Hey all-

I am having problems using a JFET as a variable resistor. For one, the information on the subject is vague. I'd like to see if anyone could give some advice on using the datasheets to find the useful resistance range, i.e. Rmin to Rmax.

The problem I am having is with the JFET tolerances. My circuit works, and works extremely well, however, 33% of JFETs don't measure to the spec I need (too high or too low R min). How do I work around this problem, especially since this is an all SMD design?

Maybe someone with manufacturing experience could help. Could I buy JFETs tested at the factory? Can you test before or during assembly and have the "bad" parts kicked out of the placement machines, or do you just have to test the whole assembly and pitch it if it doesn't test well? It would seem like this is a common problem with JFETs and I'm not the first person trying to overcome this.

Thank you in advance for your help.

 

Post a diagram of your circuit please.

 

Thanks for taking interest Dodgydave!


The JFET (an MMBF5484) is acting as R1 of a voltage divider (with a 100 ohm resistor in series before it), and our load would be R2 of the divider. I'm using 2x 1Meg feedback resistors as described in various sources on the subject to linearize the JFET. The control voltage is a negatively rectified input signal which is averaged by R2 and C1.

I made up 25 boards, and in the majority the channel resistance is approx 305 ohm, but there's a small group around 275 and small group around 350. Those small groups are throwing the whole thing off and are the problem I need to overcome.

I see two potential solutions, which is why I asked the questions in my first post. I could use a JFET with a higher R-on, and strap a fixed resistor across the JFET thus minimizing the tolerance error. However, I haven't found a reliable equation for calculating R-on Min and R-on Max which is crucial. Anyone know these? I found the MMBF5484 by trial and error.

The other solution is to fix it during manufacture, but I don't know what kinds of quality control steps can be taken here, that's question #2.

 

Thanks for taking interest Dodgydave!


The JFET (an MMBF5484) is acting as R1 of a voltage divider (with a 100 ohm resistor in series before it), and our load would be R2 of the divider. I'm using 2x 1Meg feedback resistors as described in various sources on the subject to linearize the JFET. The control voltage is a negatively rectified input signal which is averaged by R2 and C1.

I made up 25 boards, and in the majority the channel resistance is approx 305 ohm, but there's a small group around 275 and small group around 350. Those small groups are throwing the whole thing off and are the problem I need to overcome.

I see two potential solutions, which is why I asked the questions in my first post. I could use a JFET with a higher R-on, and strap a fixed resistor across the JFET thus minimizing the tolerance error. However, I haven't found a reliable equation for calculating R-on Min and R-on Max which is crucial. Anyone know these? I found the MMBF5484 by trial and error.

The other solution is to fix it during manufacture, but I don't know what kinds of quality control steps can be taken here, that's question #2.

That circuit is usually used as part of an automatic level control circuit. If you want to have constant amplitude at the input, you need a high gain threshold detector in front of it.
See the attachment.This is a Wien bridge oscillator. The voltage divider on the base of Q12, plus the Vbe of Q12 and the forward drop of D1, set the threshold. When the peaks of the sine wave at the output of U1 exceed this voltage, C3 charges to raise the gate voltage of the FET (p-channel) to whatever level is required to get the ON resistance to the level that stabilizes the oscillation amplitude. The wide range of Vgs required to get the correct Rds is accommodated by the very high gain of the threshold circuit.

 

Thanks Ron-

I actually need it to go the other way, roughly constant output. I don't care what the input is, I need the output to be determined by the input but not in a linear fashion like a pure divider. Unfortunately, I can't add many components either.

For my use, it seems from experimentation that all that matters is the specific starting Rds of around 305 ohms (+/- 5 ohms), when I measure Rds in circuit with no input. That's it, the JFET seems to take care of the rest perfectly when that condition is met. If not, it doesn't work correctly.

 

Thanks Ron-

I actually need it to go the other way, roughly constant output. I don't care what the input is, I need the output to be determined by the input but not in a linear fashion like a pure divider. Unfortunately, I can't add many components either.

For my use, it seems from experimentation that all that matters is the specific starting Rds of around 305 ohms (+/- 5 ohms), when I measure Rds in circuit with no input. That's it, the JFET seems to take care of the rest perfectly when that condition is met. If not, it doesn't work correctly.

The circuit I posted does give constant Rds, because that is what is required to keep the oscillator running, but not to saturation of the op amp.
If you provide the big picture of your application, including input waveform, amplitude, and frequency, and the desired output, we might be able to help.

 

Ron I think you misunderstood. I need Rds to vary with input. The problem I am having is that the tolerance of JFETs is poor and I need to overcome this.

For my use, it seems from experimentation that all that matters is the specific starting Rds (or Rds min) of around 305 ohms (+/- 5 ohms), when I measure Rds in circuit with no input. That's it, the JFET seems to take care of the rest perfectly when that condition is met...

There's two questions I'm trying to find an answer for:

One, how do predict a JFET's useful range of Rds from the datasheet? If I am setting out to pick a JFET with a useful variable resistance, for example 300 ohm to 500 ohm, how do I determine this? What is the formula(s) that say(s) "if Vgs = 0, then Rds = x, but when Vgs = -3, Rds = y"?

Two:
If we find the correct JFET that will give us our desired x to y range, how do we deal with poor tolerance, when the JFET's x and y are really x/y +/-25%? Ron I think you were suggesting in your circuit to use feedback to null out this tolerance variation, but I can't. So what can be done during manufacture to get rid of those tolerance variations?

So, I have 2 options, swamp the JFET with a low tolerance fixed resistor (which requires an answer to question number 1) or fix this somewhere during the manufacturing process (which requires an answer to question number 2).

 

Like I said, if you won't give us the bigger picture, then you're probably out of luck here. There are generally several ways to skin a cat. I'm not nosy, or trying to steal your idea. There have been many cases on this forum where the OP was overlooking a totally different solution to his problem.

 

Maybe someone with manufacturing experience could help. Could I buy JFETs tested at the factory? Can you test before or during assembly and have the "bad" parts kicked out of the placement machines, or do you just have to test the whole assembly and pitch it if it doesn't test well? It would seem like this is a common problem with JFETs and I'm not the first person trying to overcome this.

Thank you in advance for your help.

Yes some placement machines can be programmed that way. All the program has to do is to push the JFET on the PCB a little. Then you need a contact pad on the bottom for supply/measurement (via PC for instance). If the part isn't OK, it is again moved somewhere (discarded), and a new JFET is picked.

Depends on the volume of the PCBs but I don't imagine this would take more than 1/2 a day efforts to work this out.

The software for the placement machine needs to be flexible so it can be interfaced to a PC (which samples the parameters), and then take different branching.

Semiconductors aren't usually high precision. Sometimes parts are adjusted directly after manufacturing using a laser, or they are sorted into categories (BJT A,B, and C for instance).

 

Thank you takao21203!

Ron, I don't think you are stealing ideas or anything; you are a respected member here. But it's not necessarily you I'm worried about.

It's a two-fold use in a hearing aid for feedback suppression and compressor. We're only dealing with low level <2V, mostly midrange audio signals and it works perfectly as is except for when the default Rds is too low, you can potentially harm ears as it gets too loud. When it's too high, you can't hear the signal well enough. These two cases happen 33% of the time from a small sample.

Now, you see why I can't add components as the board must be no bigger than a dime, and we can't get into the mic/gain system to apply feedback since it's being added to an OEM part. We can only insert between the preamp source and the speaker.