I've seen similar effects with two different metals exposed during an etch.
(In one case, a contact to Au made Cr etch much faster.)
I believe that you are seeing a galvanic effect in which the more noble
metal (Au) results in the less noble metal (Ni) being etched faster than it
would alone.
This is listed in Part I of the etching papers along with many other
possible reasons for etch-rate variation.
One solution might be to make another mask to cover the plated nickel,
oversized for each feature by a few microns.
During etching, only gold is exposed. This of course leaves a ring of
nickel, which you may not be able to tolerate.
Another solution is ion milling.
    --Kirt Williams

----- Original Message -----
From: "Sven Holmström" 
To: "General MEMS discussion" 
Sent: Monday, May 21, 2007 4:03 AM
Subject: [mems-talk] Au-etchant attacking plated nickel


> Dear all,
>
> We selectively plate nickel onto gold coated silicon wafers. Later we etch
> the gold using standard gold etch (Aldrich, Potassium Iodide etchant). We
> have sometimes problem with the gold etchant etching also the nickel.
> According to data sheets and literature KI is not supposed to etch nickel
> (for example Etch Rates for Micromachining Processing—Part II by
> Williams,
> Gupta and Wasilik.).
>
> The keyword is 'sometimes'. The behaviour is very erratic. We have
> produced
> several nice pieces without any problems, but then at some times the
> nickel
> is etched even faster than the gold! (the nickel etch also in general seem
> uneven, etching from the sides in concentric half circles rather than
> evenly). It seems to us like the nickel etching is mainly depending on the
> individual nickel baths. We use all the time standard Nickel Sulphamate
> baths, but since we mix them by hand slightly individual differences are
> unavoidable. The concentric half circles also to tend to point  to
> contaminations being a problem.
>
> I wonder if anyone else has encountered this problem and in which
> direction
> we should go to get control of the process.