Close

So you were told to use gold finish on your microwave projects...

A project log for Microwave stuff - what to do and what to avoid

A project for storing facts, general rules, not very well known information only found in scientific books/articles and forgotten knowledge.

ms-bossMS-BOSS 02/04/2020 at 16:378 Comments

Usually, most people who make some microwave stuff think that the best finish they can apply to their boards is gold. It looks good, doesn't change over time and since most HF (high frequency) energy flows through the surface due to the surface effect, it's best to have some protection on the copper traces. In this case, gold.

Some of these rules are right, some are OK and some are simply off.

Available finishes

First of all, let's list the usual finishes available from almost all manufacturers, even the Chinese ones:

And now the less usual ones, or at least less used ones:

Solderability

Now, let's mention the solderability of the more usual finishes, listed from best to worst:

  1. HAL
  2. Leadless HAL
  3. ENIG
  4. OSP
  5. Bare copper
  6. Coverlay
  7. Immersion tinning
  8. Hard gold

That may sound strange, doesn't it? You were probably expecting the gold finishes on the top places. And that's the first common myth. Gold isn't the best solderable finish. HAL is usually the best one.

Repeatability

This means repeatability of thickness, surface roughness, resistivity and how precisely it is controlled over the board:

  1. Bare copper
  2. Coverlay / OSP
  3. ENIG/EPIG/ENEPIG
  4. Immersion tinning
  5. Hard gold
  6. HAL / leadless HAL

Again, gold isn't the best one. Bare copper and its "covered" variants have the lowest tolerance, roughness and best repeatability. ENIG and its variants suffer from tolerances of the thick nickel layer which usually has rounded edges. For example, let's show an image. Its source is IEEE-Xplore. If you cannot open this because you do not have the magical paywall access, you can find this article on Z Library, just try to find its name. I won't be inserting link to Z Library, because I do not know how prudent HackaDay is when it comes to links to illegal sources of science...

As you can see, the trace is nowhere near the ideal rectangular shape everyone wants to see. If you look closely on the Cu trace, you can see the effects of etching. And the Au/Ni part makes the Cu trace more rounded and thicker than it should be, which is worth noting.

HAL isn't very well controlled process in this point of view. The lead dissolves partially the copper beneath and when blown off using the hot air, it is not very flat. Look at these two pictures.

As you can see, it is not flat at all. That's the main problem with HAL - nonpredictability.

HF performance

Now, let's list those finishes in order of suitability for microwave stuff:

  1. Bare copper
  2. Coverlay / OSP
  3. HAL / leadless HAL
  4. EPIG
  5. Immersion tinning
  6. ENIG/ENEPIG

OK, this is starting to be confusing. Why is ENIG never on the top? Why not even in the top 3? The answer is nickel. Maybe that was too short answer. Let's just state that copper or its covered variants have very low losses on microwaves if the copper is smooth. The reason is that rough surface increases the length of the trace virtually and that increases the losses. HAL is useable, but its impedance may slightly vary due toits non-flat surface. However, it is more or less smooth. EPIG is a bit worse, because the surface is "grainy", but OK. ENIG and ENEPIG are not very good because of its nickel layer and surface roughness.

Why is nickel such a b*tch?

Nickel is ferromagnetic. That means it does strange and not very pleasant stuff with microwaves. And what is even worse, it is not even consistent at doing so. The main problem is that its permeability drops quite fast between 2 and 3 GHz from 5.7 to something about 1.4. And it even shows off resonance in this frequency range. There is nothing you can make about this, this is a material property. Of course, you could make each trace shaped like a filter which would make the losses flat, but that would be difficult, take a lot of space and would increase losses over the whole used band.

Let's look at the losses of ENIG trace, again taken from the same article as before.

There is the resonance. As you can see, the steepness of the loss rise lowers above 3 GHz. That's because above this frequency, the nickel layer stops behaving as a proper ferromagnetic material.
The same happens for group delay which is where the researchers smelt something stinky about ENIG.
A resonance! And quite apparently visible. They tried to model the ENIG trace using Debye dispersion model which proved to be insufficient and so the used a mixed model constituting of the Debye dispersion model and Landau-Lifschitz's permability dispersion model. This way, they were able to model the resonance and both the losses and group delay. If they are speaking truth, they are the first ones to have ever done so. Let's look at the complex permeability of nickel in ENIG. Sorry for the utterly ugly Excel graph, it is taken from the article as-is.

