Back to back Jfets

I don't mean to put words in Kevin's mouth, but I'd guess he is referring to something similar to NXP's BF862 datasheet figure 3 where it shows that Idss and Vp are related. If you connect g-s and measure Idss, the graph gives Vp. Also, according to the data sheet, Vp can range from -.3 to -1.2V. So, putting it all together, Idss can range from 0 (silly) to 32mA.

Am I incorrectly interpreting the data sheet?

JohnS

About 2/3 of my recent ones have op amps forcing them to run at IDSS, so everybody's bias conditions are the same. The other ones run at 10-12 mA, which I agree is a good choice. Usually the customer wants to use some gross huge PD at low current, and sometimes that's even the right answer. ;)

My most recent one was 1600 pF and 20 nA full scale. The interesting thing with that one was that it was a battery powered instrument, so the measurement had to be made within about 0.5 seconds of turning the thing on. Without the op amp, the thermal offset voltage transient in the bootstrap was enough to cause serious measurement error due to the differentiating action of the capacitance. Usually that's not a worry.

I usually use some extra circuitry to get the gain closer to 1.000. Sometimes the actual measured gain is better than 0.999, which is hard to do with a resistor in the source. The source impedance is around 25 ohms for one FET, so getting to 0.999 would need a bias resistor of 25k to -250V.

Some shunt feedback (like the PNP wraparound trick for 78xx regulators) will reduce the source impedance a lot, so a resistor can work OK, or else a BJT current source. It's important that the source bias current be sub-Poissonian, or it'll dominate the noise. The nice thing about shunt feedback is that the FET runs with a bit of voltage gain, so that the loop suppresses the wraparound's noise.

You've mentioned that before--I've been meaning to try it. I have a few hundred pHEMTs of various kinds in stock.

Cheers

Phil Hobbs

They're related from the device physics and modelling POVs, of course, but I'm talking about guaranteed data sheet parameters. Kevin was claiming that some devices wouldn't pull their weight, and that's just not so unless you're pretty far from V_GS=0. (Maybe his circuits are all current-starved, or something.)

If you look at Fig 7, you'll see that the max, min, and typical curves all have about the same slope at VGS=0, i.e. they have very similar transconductances at I_DSS. I actually sucked in the curves using Datathief, and the transconductances are (top to bottom) 39 mS, 39 mS, and 32 mS. (So their lower-limit curve is actually slightly out of spec.)

From crunching the data from the curves in Fig 8, it turns out that the transconductance is fractionally higher at V_DS ~ 2.5V. That's where I usually like to run them, because it reduces both power dissipation and gate leakage.

It's the transconductance that you mostly care about, because that's what sets all the stage gains, bandwidths, and governs the noise statistics. Each FET contributes a noise current of e_N * g_M, and the results add RMS fashion, so if the transconductances are similar, the noise voltage goes down as 1/sqrt(N).

Cheers

Phil Hobbs

Now I'm really lost. Are you indicating that typical values are guaranteed parameters?

Fig 7 is typical data, it says beneath the graph. So the max and min are typical max and min.

No, I'm not saying the whole curves are guaranteed. What's guaranteed is that 35 mS minimum transconductance at 0V GS, and that's all I need to establish the point at issue.

Cheers

Phil Hobbs

On a sunny day (Sun, 16 Nov 2014 12:53:36 -0800 (PST)) it happened Phil Hobbs wrote in :

Phil, I 100% agree, the gain adds up. This is why I had so many in parallel in the 'lighting a LED with a candle' project.

For RF if you have more in parallel, the noise goes down as you pointed out. We should always see things in the perspective of the application. Slimulations are often a distraction from reality. In fact those are ONLY useful if you already know what you are doing, know what to look for, and HAVE the real hardware to check against. That is my expericence, and again very recent too with my QAM modulator driver. Its great to see filter curves... But I still measure those before and after the slimulations.