Jason Mills at Sandvine Inc was testing a complex system involving a long backplane with daughtercards at each end. He was running a number of 3.125-Gbit/sec XAUI signals through the system. The trace configuration for each signal comprised approximately 5 in. total of inner-layer routing divided between the two daughtercards, plus 20 in. of outerlayer routing on the backplane. The 5.5-mil differential microstrip traces on the backplane were built on an ordinary FR-4 substrate using a process that his manufacturing engineers called SMOG (solder mask over gold).
The SMOG process, as used at Jason's company, first passivates the microstrip traces by covering them with an inert coating through which corrosion will not penetrate and to which lead-free solder will readily adhere. Pure gold would seem ideal for that purpose. It is impervious to corrosion and bonds well to lead-free solder. A microscopically thin layer of gold, only 5 microinches thick, should be all that is necessary. Unfortunately, a thin layer of gold placed in direct contact with copper will slowly diffuse into the copper surface. Eventually, the thin layer of gold diffuses so deeply into the copper that it no longer provides a passivating effect. Shortly after that, the underlying copper corrodes.
To avoid the diffusion problem, place a barrier layer of nickel 120 microinches thick between the gold and copper layers. A plating of 120 microinches of nickel, followed by 5 microinches of gold, does the trick. That is the idea behind a process that some people call "ENIG". Jason's SMOG process starts with the ENIG layers of thick nickel followed by a thin gold overlay, and tops that with a final layer of soldermask.
Either the ENIG or SMOG process will keep your metallurgists happy, but what about signal quality?
As explained in my article “Nickel-Plated Traces,” nickel plating substantially increases the high-frequency resistance of a PCB trace. It lengthens the step response of the trace, exacerbating both intersymbol interference and jitter. Microwave engineers have known this for decades.
When Jason got his first boards back and observed poor eye quality, he quickly surmised that the nickel might be hurting him. That’s a good guess, but frankly, a lot of people might have guessed that. Jason’s next move is what distinguishes him from other engineers. He wanted a way to test his hypothesis, directly and convincingly, in a few minutes, without respinning the boards. Here’s what he did.
First, he measured the system eye pattern. Then, he physically sanded the board, using #220-grit sandpaper, completely removing all of the solder mask, gold, and nickel covering the long, exposed sections of the microstrip traces on his backplane, and reshot the eye pattern (figures 1and 2). Brilliant.
Sanding removes solder mask and also thins the traces; both effects incrementally raise the differential trace impedance. Sanding also introduces a considerable amount of surface roughness. Despite those disadvantages, Jason’s sanded traces show markedly better performance than the original SMOG trace.
In the final system, Jason used a solder-mask over bare-copper process to cover the long copper microstrips and eliminate nickel from those regions. He then treated all of the exposed soldering pads with a different process to make them solder friendly.
It hardly matters what process you use for soldering pads, because the soldering pads are electrically quite short. An ENIG process would work fine for solder pads, as long as the ENIG is confined to only the pad and the long trace runs are passivated some other way.
Any two-step approach requires additional patterning and masking steps that make the board a bit more expensive, but in this case the expense seemed well justified. As expected, Jason's final system performed even better than his quick and dirty sanding experiment.