Conductive anodic filament is a copper salt that grows from an anode to a cathode along the boundary between epoxy and glass inside the laminate, and it needs three things at the same time: absorbed moisture, a DC bias, and an unbroken glass fibre path between the two conductors. Remove any one of the three and it does not grow. That is why the fix is a spacing rule plus a fabrication and storage control, and not a conformal coating applied at the end.
The failure is slow and it is invisible. Insulation resistance falls over hundreds of hours, a leakage path appears on a humid day, and the board that passed every functional test in the factory fails in the field with a burn mark between two plated holes that were never meant to touch.
Why hole to hole is the worst case
CAF was identified in 1976 and the failure order has not changed since: hole to hole first, then hole to plane, then line to line. Drilling is the reason. A drill bit cuts through glass bundles and, if the bit is dull or the spindle runs out, it separates the resin from the fibre along the wall instead of cutting cleanly. That separation is a capillary. Moisture wicks into it, the DC bias drives copper ions along it, and the filament grows from the anode side, which is why the short appears at the positive hole.
Line to line is rarer because a track sits on a flat surface where the resin is intact. Hole to hole fails first because every drilled hole is an entry point into the same glass weave, and two holes on the same bundle are already connected by a path that only needs water.
The numbers behind the spacing rule
IPC-TM-650 method 2.6.25 is the test that puts a number on it. Coupons carry plated holes at edge to edge spacings of 0.27, 0.38, 0.51 and 0.65 mm, they are held at 85 C and 85 percent relative humidity with a DC bias of 10 to 100 V applied, and the failure criterion is insulation resistance falling into the megohm range. Laminate suppliers run the same structure at 65 C and 85 percent humidity for 500 hours as an accelerated screen. The results move in one direction: closer spacing fails sooner, and higher bias fails sooner.
Practice follows the data. Keep 0.5 mm of edge to edge spacing at 50 V, and push to 0.75 mm or more above 100 V. Stagger the holes rather than putting them in line on the same glass bundle, because a diagonal path is longer than a straight one through the weave. Where the layout cannot give the spacing, the answer is a CAF resistant laminate with a higher Tg, not a wider safety margin in the test.
Temperature matters as much as spacing. Above Tg the resin expands at 200 ppm per degree or more and opens the fibre interface, which is why a lead free profile peaking at 235 to 250 C is harder on the laminate than a tin lead profile at 205 to 220 C. Six reflow passes and several lamination cycles on a high layer count board accumulate that damage, and the CAF failure shows up months later in a humid installation.
What makes it worse in production
Drill quality and desmear. A dull bit, a high hit count, excessive spindle runout, or a desmear step that etches glass and leaves the resin surface rough. All of them open the interface that the plating was supposed to seal.
Moisture taken up after the final bake. A board stored out of its dry pack in a humid warehouse absorbs water into the laminate. Reflow turns that water into steam at the interface, a micro delamination forms, and the delamination is the channel CAF uses. Dry pack, humidity indicator cards and a bake to J-STD-033 before a second reflow are cheap compared with a field return.
Flux residue left on the surface between two biased conductors. Flux is hygroscopic and mildly conductive once it takes up water, and the published CAF work names flux residue as an accelerator of the failure. A no-clean process that leaves residue between a 48 V trace and its neighbour is a reservoir sitting on top of the same interface.
Three decisions that create the problem
Tightening hole to hole spacing to gain routing. The spacing is usually reduced on signal vias, which are harmless, while the pair that actually carries the DC bias is left at the same distance. Check the spacing on the highest biased pair, not on the average via field.
Treating a CAF resistant laminate as a substitute for spacing. A better laminate buys margin, it does not buy back 0.2 mm. Once the spacing is below the process floor for the voltage, the failure moves to a slower version of the same curve.
Accepting a laminate claim without a coupon. Ask for the test report at the spacing your stack actually uses, at the bias your product actually applies. A report at 0.65 mm does not cover a stack that runs at 0.3 mm.
What SUPERB PCB controls on the panel
Drill and spacing are checked against the drill table before release, mechanical drilling starts at 0.15 mm with hole position held to 0.075 mm, the aspect ratio is 10 to 1 on the standard line and up to 25 to 1 on the high layer count line, and Tg 130, 150 and 170 laminates are stocked with CAF resistant grades available on request. Plasma treatment is used on PTFE and mixed dielectric builds where the interface is hardest to seal, and panels ship with a cross section report so the plating and the resin interface can be inspected rather than assumed.
The honest boundary: qualifying a stack against CAF is a biased humidity test of 500 to 1000 hours at the laminate supplier or at a test house. What we supply is the coupon, the laminate report and a process that does not open the interface. The qualification itself belongs in the customer reliability plan, and the spacing rule is the part that can be fixed today. Stack and process limits are on high reliability PCB, the general window on PCB capability, and the laminate and process parameters on FR-4 process parameters.
What to send
Send the stack with laminate and Tg, the highest DC bias on the board, the edge to edge spacing between plated holes on that net pair, the number of reflow passes, the storage and dry pack requirement, and the reliability standard the product has to meet. That is enough to say whether the spacing holds or whether the laminate has to change.