As you can see, for a while, the permeability even drops under zero for a quite large frequency range. And since the permeability is mostly imaginary in this range, you can expect quite a lot of strange things happening, like a lot of losses and resonance in the losses.

As you can see, ENIG finished traces can have losses near 0.5 dB/cm at 2.5 GHz (sorry, I am not going to give you values in dB/inch, because inches are not measurement unit, but a heresy which has to be stopped).
That means if you have about 2 cm distance between the u.fl connector on your WiFi modem in your notebook and the transceiver, you are losing 1 dB of power. That may not sound like much, but this means 20% of power. That means about 11% loss in useable distance of the WiFi. Now, it doesn't sound like such a little problem, eh?
Of course, a lot of the losses is caused by the substrate losses, finite conductivity of copper, gold and nickel. However, the nickel layer causes a lot of trouble. Let's compare it with another research article. This time it will be this article. Let's first have a look at the surface finish in cross cut. As you can see, the top of the trace can be quite flat, however the bottom surface is quite rough. And this causes a lot of losses, too.

Let's look at losses of differential pairs. The article mentions even single traces, but the losses are not as large as when using diff pairs. HFSS means simulated results.

The ENIG losses are more than twice as large when compared to lead HAL, which means that if you make two reference traces made with both the HAL and ENIG and then another, longer one whoch would have twice as much loss with the HAL, the ENIG one would already have about five times larger losses. If you think about these numbers in these real-world scenarios, it looks quite scary. And even then, most microwave engineers use ENIG daily, because everyone does.

As you can see, it looks that OSP isn't that bad as you would probably expect.

There is another interesting article which has data up to 67 GHz. It even list one more not very usual finish, the ISIG or Immersion Silver Immersion Gold which looks to be really good. Does anyone know of any manufacturer which would make this finish?

That ISIG looks real good. As you can see, the losses per length are more than twice as large for ENIG than for bare copper. Coverlay is somewhere between and EPIG is quite near ISIG and bare copper. The renonace bump on losses in ENIG is not visible, however that could be due to scale, different manufacturing process, different nickel alloy used or the traces could be smoothed. At least they look suspiciously smooth to me. The paper mentions the ENIG losses look somewhat like square root of frequency up to about 15 GHz. In 1935, Landau and Lifchitz proposed a theory that the corner frequency for nickel would be around 15 GHz. Maybe they were right. However, the researchers behind this article have very weak theory on why these losses happen in nickel, even though they reference the article about the permability modeling. As if they even did not bother to read an article they reference.

Should I avoid using ENIG?

You know the rule for questions in headlines. This time, the answer is "depends on your usecase" and not exactly "no". To make a few simple rules:

I am making something that operates on units or tens of GHz. Analyze the losses for several different finishes and then choose the one which is the most suitable for you. It may be the best one or the cheapest one depending on your needs.

I need to make some microwave measuring device that has an over-the-top performance and cannot withstand any losses which are not necessary. ENIG and ENEPIG are out of choice. Don't use them. If you need to have each device consistent with each other and very precisely controlled impedance on traces, avoid HAL.

I want to make some microwave stuff, but want it to be super cheap while still useable. HAL is probably your choice. OSP or bare copper can be fine too, but cost usually more. Avoid ENIG as it drives the price up significantly.

I am making general microwave stuff which has no special needs. Choose whatever you like, anything is good enough for you.

I am making a sub-GHz circuit which won't have insanely long signal traces all over the board anever will make anything over the GHz mark. You didn't have to read this article. Sorry for your time.

Discussions

Austin MacDonald wrote 09/29/2023 at 21:13 point

Thank you for making this article. I appreciate your candor and well structured commentary. This information will come in handy when selecting surface treatments for our low cost TDR pulser and sampler. However, I do wish you discussed the effect of the actual board substrate on the losses since this alone could be the cause of most losses. Since the EM wave/ energy does not actually travel in the copper but instead in the dielectric. Also, this surface finish discussion does not apply to inner layer wave guides such as strip-line since it only applies to outer layers. This would be a good distinction to make as well in your conclusion. 

