Measuring J201s with a new toy - Are Tayda's SMD J201s awful?

[schrectacular]

You're really going to town with this, Rob! Luckily I left the pedals that have these in them open and on my desk. I'll take some more measurements tomorrow with some larger resistors and see what we come up with. May also be a good excuse to mount the rest of the SMD chips I have.

[Rob Strand]

You're really going to town with this, Rob! Luckily I left the pedals that have these in them open and on my desk. I'll take some more measurements tomorrow with some larger resistors and see what we come up with. May also be a good excuse to mount the rest of the SMD chips I have.

I've done similar things in the past so I know where to look for trouble.

Based on the same J201 JFET example and same +/- 0.01 random deviations the added resistor which gives the most accurate parameters (actually least variance) is about 1.5k.    With 1.5k the deviations in the estimates about half what you get with 680 ohm.  So very worthwhile for a simple tweak.

I can't say it's going to work out that way for all JFETs but it's a good stab just the same.

The hanging question is for *matching* JFETs for a phaser is it better to:
-  match the "direct" VGS measurement without the resistor, or,
- match the VGS value with the added resistor, or,
- match the VP parameter from the whole calculation process. 
My guess is as far as matching numbers goes, probably the second one:   match VGS value with the added resistor.   While it's the best one it still might not be any good for matching - that's extremely likely since even 10k is too small.
Keep in mind the actual VP value is always best estimated from the last method since it's the only estimate which attempts to be VP.

[schrectacular]

Mind giving a brief explanation about why the variance gets worse for a a higher resistance? Why is the minima at 1.5k? Maybe I understand - as we add more resistance we are lowering our tested values. As we get lower results from the tests we are getting less accurate measurements; 2 to 3 is a change of 50% but the same nominal value higher up is a much lower percentage and therefore less error. Kind of like how I'm getting closer in age to my father every year if you consider the year as a percentage of life lived rather than as a fixed amount of time.

So I suppose 1.5k is far enough away from 680 to give us a good estimate of our slope, but also not too low that our error bars on the measurement get too high. That's my intuition anyway.

I'm about to test again with 1.5k, will post results soon.

[schrectacular]

Updated with 1.5kΩ results:

Id (direct)	Vg (direct)	Id (680Ω)	Vg (680Ω)	Id (1.5kΩ)	Vg (1.5kΩ)
.41mA	.29V	.28mA	.20V	.21mA	.15V
.38mA	.27V	.27mA	.19V	.20mA	.14V
.37mA	.26V	.26mA	.18V	.20mA	.14V
.35mA	.24V	.25mA	.17V	.18mA	.13V
.33mA	.23V	.23mA	.16V	.17mA	.12V

[Rob Strand]

Mind giving a brief explanation about why the variance gets worse for a a higher resistance? Why is the minima at 1.5k? Maybe I understand - as we add more resistance we are lowering our tested values. As we get lower results from the tests we are getting less accurate measurements; 2 to 3 is a change of 50% but the same nominal value higher up is a much lower percentage and therefore less error. Kind of like how I'm getting closer in age to my father every year if you consider the year as a percentage of life lived rather than as a fixed amount of time.

So I suppose 1.5k is far enough away from 680 to give us a good estimate of our slope, but also not too low that our error bars on the measurement get too high. That's my intuition anyway.

Yes, exactly that on both accounts.   You can see the smaller values for 1.5k in your new measurements.

The sources of error are:
- rounding error due to the display of two digits
- measurement error of the unit (even if all digits were reported the values would be off).
- Also the assumed values of the added resistor and resistor inside the unit.

It just so happens the 1.5k finds the best balance.    I might add it's only true for the example J201 JFET I used.  If the JFET had larger VP and/or IDSS then the GM328A would report the values with more digits in relative terms and that would probably make the best value for the added resistor higher.

I used a -0.01 to 0.01 deviation which is a little aggressive.  If the unit reports rounded values then it should be -0.005 to 0.005.  However that doesn't allow for measurement errors.    I tried -0.007 to 0.007 and the best resistor went upto 1.8k, however the estimation errors for 1.8k weren't much better than 1.5k.

Here's the results for the two cases.

I've added the Vgs2 value so we can see the voltage on the source is higher than the reported value on the unit and that the voltage isn't so high to make us think the unit won't measure correctly with the resistor added.

