Differences Between Electroless Gold and Electroless Silver in PCB Fabrication

The difference between electroless gold and electroless silver is usually explained as cost against performance, and that explanation misses the physical reason the two behave differently at high frequency: one of them contains nickel and the other does not. ENIG is a 3–6 µm nickel barrier with 0.05–0.10 µm of immersion gold on top, and at 10 GHz the skin depth in nickel is 1.33 µm, so the current runs in the nickel and never reaches the gold. Immersion silver deposits 0.12–0.40 µm of silver directly onto copper by displacement, with no nickel at all. That single structural fact explains the loss difference, the cost difference, the failure modes, and why one is the right finish for a 28 GHz board and the other is the right finish for a connector edge.

The earlier version of this page described electroless silver as “nickel plating followed by a chemical deposition of silver”. That is not how immersion silver works and it inverts the only difference that matters. Immersion silver is a displacement reaction: the silver ions in solution are reduced as the copper beneath them dissolves, and deposition stops by itself when the copper is covered. There is no nickel, no catalyst and no external current. ENIG is the opposite arrangement: an autocatalytic nickel deposit 3–6 µm thick, followed by a thin immersion gold layer that exists to stop the nickel oxidising, not to conduct.

The naming is loose in ordinary use and worth tightening once. “Electroless gold” is almost always ENIG — electroless nickel, immersion gold — where only the nickel is truly electroless and the gold is a displacement layer. “Electroless silver” is immersion silver, also a displacement layer. Neither finish is electroplated, which is what distinguishes both of them from hard gold, a subject that causes its own confusion (section 9).

At high frequency the current occupies a surface layer whose thickness falls as the square root of frequency, and the resistance a conductor presents is therefore ρ/δ per square rather than ρ/t. For the four metals involved at 10 GHz the numbers are close enough to compare directly.

Two things follow, and they are the whole of the high-frequency argument. First, silver is not meaningfully better than copper — a 6 % advantage — so the benefit of immersion silver is not that silver conducts well but that nothing worse sits between it and the copper. Second, the nickel in ENIG has twice the surface resistance of copper, and it is the layer the current actually uses, because the gold above it is thinner than a tenth of a skin depth at any frequency above a few gigahertz.

The gold thickness deserves its own calculation, because “gold is a good conductor” is true and irrelevant at 0.075 µm. Its sheet resistance as a layer is ρ/t, not ρ/δ, whenever the layer is thinner than the skin depth: 2.44×10⁻⁸ / 0.075×10⁻⁶ = 0.33 Ω per square. That is twelve times the surface resistance of the copper underneath and six times that of the nickel it covers. The gold is a corrosion barrier of negligible conductance, which is exactly what it is designed to be.

The penalty is not applied everywhere, and knowing where it applies prevents both over-reaction and wishful thinking. Three cases decide it.

Outer-layer microstrip. The finish sits on the conductor surface that faces away from the reference plane. That surface carries the smaller share of the current — typically 20–30 % of the trace current for a microstrip at W/h ≈ 2 — so a finish with twice the surface resistance adds far less than a factor of two to the trace loss. Measured on 10 GHz microstrips, the difference between ENIG and immersion silver is typically 0.03–0.10 dB/in, and it grows with √f because the absolute surface resistance does.

Stripline. The conductors are buried in laminate and never see the finish. The difference between the two finishes is zero, and any specification that pays for immersion silver on a stripline-only design is buying nothing electrically.

Ground plane and connector interface. Where the current is forced through a plated surface — a press-fit pin, a pogo contact, a connector launch — the finish is in the current path completely, and the nickel becomes a series resistance rather than a surface effect.

Put into a budget, a 10 GHz microstrip on RT/duroid 5880 at 0.254 mm loses about 0.22 dB/in with smooth copper. Adding a finish moves that to roughly 0.25–0.32 dB/in with ENIG and 0.22–0.24 with immersion silver, which on a 100 mm feed is a difference of 0.1–0.3 dB. That is small in absolute terms and decisive in a link budget with 1 dB of margin, which is the honest way to present it.

