I am looking for a circuit that can, on input, remember a certain voltage and output that voltage indefinitely even after the input has been taken away. The circuit should not change its output until a new input has been provided.

I understand that such a circuit can be made by digitally sampling the input up to some arbitrary resolution, but I would like to know if a simple analog solution is possible.

I would also like to keep this solution purely electronic, as I can also envisage a mechanical solution in which a feedback circuit mehanically controls a potentiometer.

Finally, I would ideally not like the circuit to rely on the passive stability of any floating inputs. The circuit should be stable for at least hours.

  • \$\begingroup\$ Simply put, it doesn't exist and I know I'm sticking my neck out so, if anyone has a good answer to this I'm ready with the upvote button LOL. \$\endgroup\$
    – Andy aka
    Aug 12, 2014 at 20:15
  • 7
    \$\begingroup\$ The best way to do this is digitally (convert A->D, store, convert D->A). It's possible to do this in an analog fashion, but it will be expensive, have limited accuracy and prone to errors from humidity etc. \$\endgroup\$ Aug 12, 2014 at 20:21
  • 9
    \$\begingroup\$ This circuit is called "sample-and-hold". Stability for hours will be difficult. \$\endgroup\$
    – markrages
    Aug 12, 2014 at 20:21
  • 1
    \$\begingroup\$ Single polarity? Can I use a relay? Aluminum electrolytics hold charge for a long time... in a temperature controlled box? Why not digital? \$\endgroup\$ Aug 12, 2014 at 22:25
  • 1
    \$\begingroup\$ @Andyaka See my comment on Sphero's answer. I think the ISD 256 ANALOG levels in a digital bit may qualify for your upvote :-) - even if it's only a comment. \$\endgroup\$
    – Russell McMahon
    Aug 13, 2014 at 9:59

6 Answers 6


This is not a practical answer unless you happen to work for a company with the resources of, say, Intersil, but the technology exists to make this work. Consider the ISL21080 type references which hold charge, hopefully for the life of the equipment they're installed into, based on a tiny capacitance isolated by quantum tunneling effects. Provided they don't get too much in the way of X-rays etc. they'll remain pretty stable for years. See, for example, this application note.

enter image description here

I might add this kind of thing gives me the willies.

For an ordinary application, digital is most likely the way to go.

  • 1
    \$\begingroup\$ This is how cheap analog voice recorders and greeting cards work. Example: kowatec.com/prod/ap/doc/apr6016-v13.pdf \$\endgroup\$
    – Phil
    Aug 13, 2014 at 2:00
  • 2
    \$\begingroup\$ @phil: No, it's not. From your link: "A maximum of 30K bits of digital data can be stored." It also mentiond 256 signal levels -- that's discrete storage, not analog. \$\endgroup\$
    – Ben Voigt
    Aug 13, 2014 at 2:58
  • 2
    \$\begingroup\$ @BenVoigt The ISD speech recorder ICs stored 256 analog levels in a "charge well" originally intended to be a digital bit store. That's analog storage in a notionally digital bit. \$\endgroup\$
    – Russell McMahon
    Aug 13, 2014 at 4:25
  • 4
    \$\begingroup\$ @BenVoigt - I didn't explain that clearly enough. In the case of ISD they took a charge well in a technology ORIGINALLY intended to be used as a 1 or 0 bit store and instead in each "1 digital bit" charge well they stored incremental charges in 256 levels (they may use a companding law). They then sensed the amount of charge stored (by specific means unknown to me) and obtained a 256 analog level / 8 bit equivalent word. By doing this they increased the amount of available speech memory by a factor of 8 as what would usually take 8 bits (2^8 = 256) was able to be stored in a single bit space. \$\endgroup\$
    – Russell McMahon
    Aug 13, 2014 at 9:58
  • 1
    \$\begingroup\$ @RussellMcMahon The gate can hold only discrete electrons, so the 'analog' voltage is limited to around 25 bits resolution for a 10pF-ish gate and 1V-ish voltage. ;-) \$\endgroup\$ Aug 13, 2014 at 10:34

EEPROM technology split into two branches in the early 1980s - one thin oxide (FLOTOX) with Intel and Seeq and the other thick oxide (Xicor). In the early days, there were weaknesses to both routes. Thin oxide leaked charge and thick oxide was inherently impossible to scale. There were other issues, but they don't apply here.

Given the thick oxide didn't "leak" electrons, I asked the designers at Xicor about the theoretical resolution of a single thick oxide cell if we discounted the limitations of the sense amps, and they said it could approach 1ppm (roughly 20 bits). Since I was also associated with LTC, which was one of the leaders in precision voltage references that were inherently power hungry, that led me to think a single EEPROM cell could be adapted to be a high precision and very low power voltage reference. My longer term thinking was this technology could be developed further for use in AI and used in conjunction with n fan-out non-blocking, reconfigurable non-volatile multiplexers.

Fast forward about 15 years - Xicor ended up developing such a device and then subsequently was acquired by Intersil. Given the inability to scale, the longer term vision is probably not practical. However, other technologies can enable the vision when combined with a software software reconfigurable mux.


