Working with JFET tolerances

[Stupid Bear]

I'm a bit confused with these tolerances for an amp in designing with a JFET input. The Vgs PinchOff is listed as min -2.0 and max -5.0. Is there a reason the limit has such a large spread? Is it based on how much current I put through it or is it just variation in manufacturing? How do I design around this? Similar tolerance spread for the Id current. How would you approach this?

[TimFox]

The usual primary parameter for a JFET is Idss:  drain-source current at zero gate-source bias voltage.
For a given production, it varies by a very large ratio;  the 2N4416 is specified between 5 and 15 mA, and the transconductance gm at 0 bias varies with that value.
Some JFETs are available binned into narrower ranges:  the Sony 2SK152 (of blessèd memory, we shall not see its like again) had a range from 9.5 to 42 mA, but was available with these "ranks":
-1    9.5 to 14.8 mA
-2  13.4 to 21.0 mA
-3  19.0 to 30.2 mA
-4  27.4 to 42.0 mA
The gm vs. Id curves for different values of Idss lie upon each other, up to the maximum value of Id for each Idss value.
https://www.silicon-ark.co.uk/datasheets/2sk152-datasheet-english.pdf
Page 229 of that reference shows the dependence of the input noise voltage on drain current, which decreases with increasing current.

If you know the maximum Idss for the devices you will use (either manufacturer's data or your own binning), you must design the bias circuit to keep the Id value below that Idss, to avoid forward-bias of the gate junction.

[exe]

What's your application?

[MathWizard]

Why is it so hard or expensive after all these years, that they don't make common JFET's or mosfet's more matched ? Who would complain if they were more matched ??

[Stupid Bear]

This is going to be the input for an audio amp. It will take in a guitar input. I was going to hold Vgs at around -1 V and the amp will run off a 9V battery. Simulating this in spice I was wondering what would happen as the battery goes down to about 6V and started getting a drop in out output and non-linearity from the FET. I can compensate but with a drop in Av. I'm trying to peg this in the sweet spot so I can use it with the full range of 6-9.6V but I don't know how to work around the tolerances being that the JFET I build with may not act the same as the model. I was curious what your thoughts were on going about the design and how you would think about the problem. Thank you.

[Kleinstein]

There is quite some scattering of the parameters in production. This is mainly a 1 parameter spread and one could use Idss or the threshold voltage as the 2 are high correlated. Another linked parameter is the on resitance, when used as a switch. It depends on the application which range threshold / Idss is preferred.
Quite often there are also related parts, roughly sorted by Idss (e.g. J111, J112, J113)
Some types come more or less sorted (usually Idss as parameter) as different subtypes (e.g. with a -3,-4,... or a color code) with a slightly smaller spread.

Quite often the parts from a single batch don't scatter all that much as the datasheet allows. This helps a little with finding matching parts.  However one still does no know which range one will get.

Usually one designs the circuit to tolerate quite some spread. If needed have a resistor that is adjusted/selected for the actual part to compensate for the spread.
In differential and similar circuits one may opt for selected / matched parts. So extra tests are done on the parts and suitable pairs (e.g. difference in the treshold < 50 mV) are selected.
The need for trim or selction is a reason what JFETs lost a bit on populartity.

They seem to be able to make well matched JFET pairs.  (e.g. JFE2140 with  < 4 mV offset).
AFAIK the threshold depend on how long / hit / far the diffusion runs. Here only minute (e.g. <1 K range) variations in the oven temperature and maybe the inital oxide thickness can make quite a difference in the result.
It is still the question why they don't have at least some common types (e.g. the PN4393)  with better sorting / testing. Instead they have a crazy number of JFETs that are nearly identical with more minor variations.

Additional sorting, matching adds a bit to the costs and one would end up with even more types.  With some MOSFET types (e.g. BSS169 and a few related) they match and sort, but somehow don't want to keep them as seprate articles. I don't understand this - at least for a few types this could add quite some value to the parts.

p.s.:
For an audio amplifier one can often use AC coupling and a reduced DC gain to get the amplifier tolerant to scattering Idss and changing supply. Another point is a trimmer / selected resistor.

[TimFox]

A good bias-set circuit for the input to an audio amplifier is a bypassed resistor from source to ground, and a gate-bias resistor to a positive voltage, with the input capacitively coupled to the gate.
You can do this on a breadboard, or a worst-case calculation using the JFET data sheet.

