of nonlinear resistance (depending from the input quantity - voltage or current), it is written in a derivative form (dV = r.dI) where r is the incremental resistance in the given point.
In 1971, Chua simply integrated Ohm's equation ( ), what gave nothing new if the resistance stayed still constant. Then he made the resistance be nonlinear (depend on the input quantity - the charge Q in this case) and name it memristance, expressed the right time integral of current as charge Q and began speculating how to express the left time integral of voltage. Inspired by his speculation about the existence of a 4-th passive element, he saw a connection with the magnetic flux Φm in an inductor and did three crucial actions changing the world for the next forty years:
- He denoted an integral of a pressure-like quantity (voltage) with a flow-like name (flux).
- He named an electric quantity (time integral of voltage) exactly with the same name as the similar integral magnetic quantity.
- He denoted the new artificial quantity exactly with the same "magnetic" letter (Φ) and with the same "magnetic" subfix (m) as the genuine "magnetic" quantity.
The results of this misleading denoting were tremendous. Any intuitive notion about this extremely simple device disappeared. People, knowing what such an integrating resistor was, ceased understanding it; they began trying to see something "magnetic" in this pure resistive device. Others began looking for some connection between Φm and the stray inductance inherent for every electrical element including resistor. The rest of them, relying on common sense and realizing that there was nothing magnetic here and the stray inductance is unrelated, began looking for some way of justifying the misleading denoting (to explain that, for the purposes of theory, it is written magnetic but it is not magnetic; it is written "flux" but it is not a "flux", etc.:) Memristor produced a sensation and media intensified all these problems...
Please, discuss and suggest remedies for correcting these misconceptions to improve the article. Circuit dreamer (talk, contribs, email) 11:11, 27 March 2011 (UTC)
- You cannot post an uncited diatribe on the talk page and then start amending the article to comply with it over references in the article. I have reverted most of what you put in (I tried to undo specific edits but you had made too many changes to allow for this). As I said in the edit summary, the Chua reference talks about magnetic flux linkages so there is no justification in the article qualifying this unless there are other sources that can be quoted. Likewise, later edits have introduced concepts into sections cited to the HP reference which cannot be found in that source. SpinningSpark 23:10, 27 March 2011 (UTC)
- And God said, "Let resistive elements be magnetic! Amen!":)
- I have a great idea about a colorful picture on the top of the article representing the situation: Prof. Chua, longing to convince his students of the magnetic nature of memristor and to evoke their interest in electrical phenomena, holds a big magnet in his hands and tries to gather a heap of HP memristors scattered on the table in front of the curious students:)
- I have also a big idea about a funny top picture for another sensational article - Deborah Chung's "apparent negative resistance": Prof. Chung creates a negative resistance in front of her students by using a humble bridge circuit hand-made by two pencil graphites (see the talk if you want to see what the idea of such "apparent negative resistance" is:)
- To second my university colleagues in their educational efforts, I give a promise to make a humble memristor of pencil graphite and to investigate it in front of my students:)
- Of course, this was only humor but the problem is serious and we have to solve it somehow. I suggest the simplest but effective solution - simply to not mention "magnetic flux linkage" in the lede; to revert the article to its old state without Blm19732008's insertion "...and is expressable in terms of a functional relationship between charge and magnetic flux linkage". I have already done it. Circuit dreamer (talk, contribs, email) 14:42, 29 March 2011 (UTC)
"Magnetic flux Φm" is neither "magnetic", nor "flux", nor "Φm"
