QEX: Gary Kaatz Observations

Gary Kaatz (W9TD) wrote to the ARRL editor with this question. The QEX editor forwarded it to me.

Gary Kaatz (W9TD)

The author assumes that all diodes will perform like their published data. This is not so, semiconductor manufacturing is not that precise. Matching between similar devices on the same IC will be very good but will vary from IC to IC or diode in this case. The published data and SPICE model are only “typical” and any physical device may vary from “typical”. In fact, when doing IC design you would use corner case models, worst case, fast, slow, etc. to insure that the design would have high yield at manufacturing. The author correctly shows how one specific physical diode can be modeled from I-V data but that cannot be construed to represent “any” other diode of the same type. It was purely coincidental that the measured diode conformed closely to the published performance.

This is an EXCELLENT observation that is well-noted in the industry. There were two very important observations that Gary submitted in his question.

I should mention that the QEX article is merely a sub-set of a much larger article that I plan to publish with either the IEEE or the SAE. While the QEX article addresses the ambient temperature ill-congruence in passing, the larger article addresses this aspect of the SPICE diode as an explicit premise. In the article, I will propose a simple method to circumvent this issue without changing the underlying model in any way. Extensive lab measurements were made.

I purchased one hundred 1N914 diodes and made rudimentary comparative measurements on ten of them. “Rudimentary” means that I measured performance at only one temperature–the ice bath. I would have to look at my lab notes but the measurements were impressively close. I was not expecting that. I then picked one of the ten to serve as the sample for the QEX article.

I spent my entire Christmas break of 2020 taking thermal measurements and drafting text. Well… almost all of the break. It was more like putting in workdays.

It should, of course, be noted that my sample of one hundred diodes came from one production. Another production could have varying results but performance within this lot was “right on the money.”

Gary’s second observation was very much astute and has come up many times over the years in the engineering community. The diode (and most other electronic components) specifications typically run all over the map! Probably most engineers do not realize this but it is one of those things where the hand is quicker than the eye. The specifications that engineers look at are consistent but where the hand is quicker than the eye is usually seen where curves are not presented but only simple tables.

Consider the example shown on the right obtained from www.Radwell.com. The elephant in the room is line-item number 3 showing 100 Volts at 200 mA. Something is just simply wrong there, likely somebody was not interested and not paying attention to anything but putting something up for sale. A bean-counter needed something looking official and this did the trick. We are talking about 20 Watts!

But the relevant item in the illustration is 1 Volt drop at 10 mA which is unrealistic.

But this is not unique to diodes. We see this routinely with Operational Amplifiers, Bipolar Junction Transistors, and the list goes on.

Gary’s observation is very well taken. My hat goes off to Gary.

Gary is also very much correct in naming differing performance from aging and component lot tolerances. Sadly, even the characterization houses such as Mentor Graphics and Synopsys vary for worst-case Monte Carlo analyses by tolerancing characterization arguments. Gary is correct (in my opinion) in that worst-case characterization must be at a component level and not argument.

The premise of the QEX article is with regard to duplicating published performance. The industry has standardized for the 1N914 diode publishing identical curves. That is all that we have to go on so characterization will naturally follow that.