[Zero999]

This is going to be the input for an audio amp. It will take in a guitar input. I was going to hold Vgs at around -1 V and the amp will run off a 9V battery. Simulating this in spice I was wondering what would happen as the battery goes down to about 6V and started getting a drop in out output and non-linearity from the FET. I can compensate but with a drop in Av. I'm trying to peg this in the sweet spot so I can use it with the full range of 6-9.6V but I don't know how to work around the tolerances being that the JFET I build with may not act the same as the model. I was curious what your thoughts were on going about the design and how you would think about the problem. Thank you.

Guitar amplifiers normally use source follower configuration for high input impedance.

How about adding regulating the current, so it will work the same irrespective of the supply voltage?

[TimFox]

For a source follower, having a current source in the source lead with a gate resistor to a positive voltage is nearly foolproof, so long as the current source is less than Idss.
However, an unbypassed resistor in the source lead is probably almost as good, and cheaper.

[exe]

This is going to be the input for an audio amp. It will take in a guitar input.

Just use an opamp. You can use a jfet one, if jfet is absolutely required.

[Zero999]

This is going to be the input for an audio amp. It will take in a guitar input.

Just use an opamp. You can use a jfet one, if jfet is absolutely required.

Or even a BJT with bootstrapping for a high input impedance.

[exe]

JFYI I'm eager to share my take on jfets. According to textbooks, they always had very poor tolerances, though a bit more predictable in the same batch. So, manual binning is necessary if single jfets were used. Or a trimpot to set the operating point for a specific device.

However, for amplifier differential input stage it's much more convenient to just use a monolithic matched pair. Those are largely extinct, and cost a fortune nowadays. However, matched jfets are used inside jfet opamps, and that's preferred way to have jfet input, though a discrete solution could in certain cases give more bandwith, as I understand.

Why I asked about application. Jfets could also be used as current sources. But, in my opinion, depletion fets are better because they were a bit easier to source and gave me better tempco and compliance. This topic is covered in detail in "current sources and voltage references" book.

So, I'm not sure why one would use discrete jfets nowadays except some very rare cases. Their loose parameters make it hard to use in mass production.

[David Hess]

Why is it so hard or expensive after all these years, that they don't make common JFET's or mosfet's more matched ? Who would complain if they were more matched ??

The fabrication process just does not allow that kind of control of certain parameters.  Bipolar transistor hfe has considerable variation as well.

[T3sl4co1l]

Specifically diffusion, you cook it a little too long or short, or hot or cold, and the charge balance goes out just a little bit.  But you're pinching off a very narrow channel, that merely a few volts is supposed to be enough to modulate, so it takes very little charge indeed.  I'm guessing the nonuniformity is as much by process as within a die, because the chemical and thermal environments just can't be controlled that evenly.

Do they ever make JFETs with epitaxy?  They ought to be able to, and could do so with better precision (see SuperJunction technology, it requires depletion of whole pillars of alternating P/N) than diffusion.  I guess I wouldn't be surprised if SiC JFETs are; in Si, they're really only important for signal purposes anymore, for which traditional methods suffice.

Also there are the bad old types, and newer ones like CPH3910 which are tighter, not to mention tightly optimized in gm/C.

Tim

[donlisms]

Just anecdotally, Alembic Series instruments used a matched dual JFET as current source and follower as the front end to an NE5534 in the days before there were JFET opamps. I like mine a lot, but if one of them ever breaks, I'll switch to modern opamps. I'm *certain* Ron Wicker-sham would have used them back in the day if such parts had been available when he was designing that stuff.

[TimFox]

The Linear Systems company (not to be confused with Linear Technologies) still makes dual matched JFETs.
https://www.linearsystems.com/product.html?category=jfetAD
Even with a spread of Idss from part to part, the two halves of the dual part match each other.

[exe]

Yeah, even TI makes ones. Those are JFE2140DR and they are the cheapest matched jfets I found, yet they cost over 4euro+VAT on mouser.

On the other hand, I've read elsewhere that precision matching is not needed. Afaik, one of designers of DMM6500 said that super-low offset voltage between inputs is not required. So, I guess, for audio precise matching is not needed. So, one may get away with much cheaper device, like MCH5908H which costs less than $1. Or may be not, I'm not an expert).

BTW, what are the advantages of using a discrete frontend comparing to an opamp?