Blm19732008, how do I persuade you that "magnetic flux Φm" used here is neither "magnetic", nor "flux", nor "Φm"? And that this term should not be used in the literal sense here as it would duplicate the genuine magnetic variable? It can be used here only figuratively, as a metaphor; we should say "this variable is similar to magnetic flux" but not "this variable is a magnetic flux"... I will try to persuade you of this obvious truth by a simple example. Imagine an inductor driven by a constant voltage source. In this arrangement, the current through the inductor is proportional to the time integral of the voltage across it (i.e., to Chua's "magnetic flux Φm"). But the current flowing through the inductor creates another (genuine) "magnetic flux Φm". It turns out that there are two duplicated variables (one "synthetic" and another - genuine); we have denoted two absolutely different quantities by the same term... Circuit dreamer (talk, contribs, email) 20:23, 7 April 2011 (UTC)
- The way to persuade is to provide a source, not by endlessly delivering lectures on the talk page. Save it for your students. SpinningSpark 00:20, 8 April 2011 (UTC)
- Do we really need sources for obvious truths? Circuit dreamer (talk, contribs, email) 02:37, 8 April 2011 (UTC)
- Yes, and this does not come under the heading of "obvious truth" anyway, at least to most mere mortals. Chua definitely speaks of magnetic flux in his original paper, and appears not to mean simply the time integral of voltage. It is not clear that the devices with V/I constitutive relations, such as the HP device are in the same class as that defined by Chua. It may be that there is some class of constitutive relations of Q and Φm which when differentiated yield hysterisis-like constitutive relations of V and I. I have searched for a simple analytic example of such a case but I am probably not a good enough mathematician to succeed. We definitely need sources discussing this issue before writing anything at all on it. SpinningSpark 16:22, 8 April 2011 (UTC)
- Do we really need sources for obvious truths? Circuit dreamer (talk, contribs, email) 02:37, 8 April 2011 (UTC)
- No one knows what kind of ideas inhabit "beautiful minds"; we "mere mortals" can only guess... Chua is a theoretical genius and definitely he had good reason to name it "magnetic flux Φm". The problem is that this term is contrary to our notion about the constituents of this name and we "mortals" have to solve somehow this contradiction.
- I like your speculations about Q/Φ versus V/I constitutive relation memristor. IMO we have to arrange the lede and the introductory part of the background section according to these two viewpoints at the same element (now, the background section is quite scattered). Circuit dreamer (talk, contribs, email) 15:48, 9 April 2011 (UTC)
- we can only guess...not so, we can read the sources. A lot of this is clarified in Di Ventra et al. 2009, and apparently avoiding directly contradicting Chua at the same time. I have obtained this paper and I think you will probably like what it says, but I don't have time to go through it in detail right now. In the meantime, please don't make stuff up yourself to put in the article. You must surely know by now that the Wikipedia way is to write from the sources. I strongly encourage you to read sources, then write; not to write, and then hope for a source to back it up. SpinningSpark 18:37, 9 April 2011 (UTC)
- It is just superb! You have guessed right: I do not simply like it; I love it since it says what I would say! I only regret that I did not read it carefully a few weeks ago when Blm19732008 kindly directed my attention to it (sorry!) IMO this article is our salvation from the vicious circle where we are now. As I can see, after almost forty years, the authors corrected tenderly (not so rudely as me:) the "magnetic flux" problem and the very Chua approved this revision by the fact of his participation. So, here is the solution - we should mention the original Chua's article but we should use Di Ventra et al. memristor definition:
- "The memristor is characterized by a relation between the charge and the flux, defined mathematically as the time integral of the voltage, which need not have a magnetic flux interpretation" (I would put flux in quotes but I know you will not accept it).
