More FET noise measurements (for EUVL)

diyAudio Chief Moderator
Joined 2002
Paid Member
...I decided to build my own test jig to get a better picture of the noise spectrum. There are two plots here...

Hello Scott, thanks for the graphs. Wouldn't a 2SK170V compare fairer since more IDSS means more Gm while the other two Toshiba fets are already strong? Or is it still apples to apples due to other reasons? I.e. you used low current for instance. Also did you figure out that HF noise rise tendency's root cause by now? Can it be the higher capacitance fets resonating with something in the set up? K170's sub 1kHz rise, is it due to generation-recombination noise or it can be 1/f as well since those two are normally room mates?:)
 
Hello Scott, thanks for the graphs. Wouldn't a 2SK170V compare fairer since more IDSS means more Gm while the other two Toshiba fets are already strong? Or is it still apples to apples due to other reasons? I.e. you used low current for instance. Also did you figure out that HF noise rise tendency's root cause by now? Can it be the higher capacitance fets resonating with something in the set up? K170's sub 1kHz rise, is it due to generation-recombination noise or it can be 1/f as well since those two are normally room mates?:)

If you just want to max out gm you might make different choices. One thing I missed is that you have to watch out for Vgs going positive on large signal swings, this might also lead you to choose a higher Idss. The noise flat top's for a while this is the GR signature. The noise is random but the average pulse witdth has a Lorentzian distribution around a certain tau. This gives the characteristic shape more like an RC (1/tau) break point. The noise lower becomes 1/f again. I have not found a source for the high frequency noise.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
If you just want to max out gm you might make different choices. One thing I missed is that you have to watch out for Vgs going positive on large signal swings, this might also lead you to choose a higher Idss. The noise flat top's for a while this is the GR signature. The noise is random but the average pulse witdth has a Lorentzian distribution around a certain tau. This gives the characteristic shape more like an RC (1/tau) break point. The noise lower becomes 1/f again. I have not found a source for the high frequency noise.

Cheers Scott. Thanks for the extra info. Is it likely the K170 is GR infested more than its measured siblings due to a narrower channel, or due to subtleties of a possibly different wafer process?
 
About Salas Q.:
While FET transconductance Yfs increases with Id, it is actually a little lower
for the higher Idss samples (compared at the same Id). Check the Yfs=ƒ(Id)
for the different Idss (parameter) chart.
What most FET manufacturers will not show you in their d/s, is that there is
a very narrow minimum of spectral En versus Id at low freq, while En at 1KHz
will have a wide valley for minimum versus Id. That Id for the low freq minimum
will not coincide with the Id for 1KHz minimum. Clearly an issue for (phono/tape)
equalizers - there will be the optimal current for the best S/N ratio for the whole
freq range.
The rise of noise at high freq. caused by Cgd: the thermal (noise component)
current from the channel through this cap. gets in the gate, produces additional
noise component at the assosiated gate impedance (Zsource, Rbias, etc.) and gets amplified in a normal manner. Another cause for high freq. En increase is the drop of transconductance with the freq.
 
Last edited:
diyAudio Chief Moderator
Joined 2002
Paid Member
Thanks.

So my gut feeling that it had to do with the higher capacitance of the wider channel fets possibly held some water after all in a way?

True, the gm relative to Idss as a ratio is following a diminishing returns trend. Thanks for reminding that. For the K Toshiba fets I have seen that Id 4-5mA Vds 4-5V is a good operating range for better noise balance in the lows when the signal comes from low Rg sources like MC carts. Any different experience on that?

Alas, low en fets are generally high gm/high Idss and degenerating them adds en from source resistors, invariably leading to shunting many of them with relatively high individual Rs so to hold current back but bring down the source resistance total also. Gives slack to match picking them in quads too, since it won't let different enough ones current hog?
 
I found this excellent reference, but fairly math intesive analysis for low frequency noise (he actually used a 2SK369 as one reference part).
Ultra-Low-Noise High-Input Impedance Amplifier for Low-Frequency Measurement Applications, 2008, Felix A. Levinzon, Member, IEEE.
I don't know if a free download exists. He found that induced gate current noise was absent up to 110KHz. If you want a tedious accounting of every noise source this is the place to go. He got amazing low frequency results and made some interesting observations. There was essentially no low frequency noise in the audio band for his sample.

Just to put numbers on this I attached the Yfs vs. Id curve from the 2SK170 ds. There really is not a meaningful variation over the entire range of Idss i.e. every 2SK170 has 20mS+-~2% at 2mA. This was studied by Columbia University (2SK369) here, Characteristics of 2sk369 Junction Field Efect Transistor for Run 2 Calorimeter Preamplier. This is available as a free download. 40 samples of each V and GR grade were tested with mean gm of 44.56 and 45.59 respectively at 6mA. The difference gets buried in the standard deviation and the noise impact is almost unmeasurable.
 

Attachments

  • k170.JPG
    k170.JPG
    11.9 KB · Views: 1,692
Last edited:
diyAudio Chief Moderator
Joined 2002
Paid Member
Thanks again for the help Scott. Thus, your 2mA test was apples to apples. To clarify with a simple example, say we have K369BL and 369V, what will be best to use for lowest noise, and running them near full Idss will be of benefit?

P.S. Found the Calorimiter. Gives the K369 0.46nVrtHz. Nice.
 
Thanks again for the help Scott. Thus, your 2mA test was apples to apples. To clarify with a simple example, say we have K369BL and 369V, what will be best to use for lowest noise, and running them near full Idss will be of benefit?

P.S. Found the Calorimiter. Gives the K369 0.46nVrtHz. Nice.

I would pick a current like they did that encompasses both, remember you only get lower noise as the 4th root of current and ask for thermal problems.

Using .6/gm as the noise resistance at .045mS you get 0.47nV, pretty good agreement!

Doubling the current to 12mA only gets you 0.39nV.
 
  • Like
Reactions: 1 user
That whould help to avoid oscilation but the problem of unlinear input capacitance is not solved that way. (One way is conventional cascoding). Reminds me a bit to avoid TIM by an input filter that makes sure that transients do not drive a slow amplifier into slew rate limiting. You loose the information in that pulse too that way but hey, i am 53 so i do not hear anything above 16kHz anyway.
 
diyAudio Chief Moderator
Joined 2002
Paid Member
Is the high gm likely to stay prone picking out of band noise, or taming the Miller multiplied capacitance by cascoding and/or degeneration should be enough measures? We don't want our mobile phone to startle us through the speakers.