[Kleinstein]

There are also not especially matched JFETs that should normally still be resonable matched.
A cheap one is SK2145  made for audio use.  With the source pins connected there is no good way to compensate a large offset though. A small offset can be compensated / trimmed at the drain side.

One could still use the dual JFET for a source follower, using one of them for a constant current sink: so use one of the JFETs with source and drain swapped ( most JFETs are symmetric and D and S can be swapped).

The discrete JFET amplifer can get  lower noise levels (epsecially when single ended and not differential) and operation from a low voltage (e.g. 3 V) and possibly power via the signal cable (e.g. the amplifier in many microphones is just a JFET).
Modern OP-amps however can replace the discrete JFETs in many cases.  One use is still AM input stages (e.g. the SK3557 and similar) for better noise performance, but AM radio is a bit on the decline.

[David Hess]

Just anecdotally, Alembic Series instruments used a matched dual JFET as current source and follower as the front end to an NE5534 in the days before there were JFET opamps. I like mine a lot, but if one of them ever breaks, I'll switch to modern opamps. I'm *certain* Ron Wicker-sham would have used them back in the day if such parts had been available when he was designing that stuff.

That depends on the application.  Monolithic duals have better matching but parasitic coupling between the transistors limits performance in some applications.

[John Coloccia]

FWIW, I rarely use a JFET (opamp or otherwise) as my input buffer for a guitar circuit. After much experimentation over many years, I always come back to a good, bipolar audio opamp, often the LM833. With a well selected cap and resistor to AC couple/filter/set input impedance, this setup to me always just "feels" very good as a guitarist. They also tend to be a lot less noisy than JFET opamps. I have designs where I stack 4 opamps in series. Try that with a TL072 and any sort of gain.   Hiiiiiissssssssssssssssssss.

I encourage you to experiment with the JFET path. You might fall in love. I always found them to be fiddly and underwhelming except for certain applications.

BTW, it's not uncommon to use a trimmer to set your bias, or to sort your JFETS so you know what you have.

[Zero999]

FWIW, I rarely use a JFET (opamp or otherwise) as my input buffer for a guitar circuit. After much experimentation over many years, I always come back to a good, bipolar audio opamp, often the LM833. With a well selected cap and resistor to AC couple/filter/set input impedance, this setup to me always just "feels" very good as a guitarist. They also tend to be a lot less noisy than JFET opamps. I have designs where I stack 4 opamps in series. Try that with a TL072 and any sort of gain.   Hiiiiiissssssssssssssssssss.

I encourage you to experiment with the JFET path. You might fall in love. I always found them to be fiddly and underwhelming except for certain applications.

BTW, it's not uncommon to use a trimmer to set your bias, or to sort your JFETS so you know what you have.

The TL072 is much cheaper than the LM833 so it's not really a fair comparison. The NE5532 is a much better op-amp to compare it with.

J-FETs have more voltage noise, less current noise and BJTs more current noise and less voltage noise.

What's better depends on the source impedance. If I remember rightly, guitar pick-ups are highly inductive, so the impedance increases with frequency.

[MathWizard]

I have to go over all those FET equations and start using JFETs more. I need to buy a big bag of cheap ones.

Years ago from ebay I bought some BJT assortment, and most of them, I just never use, they never seemed that different. I should read their datasheets a bit harder. Like whats the real difference between a 2N3904 for audio, and some S90xx series that's in a lot of radio/audio stuff.

I own multiple books on semi-conductors/transistors, and classical and quantum electrodynamics. But it's page after page of equations. And I do love equations, but I hope I live long enough to learn a 1/4 of it.

I just learned again, the other day that Faraday was at the center of calulating the # of atoms in a Mole. Electro depositing different metals I believe, and monitoring the total current over time. And comparing different elements, and I'll have to look that up.

[exe]

I need to buy a big bag of cheap ones.

Please post a link if you find one.

[Sengcid]

Discrete JFETs of the same type are better matched than datasheets might suggest. IDSS and VGS(off) are correlated. Probably OTT for novices are a couple of monographs on JFET theory and low noise pre-amps plus more via the link.

Part 5 monographs 3 and 4

https://drive.google.com/folderview?id=1TbomXeoBbIe-IVADaOOmyyLH9le-3wAt

[David Hess]

BTW, it's not uncommon to use a trimmer to set your bias, or to sort your JFETS so you know what you have.

In single ended AC coupled circuits, the bias can be set by controlling the drain current, producing as specific transconductance, and ignoring Vgs.