- How simple it was: first, to name the variable; then, to define it mathematically and finally, to distinguish it from the similar magnetic one. Circuit dreamer (talk, contribs, email) 22:44, 9 April 2011 (UTC)
- It is just superb! You have guessed right: I do not simply like it; I love it since it says what I would say! I only regret that I did not read it carefully a few weeks ago when Blm19732008 kindly directed my attention to it (sorry!) IMO this article is our salvation from the vicious circle where we are now. As I can see, after almost forty years, the authors corrected tenderly (not so rudely as me:) the "magnetic flux" problem and the very Chua approved this revision by the fact of his participation. So, here is the solution - we should mention the original Chua's article but we should use Di Ventra et al. memristor definition:
"Mem" elements do not obligatory possess hysteresis
I do not share Chua's, Ventra's and others' (including this article) assertions about the (unconditional) existence of hysteresis in memresistive, memcapacitive and meminductive elements (see for example [3]). Hysteresis is a rate-independent phenomenon like nonlinear resistance. This means that there is no matter if we drive a hysteretic element by slow (DC) or rapidly (AC) changing input quantity; it will show different forward/backward paths (what is the definition of hysteresis) of its IV curve in both the cases. This is not true for a volatile memristor, memcapacitor and meminductor that exhibit "hysteresis" only in the case of slow changing input. If we wiggle the input rapidly, their bizarre ("pinched") hysteresis loop will become a humble straight line (for example, imagine an inert thermistor).
If we assume that volatile "mem" elements possess hysteresis, we should accept that the ordinary ("not mem") capacitors and inductors possess hysteresis as well. For example, if we drive a capacitor with low frequency AC input current, measure the voltage across it as an output and draw its "IV curve", we will probably see a "hysteresis" loop (I am not sure if it will be "pinched"; please, check it:)
It seems only nonvolatile "mem" elements (i.e., true memory elements) should possess hysteresis since their IV curve does not depend on the input rate? Please, discuss this extremely interesting topic; I would like to know if I am right. I have copied this text to hysteresis talk as well. Circuit dreamer (talk, contribs, email) 09:36, 10 April 2011 (UTC)
Invalid Equations for Resistance, Capacitance, and Inductance
Capacitance is C = Q/V, not dQ/dV, Inductance is L=Φ/I, not dΦ/dI, Resistance is R=V/I, not dV/dI. dV/dI has a name and is called differential resistance, which is not the same as resistance. Only when the resistance does not depend on the current, can we solve for V and take the derivative over the current giving dV/dI=R. Similar statements apply for C and L. The memresistance of a charge-controlled memristor, on the other hand, is defined as M = dΦ/dQ, and MUST depend on Q, to make it different from a regular resistor. If M did not depend on Q, then the memristor would just be a regular resistor, with a constant resistance that does not depend on the current or charge. Memresistance is not analogous to resistance, capacitance, and inductance. If you want to compare to resistance, capacitance, and inductance, then you must compare to Φ/Q, which is not the same as M. Φ/Q = ($MdQ)/Q, the ratio you want is the integral of M over Q, divided by Q. I don't recall Chua giving a name to this ratio in his original paper. Humanoid (talk) 21:16, 17 May 2011 (UTC)
I have fixed the invalid equations Humanoid (talk) 03:38, 25 May 2011 (UTC)
- M = R = V / I (constant resistance) is not interesting here. As it was written, the table included the nonlinear quantities as well. R = V/I, C = Q/V and L = Φ/I are special cases of R = dV/dI, C = dQ/dV and L = dΦ/dI. Circuit dreamer (talk, contribs, email) 04:14, 25 May 2011 (UTC)
- R doesn't have to be a constant. Even in a regular resistor, R changes depending on the voltage (or current), and only remains approximately constant within small intervals. You can write R(I) = V(I)/I or R(V) = V/I(V) to emphasize that R may depend on I or V. And for M, you may write M(Q) or M(Φ) to emphasize that it depends on the charge, or magnetic flux. The basic quantity of interest in resistors is their resistance, not differential resistance. Resistance is not a special case of differential resistance. Humanoid (talk) 18:55, 26 May 2011 (UTC)
Linear vs. constant
Have I misunderstood something, or is the following sentence from the Background section in error?
- "However, as mentioned above, if it has no non-linearity then it is the same as a standard resistor."
It seems clear from the more mathematical sections below that if M(q) is constant then it's the same as a resistor (and M=R), but if M(q) is a non-constant but linear function then it's not.
89.204.153.129 (talk) —Preceding undated comment added 16:21, 28 August 2011 (UTC).