If immersion silver is cheaper and electrically better, the obvious question is why ENIG is the default. Five reasons, all of them mechanical or metallurgical rather than electrical.

Aluminium wire bonding. Gold wire bonding needs a gold surface; aluminium wedge bonding works on ENIG and not reliably on silver. Any board with a bare die or a chip-on-board assembly is ENIG by default. Edge connectors and contact surfaces. A surface that will be inserted, wiped or pressed needs wear resistance, and that means nickel under gold — though for a connector the answer is usually hard gold rather than ENIG. Shelf life. Gold does not tarnish; silver does, and a board that sits in a warehouse for a year needs the protection. Fine-pitch solderability. A flat, hard, oxidation-free pad is easier to print and reflow on than a soft silver surface, particularly at 0.4 mm pitch and below. Multiple reflows. ENIG survives several lead-free cycles without the surface changing; silver dissolves into the solder and thins with each one.

There is also a perception reason that is worth naming because it is not a reason: ENIG is what everyone specifies, so it is what everyone knows how to inspect. A fabricator with a well-controlled ENIG line and a poorly controlled silver line will produce better boards with ENIG regardless of the physics, and the control of the line is a legitimate part of the specification.

Immersion silver wins on three counts and each has a number. Loss: no nickel barrier means the extra surface resistance is close to zero, worth 0.03–0.10 dB/in at 10 GHz on an outer microstrip and more at 28 GHz. Cost: two process steps fewer, no nickel bath to analyse and no gold in the bath, which is why silver is the cheaper finish on a per-square-metre basis by a margin that varies with the gold price. Planarity and solderability: 0.12–0.40 µm of silver follows the copper topography exactly, with no nickel edge build-up at fine lines, and solderability to SAC305 is excellent on a fresh surface.

The conditions attached to it are equally concrete. It must be packed with sulfur-absorbing paper, because atmospheric sulfur tarnishes the surface within weeks in an uncontrolled environment — the tarnish is a cosmetic and solderability issue rather than a conduction one, since silver sulfide is still conductive, but it fails visual acceptance and it worries inspectors. It should not be used where the assembly will be stored humid and biased for long periods, because silver migrates under those conditions. And it should not be used as a wire-bond surface.

Two costs are not in the per-square-metre price and both favour gold. The first is shelf life. A silver surface begins to react with atmospheric sulfur within weeks, so it is supplied with sulfur-absorbing interleaf and a stated storage window, typically six to twelve months in the original packaging and considerably less once the bag is opened. A gold surface does not react at all, which is why boards that will be assembled in a different country, months after fabrication, are usually finished with ENIG regardless of the electrical argument.

The second is rework. A silver surface can be reflowed more than once, but each cycle dissolves some of the 0.12–0.40 µm into the joint, so a board that will see three reflows effectively starts with less silver than the drawing says. ENIG loses nothing to the first reflow, and the failure it does have — embrittlement above 0.1 µm of gold — is a thickness specification rather than a wear mechanism. Solderability is verified to IPC J-STD-003 by wetting balance or dip-and-look, and the test is worth writing into the purchase specification for either finish, because it is the only check that catches a bath that has drifted.

A third consideration belongs to the laminate rather than to the finish but shows up in the same failure analysis. On a PTFE board the adhesion between copper and laminate is the weak link, and a plating line that is aggressive to the laminate surface will produce a delamination that is later blamed on the finish. The interaction between surface treatment and copper adhesion on RO5880 specifically is covered in preventing copper foil delamination on RO5880 boards, and the acceptance tests that would catch it are in testing steps in high-frequency PCB production.

Both finishes fail, and they fail in ways that look like different problems on the bench. Knowing which one produces which symptom shortens a lot of investigations.

