I haven't updated for a long time (since March 15). Now I updated and found that Linear Systems' JFET models have been changed. From weakly wrong noise models they replaced them with ultra-noisy models! I have the correct noise models from this company have the suffix _n. I customized the models to match the datasheet. I had to add the suffix so that my correct models would not be replaced by the wrong ones when updating, as I had already done several times.

Bordodynov.

I always do a backup of my stuff before updating, but I haven't updated in a long time, because wouldn't of they done it right the first time?

My files are tools and one should take care of them.

开云体育

I have written about this several times in this group, starting with #138735. I have also complained to ADI about it a number of times. All to no avail.

It seems that the general problem with the affected models is that the Kf parameter in standard.jft is 1e6 times bigger than the same model available directly from Linear Systems. It looks like a transcription error made by someone at ADI, or whoever they have outsourced the standard libraries to.

Can you share the "good" jfet models or point me to where I can find them?

Thanks?

Yes, this is a "known" problem, and it's been known for a rather long time.? It was before Tony's message in 2022.? I don't recall exactly when it was first discussed in this group. but it was quite a while ago.? LTspice used the defective JFET models from Linear Systems (Linear Integrated Systems) starting pre- or early-COVID.? It was not only updates where you got the bad models, you got them even just by installing LTspice XVII.

? ? "It looks like a transcription error made by someone at ADI, ..."

Actually, I believe the error was made by someone at Linear Systems.? See below.

ADI was not innocent.? They accepted the bad models without questioning them.

Linear Systems has (good) SPICE models of most of their JFETs.? You can find links to them on the webpage for each JFET, and those models are OK.? But at some point, someone at Linear Systems merged all the JFET SPICE models together into one file, and that was when the mistake happened.? Then they shipped the merged file off to ADI for inclusion in LTspice.

The biggest problem is the Kf noise parameter.? I'm guessing that the person who merged them had not seen numbers that small before (e-18).? Probably through recklessness, he altered them.? e.g., "Kf=45.61E-18" morphed into "Kf=45610f" which made it one million times larger!? Maybe he thought he was being "clever" by changing from scientific notation into a letter multiplier, but he went the wrong way, moving the decimal point in the wrong direction!? (As you can tell, I am only guessing.)

ADI could have made a thorough check, testing every JFET model in the file on a handful of test jigs, but that's a lot of work and hardly anyone does .NOISE analysis anyway.? I'm guessing ADI assumed the Linear Systems models must be OK, so they were incorporated into LTspice without question.

? ? "Can you share the "good" jfet models or point me to where I can find them?"

The "good" Linear Systems JFET models are here:??Go to Linear Integrated Systems's website??and look up a JFET in the product selector.? Go to that JFET's page.? Most of their part webpages have a link to THAT one part's SPICE model.? These individual SPICE models are OK.? (Yes, there are questions about some of the other parameters in some of the models, but at least the Kf parameter is "OK".)

If you find a link on Linear Systems's website to download all their JFET SPICE models in one file, stay away from that one!? It is the "smoking gun".

Bordodynov has many SPICE models where he has tweaked them to fit measurements.? Trust his models.

Andy

开云体育