  Are you sure? yes | no

Qbort wrote 05/29/2020 at 19:01 point

Amazing article. You probably saved me a lot of pain as I was indeed thinking I'd use ENIG for my first microwave project. Keep the good work!

  Are you sure? yes | no

john benham wrote 02/22/2020 at 20:15 point

Oh, I forgot, nice pics of the dendrite growths on the bottom of the traces. Would be so much nicer of we could use smooth rolled copper and get rid of the additional surface roughness losses. But then there's that pesky board delamination issue. Some smart person really ought to solve that  sometime.

  Are you sure? yes | no

MS-BOSS wrote 03/11/2020 at 19:42 point

It could theoretically be also caused by the cutting and polishingprocess used to get the cross-section. Sadly, the paper doesn't mention the method used for getting the cross-section.

  Are you sure? yes | no

john benham wrote 02/22/2020 at 20:06 point

Hmm, this for me was an interesting trip down memory lane. It took me back to my time in the late 90s/early 2000s at a start-up that had been acquired by Intel.  We had a technology that used electrically small directional couplers integrated into the  PCB channel traces to distribute multi Gb/s signals  on multi-load buses with extremely good low error rates. Of course, being a new idea it had every objection under the sun thrown at it, everything from manufacturability to testability. So I spent a few years of late nights running HFSS and coming up with several new coupler designs that  were insensitive to every manufacturing variation known to mankind and which could be turned out by any mom and pop PCB manufacturer in their home garage in Taiwan. Unfortunately risk aversion being what it is, I gather  the technology only got used in a few power/data rate critical niche applications despite its  proven performance advantages. 

As a result of all the simulation work and lab measurements we got to know  quite a lot about the  through and return loss mechanisms in various  PCB  board structures. I  recall coming in one morning to find the lab manager in a tizzy - the latest  test boards were showing unexpectedly high through losses - did we have a problem with the latest coupler design?  It turned out the spec called for a couple of um thick nickel flash passivation on the traces - a small fraction of a skin depth. When we sectioned a board we discovered  the nickel layer was something over 10um thick.  A quick HFSS  simulation confirmed that YUP, that would do it for you. It turned out the line operator had forgotten to set his timer and allowed the boards to spend much longer than required in the plating bath - he'd just assumed nobody else in the world would ever be any the wiser about his mistake. Nickel is conductive, so what's the problem?

I'm more than a bit skeptical of the Real/Imaginary  permeability curves. Just thinking about the analogy with the loss behavior of magnetic  ferrites I find the negative u' a bit unsettling. In my experience cobbling together loss mechanisms from behavioral models without tying them to the underlying physics is likely to end in tears. I wonder if the model is causal, are Zreal(s) and Zimag(s) a Hilbert transform pair? Perhaps if I ever get the time and energy I'll think about this some more.

  Are you sure? yes | no

MS-BOSS wrote 03/11/2020 at 19:40 point

Hi, sorry that I didn't answer. HaD didn't show me any notice of a comment...
In fact, I don't know much about the model. I take it as an "effort to somehow simulate reality without being physically correct, because the purpose is to make simulations more precise, nothing more.
Thus, I am not unsettled by the negative permeability. After all, negative permeability CAN happen. Even pure carbon has permeability below zero.

  Are you sure? yes | no

Andrew Nambudripad wrote 04/29/2020 at 15:07 point

I think this is the first time I've logged into HAD in 4 or 5 years, but I just want to tell you that it's projects like this and comments like yours that make HAD worth reading. Thanks for keeping this place rockin

  Are you sure? yes | no

MS-BOSS wrote 05/01/2020 at 17:31 point

Thanks :) Well, I remember HaD from times when it wasn't a blog about "How to stuff Arduino/RaspberryPi into X", but mostly about real hacking, reverse engineering and telling others how to avoid getting your fingers burnt and I would like these times to come back.

  Are you sure? yes | no