For a card edge that mates more than a few hundred times, the working range is 30 to 50 microinches of hard gold over 100 to 200 microinches of nickel, which is 0.76 to 1.27 micrometres of gold over 2.5 to 5.1 micrometres of nickel. Immersion gold on the same board is 2 to 3 microinches, roughly one fifteenth of that, and it is a solderable finish rather than a contact finish. Thickness and hardness belong in the same sentence: soft pure gold at 50 microinches wears faster than hard gold at 30.
What the standards do and do not say
ASTM B488 defines the deposit, not the application. Type II is 99.0 percent gold or better, which is the hardened grade used on contacts. Hardness codes run from B at 91 to 129 Knoop, through C at 130 to 200, to D above 200. Coating classes are minimum thicknesses of 0.25, 0.50, 0.75 and 1.0 micrometre. A drawing can therefore be completely compliant with B488 at 0.25 micrometre, which is 10 microinches and too thin for a connector that gets cycled.
Mating cycle life is not in ASTM B488 either. It belongs to the connector specification or the customer durability requirement, tested to IEC 60512-9-1 for mechanical durability, with a contact resistance limit as the pass criterion, usually a rise of 10 to 20 milliohms over the required number of cycles. IPC-6012 qualifies the board, IPC-4552 covers how the nickel and gold are measured, including the 1.5 mm by 1.5 mm spot that XRF readings are taken over. None of the three converts a thickness into a cycle count, and any table that does should be read as a starting point rather than a warranty.
Why the nickel underplate matters more than the gold
Gold on copper is not stable. Copper diffuses into the gold layer, the surface oxidises and contact resistance climbs, and the nickel underplate of 2.5 to 5 micrometres is what blocks that diffusion. The nickel also carries the mechanical load: it is the hard layer under a thin gold film, and it is what keeps the contact surface flat as the connector wipes across it.
The failure mode of a thin deposit is porosity rather than wear. Plating literature puts the porosity knee of gold around 0.5 micrometre, so at 2 to 3 microinches of immersion gold the deposit is porous as plated and the nickel is exposed in patches. In clean, dry air that is harmless. In sulfur-bearing air, in a coastal site, or in a cabinet with rubber gaskets, those patches corrode and the corrosion creep along the finger surface, which is why an edge that measured fine at goods-in reads high resistance two years later.
Wear is the other path, and it is abrupt. Once the gold is gone the contact sits on nickel oxide, and contact resistance moves from under 20 milliohms to over 100 milliohms. It reads as an intermittent link under vibration, not as a dead board, so it survives functional test and fails in the field.
Geometry decides as much as thickness
The insertion edge needs a bevel. A 20 to 45 degree chamfer, on both faces, is what lets the finger slide into the connector instead of shaving the plating off the leading edge and pushing a copper burr into the spring. A card edge also needs a routed profile, not a scored break: a V-score leaves a rough web that cannot hold a chamfer, and that is a panel decision made long before plating (the panelization article covers the cost side of that choice).
Gold has to stop before it reaches anything that takes solder. A joint above about 3 percent gold by weight is brittle, and the boundary is where rework starts to lift pads. On our line the gold area is defined on the drawing with a 0.5 mm clear zone to the nearest solder pad, and finger to finger spacing is held at 5 mil or more, because immersion gold bleeds under the mask and two fingers that touch are a short on the mating connector.
Three mistakes that reach the panel
Calling out ENIG on a mating edge. ENIG is 2 to 3 microinches of pure gold, soft at 60 to 90 Knoop, and it is there to protect nickel from oxidation before soldering. On a card edge it wears through in tens of cycles. It is the right finish for the solder side of the same board and the wrong one for the fingers, which is why mixed boards need selective plating rather than one global finish.
Specifying the gold and leaving the nickel to the shop. A drawing that says 30 microinches of gold and nothing about nickel gets whatever the line runs, and the diffusion barrier is then an assumption. Nickel minimum, its measurement locations and the gold boundary all belong on the same call-out.
Buying thickness instead of hardness. Soft gold at 50 microinches sounds better than hard gold at 30 and wears faster, because hardness, not thickness, sets how the deposit responds to the wipe and the contact force. Type, hardness code and thickness class are three fields on the drawing for that reason.
What SUPERB PCB supports, and where the line is
The published finish list is HASL, lead-free HASL, ENIG, OSP and electroplated nickel-gold. Electroplated nickel-gold runs 120 to 200 microinches of nickel with 8 to 12 microinches of gold, selective to the finger area, with a 5 mil minimum finger to finger spacing, 0.15 mm profile tolerance and 0.20 mm minimum trace to board edge. ENIG runs 100 to 200 microinches of nickel with 2 to 3 microinches of gold. Both are reported from XRF coupons per lot.
The honest boundary: our standard electroplated deposit is 8 to 12 microinches, which suits a board that is mated a handful of times. A connector specification asking for 30 to 50 microinches of hard gold, hardness code C, over 200 microinches of nickel is a heavier plating job, and it is confirmed with the plating house before we quote, with the XRF result reported against the drawing. We will not print 50 microinches on a drawing and ship 12.
What to send with the edge specification
Mating connector part number and its contact force, the number of mating cycles the product has to survive, gold thickness class and hardness code, nickel minimum and where it is measured, bevel angle, the gold area boundary, and whether the edge is lubricated or cleaned before packing. With those seven items the finish is fixed before the panel, and the conversation after the first lot is about yield rather than about a specification that was never going to hold.