Rs_add    680   
JFET    Vgs2 [V]  VP [V]  IDSS [mA]
J201    0.3898    0.86    0.93
J201    0.3730    0.93    0.76
J201    0.3562    0.86    0.76
J201    0.3395    0.88    0.66
J201    0.3159    0.75    0.69

Rs_add    1500   
JFET    Vgs2 [V]  VP [V]  IDSS [mA]
J201    0.4650    0.91    0.89
J201    0.4400    0.89    0.78
J201    0.4400    0.94    0.71
J201    0.4000    0.81    0.71
J201    0.3750    0.74    0.69


You can see there's good general agreement which means the added resistor method works.   
The worst error between the two sets is about 9% which is telling us how far we can trust the numbers.
We know the second set are probably more accurate.

If you really wanted to push the envelope you could get a few estimates with "good" resistor values, say 1k5, 1k6, 1k8, 2k0 (whatever) and average the results.   Its a bit like pulling teeth but multiple measurements is a way of  getting around noisy data.

[schrectacular]

This is super helpful, Rob. It's also nice to set expectations on what this thing can do. I suppose that we'd expect it to be more accurate measuring resistors and capacitors.

[PRR]

Igss and Vgs(off) are really trivial to measure directly with batteries and a meter. (And the JFET data-sheet!!) I appreciate exterminating fruit-flies with a sledge hammer but that's not the only way.

[Rob Strand]

This is super helpful, Rob. It's also nice to set expectations on what this thing can do. I suppose that we'd expect it to be more accurate measuring resistors and capacitors.

Igss and Vgs(off) are really trivial to measure directly with batteries and a meter. (And the JFET data-sheet!!) I appreciate exterminating fruit-flies with a sledge hammer but that's not the only way.

I think idea of getting better numbers out of the GM328A unit is fine for biasing a JFET stage but it's not going to be good enough for matching JFETs for a phaser.

I agree building with PRR that a purpose-built JFET test jig is going to produce better numbers.

The way we got here is the initial assumption was the GM328A produced numbers like a jig.   Clearly it doesn't.   So we pushed our luck to improve the results.   It turns out those results still have limitations.  So we end-up having to fall back on purpose built jigs.

Even those purpose built jigs can put some doubt in your mind about what the real parameters are.  If you use different source resistors on the test jig the measured Vgs values don't agree with the true VP parameter.   You can apply corrections like the table here,

https://www.diystompboxes.com/smfforum/index.php?topic=128021.msg1231427#msg1231427

but ieven after the corrections you will find the numbers from different source resistor values don't always agree.   For matching it's probably OK but as far as finding the actual parameters there's uncertainty.

IDSS measurements have problems where the device gets hot.    Most JFETs have a built-in source resistance (around 20 ohms) which causes the IDSS measurement to be off from the real value.

See fig 5,
https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.329.8164&rep=rep1&type=pdf

The article also mentions how Lambda affects the results (fig 2).

[PRR]

> .... put some doubt in your mind about what the real parameters are.  If you use different source resistors on the test jig the measured Vgs values don't agree with the true VP ...
> ...even after the corrections you will find the numbers from different source resistor values don't always agree....
> IDSS measurements have problems where the device gets hot.

We can always have doubt. And we can address it. Read the specs! Vgs(off) is commercially specified at some low but non-zero current.

So ideally you want a current source. (For J201, a VERY small current source.)

But for selecting "similar" parts it is sufficient to get in the nominal range. For J201 and assuming you expect a large group around 1V, then 1V/10nA makes 100Meg(!) source resistor.

Yes, this seems to eliminate common 10Meg meters. But 90Meg in series gives readings like 0.1V which is a good point on a 199mV digimeter.

Any temp-sensitive reading, you define "cold" or "hot". My furnace measures hot. Transistors may want pulse measurements. Peak-catchers were always cheap; today digi-scopes are also. However J201 Idss maxes at 1mA, which at 20V test condition is 20mW, which is 6.1 deg C rise. OTOH J111 goes to 20mA, at its 15V test case is 300mW, which is awful hot. "Pulse test: pulse width <300uS, duty cycle <2%".

But we are practical men/women/others. We don't need an abstract match, we need parts which, in a specific job, will play well together. For the classic FET-input opamp, we build the opamp and swap for a null. For phasers, we pick two resistance values near the extremes of the sweep, pick a series resistor which in a resistor divider will give say 0.33 and 0.66 of a small DC voltage. In this case we don't have 100Meg impedance or many-milliWatts of heat.

[italianguy63]

I went through this about 5 years ago.

I ended up building a socketed test rig... base on R.G.'s circuit, and tweaked by Runoff Groove.

It worked great.. still does.  I don't need it anymore.  I am going to make the rig available in the For Sale section.

Rock on!!
MC