This device does exist although it's not readily available in single unit quantities, its output amplifiers will get in the way and it's very non-linear.

It is a Floating Gate MOSFET, used in Flash memory, EEPRom and the ilk. The programming charge can be variable though somewhat unpredictable as the FN tunnelling (Fowler Nordheim) will be variable across the die. While non-linear it is a proportional effect so you could imagine designing a circuit that linearized the programming effect (of Vth shift). It will be stable over weeks to months so it meets the requirements of hours that you say you'd need.

But a lot depends upon the specifications that you need, how much drift is acceptable etc.

Just to be clear here, I am talking about the individual device/transistor not the complete component as the support circuits of a Flash will prevent you from operating the cells in this way.

Here are 3 references from an EDN article talking about a company called GTronix which was acquired by National Semi (now TI).

Lee, BW, BJ Sheu, and H Yang, “Analog floating-gate synapses for general-purpose VLSI neural computation,” IEEE Transactions on Circuits and Systems, Volume 38, Issue 6, June 1991, pg 654.

Fujita, O, and Y Amemiya, “A floating-gate analog memory device for neural networks,” IEEE Transactions on Electron Devices, Volume 40, Issue 11, November 1993, pg 2029.

Smith, PD, M Kucic, and P Hasler, “Accurate programming of analog floating-gate arrays,” IEEE International Symposium on Circuits and Systems, Volume 5, May 2002, pg V-489.

THere is another class of device that is called a MNOS transistor (Metal Nitride Oxide Semiconductor) in which tehere are two dielectrics in the gate, one of which is Si3N4 which has a lot of traps. This device operates very similarly to the flash cell above.


I left a comment and thought about it for a minute and will say with hopeful certainty that it does not exist - some drift away from the "sampled" voltage is not only likely but a certainty. Resolution is critical it seems, (as implied in your question) and this is why I say it doesn't exist. Noise is another factor that will reduce the fidelity of what you have sampled.

Even a digital system (with more than enough resolution) will be inaccurate in reproducing the voltage you apparently "stored". Anything taken to limits will be a problem. The potentiometer idea (suggested in the question) is also flawed because it relies on the reference voltage across its terminals being kept (or reproduced) - you cannot know how these things minutely drift but, again, it's all down to accepting an error or rejecting that error.

  • \$\begingroup\$ Minute drift and noise is to be expected in all circuits, and I am not concerned with extremely precise reproduction, just something that will be a least slightly more stable than a free floating input. I would be satisfied with a digital memory solution, but was wondering if a more direct solution existed. Thanks for the input. \$\endgroup\$ Aug 12, 2014 at 20:37
  • \$\begingroup\$ @user2640461 I have a number of these still - I used to make "talking" disability aids using these as the speech store. 8 bits/256 levels stored as analog level in a usually-digital bit. studyelins.blogspot.com/2012/11/… \$\endgroup\$
    – Russell McMahon
    Oct 21, 2020 at 0:16

Yes, the ISD chips work this way. In fact an inventor in the 1990s claimed to have found a way to store an entire 1 hour movie in a 16MB analog memory chip.

The problem turned out to be (Yup you guessed it!) voltage drift over time. Sure the chip would store your movie just fine for a day, possibly two but even powered eventually it would degrade beyond use because the individual values stored could not be recovered without referring to the original file. I actually looked into using this to store SSTV signals but ran into the same problem, conventional floppy disks or even VHS tape were far more reliable.

  • 1
    \$\begingroup\$ Interesting hack: I came up with this while tinkering with the idea of using ZnS glow in the dark material as a way to store data. \$\endgroup\$
    – Conundrum
    Oct 10, 2016 at 7:39

Here's the best I can do that approaches those (unrealistic) requirements:


simulate this circuit – Schematic created using CircuitLab

The "sample" switch should be a mechanical relay in which the contacts are truly disconnected from the input. This will eliminate parasitic currents that will cause drift if using a semiconductor switch. However, in practice, you will like use a high-performance analog switch to keep the design solid-state. You want a switch with a very high impedance in the off state. Ironically, the on-state resistance isn't important. You also want the lowest leakage/bias currents possible for long term stability.

Now for the amplifier treatment. You're going to want an ideal amplifier; one that comes from a first year analog design textbook. Unfortunately, Digikey has been out of stock of those for some time now. So the best substitute you can do shall be as close as possible to these specs for long term stability:

  • Infinite open loop gain
  • 0pA input bias current
  • 0pA input offset current
  • infinite CMRR
  • 0ppm/degC temperature coefficient
  • 0pV input offset voltage

There may other specs I'm missing but the point is these non-idealities compromise the long-term performance (and short-term if the specs are poor enough). Sure, you can get fancy and do some tricks to mitigate these effects at the expense of circuit simplicity, but you can't eliminate them and the output will still drift.

C1 ideally will have no leakage current. In reality, this will be a high-quality film capacitor of relatively large value as to not be so influenced by the amplifier parasitics.

Then, put the whole circuit in a iso-thermically controlled crucible to meet temperature drift performance.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.