Thermal Fatigue

Thermal fatigue is the progressive cracking of solder joints and interconnects caused by repeated heating and cooling cycles rather than a single overstress event. Each power-up and power-down expands and contracts the materials in an assembly. Because a solder joint, the component body, and the board laminate all expand at different rates — a coefficient-of-thermal-expansion (CTE) mismatch — the joint is forced to absorb cyclic shear strain. Over many cycles this initiates and propagates cracks until the connection becomes intermittent, then open. Nothing "broke"; the assembly simply wore out, one thermal cycle at a time.

How engineers model it

Solder-joint fatigue life is commonly described with Coffin-Manson-type relations, in which life correlates inversely with the strain amplitude experienced each cycle. The practical corollaries follow directly: larger temperature swings shorten life; larger components shorten life, because CTE mismatch accumulates over a longer span — which is why a big BGA package like an ASIC dies before the 0402 resistor beside it; and stiffer joints concentrate rather than absorb strain. Cracks usually nucleate at the joint-to-pad interface where shear strain concentrates, then grow with each cycle until resistance rises and the fault appears — classically as a board that fails when cold and recovers once warm, or the reverse, as expansion opens and closes the crack.

Why mining hardware is exposed

A hashboard is close to a worst-case exhibit: dozens of large hash chips dissipating serious heat, clamped between heatsinks, on a thick copper-heavy laminate. Every trip between ambient and full operating temperature is one fatigue cycle, and operational patterns multiply them — curtailment schedules, unstable power, aggressive auto-restarts, and seasonal on/off operation can put thousands of extra cycles through the joints that a steadily running machine never sees. Heat-reuse deployments that cycle miners with a thermostat face the same arithmetic. The failure signature on the bench is familiar: a chain that drops chips intermittently, an X-pattern of failed chips near the hottest zone, or a board that passes cold-start tests yet falls over under thermal load. The repair — reflowing or replacing the affected chip via BGA rework with a hot air rework station — restores the joint, but the neighbouring joints carry the same mileage.

On the bench, fatigue cracks announce themselves to the right instruments before they fail outright: a thermal camera shows a chip running oddly hot or cold relative to its neighbours as the joint's resistance drifts, gentle flexing of a powered board can make an intermittent chain appear and vanish, and chip-enumeration tests that pass cold but fail warm point straight at a heat-opened crack. The diagnosis matters because a fatigue failure is positional evidence — it tells you where the board's thermal and mechanical stress concentrates, and the next failure is usually nearby.

Slowing the clock

You cannot repeal materials science, but you can starve it of cycles and amplitude:

• Avoid unnecessary power cycling — a miner that runs continuously accumulates fatigue far more slowly than one bounced daily; prefer throttling via underclocking over full shutdowns where the economics allow.
• Keep operating temperatures moderate and stable; the swing matters more than the average, so clean airflow and dust management do double duty.
• Fix cooling problems promptly — a failing fan converts steady-state operation into a slow thermal oscillation.
• On the bench, use a preheater and gentle profiles so the repair itself does not pre-age every joint around the one you fixed.

Related wear-out mechanisms are covered in our entries on solder joint fatigue, the joint-level view of the same physics, and electromigration, the current-driven counterpart that gnaws at the silicon itself.

Trace intermittent faults with the diode & voltage reference.