HDI Failure Beyond Microvia Cracks: Resin Starvation and CAF
When engineers discuss reliability in high density interconnect boards, the conversation usually goes straight to microvia cracks. Thermal cycling, expansion mismatch and copper fatigue are real mechanisms, well documented and increasingly well managed through better materials and tighter design rules. They are, however, only part of the story, and some of the most expensive field failures come from mechanisms that standard qualification testing never provokes.
Why Microvia Cracks Are Only Part of the Story
A microvia crack is a mechanical failure with a clear signature: it appears after enough thermal cycles, it concentrates at the via wall or the target pad, and it progresses predictably. Because it is visible in cross section and reproducible in a thermal shock test, it can be designed against. The failures that surprise teams are the ones that leave no such signature in qualification.
Those mechanisms usually involve chemistry and process rather than pure mechanics. Resin that never filled a cavity, contamination from an earlier process step, moisture that entered an incompletely cured structure: each can sit dormant through months of testing and then produce a failure in the field, when the combination of heat, humidity and voltage finally completes the circuit that should not exist.
Resin Starvation in Sequential Lamination
During lamination the resin in the prepreg flows to fill cavities, encapsulate copper features and bond the layers. When there is not enough resin available, the result is unfilled voids, incomplete bonding and poorly encapsulated buried structures. This is particularly common in HDI constructions because sequential lamination cycles consume resin progressively, and the second or third press has less flow available than the first.
Dense copper features make it worse. Filled via pads and heavy copper areas occupy volume that resin would otherwise fill, and fine pitch structures create narrow gaps that a high viscosity resin cannot penetrate completely. Designing with copper balance in mind, and choosing prepreg with a resin content matched to the copper volume, is the primary defence. Our notes on HDI lamination structure describe how those layers are built up.
Copper Balance and Prepreg Selection
Copper balance is a design constraint that is easy to ignore because modern layout tools do not enforce it. A board with dense copper on one side of a layer and sparse copper on the other presses unevenly, and the voids that form are not distributed uniformly. Balancing copper across each layer, and adding thieving bars where the density is genuinely uneven, improves both flatness and resin flow.
Prepreg selection then has to match. A higher resin content fills a large cavity volume more reliably but shrinks more on cure, while a low resin content laminate is dimensionally stable but less forgiving of topography. The relationship between the two is described in our comparison of prepreg and core, and it is the reason that stackup decisions belong in a conversation with the fabricator rather than in a template.
Moisture, Voids and Electrochemical Failure
A void is not a failure in itself. It becomes one when it absorbs moisture and the moisture, together with ionic contamination and an applied voltage, forms a conductive path. Polyimide and epoxy both absorb water, and the absorbed water turns to steam at reflow temperature, which is how a small void becomes a delamination blister that then extends under thermal cycling.
Controlling moisture means controlling bake schedules and floor time, and it applies to the bare board as much as to the components. Our guidance on moisture sensitive devices on assemblies covers floor life and rebaking, and the same discipline should apply to HDI panels before each lamination cycle and before assembly.
<img src="https://www.gopcba.com/wp-content/uploads/2026/08/PCB脉冲电镀.png" alt="Microsection of conductive anodic filament growth between PCB conductors” />
Conductive Anodic Filament Growth
Conductive anodic filament growth, usually shortened to CAF, is an electrochemical migration path that forms inside the laminate between two conductors at different potentials. It requires three ingredients at once: moisture, ionic contamination and a voltage gradient. Given enough time, a filament grows along the glass fibre bundles and eventually bridges the conductors, producing a short that appears spontaneously after months in service.
The risk grows as spacing shrinks, which is exactly what HDI demands. Small holes, thin dielectric layers and tight conductor spacing all bring the conditions closer together. Woven glass style, resin content and the quality of the drilling and desmear process all influence susceptibility, which is why CAF resistance is a material and process property rather than a design parameter alone.
Contamination and Process Residues
Every wet process step can leave something behind. Plating chemistry trapped in a blind via, flux residue that was not fully removed, handling oils from bare fingers: all of these provide the ionic species that electrochemical failure needs. Cleanliness is not a cosmetic requirement but part of the electrical specification for a high density board.
Cleanliness also has to survive the assembly process. A board that leaves the fabricator clean can be contaminated by a marginal cleaning step after soldering, especially where flux is trapped under a component body. Verifying ionic cleanliness by measurement rather than by appearance is worthwhile on any product where CAF or leakage current is a concern.
Ionic contamination is measured rather than observed. The standard approach is to extract residues from a sample board in a known volume of solvent and measure the conductivity of the resulting solution, producing a figure that can be compared with a specification and tracked over time. Appearance tells you nothing about ionic content: a board that looks clean can fail the measurement while a slightly discoloured one passes comfortably. Where a product sits in a humid environment with tight conductor spacing, that number is worth recording for every production lot, because it is one of the few process measurements that correlates with long term electrochemical reliability rather than with immediate assembly yield.
Qualification That Finds What Standard Tests Miss
Standard qualification applies thermal cycling and moisture sensitivity tests, and it is good at provoking mechanical failure. It is much weaker at provoking slow electrochemical mechanisms, which may need thousands of hours of biased humidity testing to appear. Where the product has a long service life and tight spacing, a biased humidity test with in situ monitoring is the honest way to assess the risk.
Design review should also include a check that the assumed materials actually match the delivered ones. A resin content or glass style substitution made late in procurement changes the failure risk without changing the drawings, and it is one of the few risks that no amount of layout review can catch.
FAQ
Are microvia cracks still the main reliability risk in HDI? They remain the most common mechanical failure and the easiest to test for. The mechanisms that catch teams out are the ones involving resin, moisture and contamination, because they behave as latent defects: they pass qualification and appear only after extended service in a warm, humid environment.
How do I reduce the risk of resin starvation? Start with copper balance, keep the number of sequential lamination cycles as low as the design allows, and select prepreg with a resin content matched to the copper volume it has to fill. Then confirm the choice with the fabricator using their process experience rather than assuming the nominal data sheet value.
What test best predicts CAF failures? A biased humidity test, sometimes called a temperature humidity bias or highly accelerated stress test, with continuous insulation resistance monitoring across fine pitch structures. Thermal cycling alone will not reveal it. The test needs to run long enough and be instrumented closely enough to detect a slow resistance drop.