The asymmetry is worth noting: ENIG failures tend to be metallurgical and appear as joint reliability, while silver failures tend to be chemical and appear as inspection or leakage. A programme that has no history with either will usually find ENIG easier to accept and immersion silver easier to defend on measurement.

The IPC documents give windows rather than single values, and each end of each window has a reason.

Two of these windows are measured in tenths of a micrometre and are therefore as much a process-control statement as a design one. A gold thickness of 0.05 µm is 50 nm, well inside the capability of a controlled immersion bath and well outside it if the bath is not analysed; the difference between a porous and a sound ENIG surface is often 20 nm.

The vocabulary causes more mistakes than the chemistry. Three finishes contain gold and they are not interchangeable. Immersion gold is 0.05–0.10 µm over nickel, is what ENIG means, and is a solderable and bondable surface. Electroless gold in the strict sense — a thicker autocatalytic gold over nickel — exists but is rare and expensive. Hard gold is electroplated gold alloy, 0.5–1.5 µm, usually with cobalt or nickel, over 3–5 µm of nickel, and exists for wear: it is specified for edge connectors, test points and anything that will be inserted repeatedly.

The mistake is to specify “gold 1 µm” on a solderable pad. That is hard gold, it will not solder reliably if it contains cobalt, and it costs several times ENIG. The second mistake is to specify ENIG for an edge connector, where 0.1 µm of gold wears through in a handful of insertions. The third is to assume that a thicker gold layer improves high-frequency performance: it does not, because the current is in the nickel either way, and it makes the joint more brittle.

1. The finish by name and standard — ENIG to IPC-4552B or immersion silver to IPC-4553A, not “gold”. 2. The thickness window with the measurement method, since 0.05 µm of gold and 0.10 µm behave differently in a joint. 3. The area to which it applies — a common and expensive error is to finish the whole board in hard gold because a connector edge needed it. 4. The packaging requirement for silver — sulfur-absorbing interleaf and a stated shelf life. 5. The reflow count, so that the finish thickness is chosen for the number of cycles the assembly will actually see.

Each line answers a question that would otherwise be settled by a failure analysis three months later. The wider subject of how finishes fit into the process, including the ones that are neither gold nor silver, is covered in surface treatment processes in PCB manufacturing and in surface treatment for high-frequency PCB production.

The differences between electroless gold and electroless silver come down to a layer that one of them has and the other does not. ENIG is 3–6 µm of nickel with 0.05–0.10 µm of gold on top; at 10 GHz the skin depth of nickel is 1.33 µm and of gold 0.79 µm, so the current runs in the nickel and the gold is a corrosion barrier whose sheet resistance, at 0.33 Ω per square, is twelve times the copper beneath it. Immersion silver puts 0.12–0.40 µm of silver straight onto copper by displacement, with no barrier at all, and its surface resistance is within 6 % of copper’s.

That structural difference produces the cost difference, the loss difference of 0.03–0.10 dB/in on an outer microstrip at 10 GHz, and completely different failure modes: black pad and gold embrittlement on one side, tarnish and silver migration on the other. It also explains why the choice is not a matter of taste. Above about 10 GHz on an outer layer, in a link budget with real margin, silver is the better conductor; where the board has a wire bond, an edge contact, a long shelf life or a fine pitch that must survive several reflows, ENIG is the better finish and the 0.1 dB is the price of reliability.

XCE PCB runs both finishes in house on high-frequency builds: ENIG to IPC-4552B with nickel at 3–6 µm and gold at 0.05–0.10 µm, immersion silver to IPC-4553A with sulfur-absorbing packing, hard gold for edge contacts, on 2 to 72 layers with impedance held to ±10 % and samples in as fast as 24 hours. The copper under the finish is the other half of the conductor, and its roughness is covered in PCB copper foils; the measurement that reveals a finish problem is in electrical characteristics in high-frequency PCB manufacturing.