Electromigration

Electromigration is the gradual mass transport of metal atoms within a conductor, driven by momentum transfer from a high density of moving electrons. Over time, atoms migrate in the direction of electron flow, leaving voids (which thin and eventually open the conductor) at the cathode end and hillocks or extrusions (which can bridge to adjacent conductors and short) at the anode end. It is one of the dominant wear-out mechanisms in the fine on-die interconnects and power-delivery metal of modern ASICs — the silent clock ticking inside every hashing chip.

Why current density matters

Electromigration is governed by current density, not just total current. The classic predictor is Black's equation, where mean time to failure scales with the inverse of current density raised to a power (typically between 1 and 2) and with an Arrhenius temperature term — meaning lifetime falls exponentially as junction temperature rises. Because mining ASICs run hot and push very high currents through tiny geometries, both factors compound: a board running near thermal limits ages its interconnects far faster than one kept cool, and the damage is cumulative and irreversible. Each process-node shrink makes the wires thinner, which is why designers add wider power rails, redundant vias, and current-density rules — and why the margin you have to burn as an operator gets smaller with every generation.

What it means at the operating level

On a hashboard, the most current-stressed metal lives inside the ASIC die and in the on-package power rails feeding each hash domain. Electromigration there manifests as slow, cumulative degradation rather than a sudden event: a chip that starts throwing hardware errors at a frequency it once held comfortably, a domain that needs slightly more voltage each year to stay stable, hashrate that sags as damaged cores are worked around. By the time symptoms are visible, the physical damage is done — there is no rework that restores thinned on-die metal, which distinguishes it from board-level faults like a cold joint that a bench tech can actually fix.

The operator's defense

The practical defense follows directly from Black's equation: control temperature and current density. Keep junction temperatures down with clean heatsinks, good fan curves, or immersion cooling; avoid sustained operation beyond stock power. Aggressive overclocking raises both voltage and current through the same fixed metal cross-sections, so it trades directly against interconnect lifetime — a trade worth making consciously, not accidentally. Conversely, underclocking for heat-reuse or quiet home operation is one of the kindest things you can do to a chip's metal: lower current density and lower temperature both push wear-out decades further out. For a home miner planning to run hardware long past its warranty — the D-Central default — thermal discipline is not just about today's efficiency; it is about whether the silicon still hashes in five years.

For related component-aging topics, see our entries on thermal fatigue, which attacks the solder joints around the chip while electromigration attacks the metal inside it — two different clocks, both slowed by the same habit of running cool and steady.

Distinguishing it on the bench

Because electromigration lives inside the die, it is a diagnosis of exclusion. Board-level suspects — failing DC-DC stages, cracked solder under a package, dirty heatsink contact — are all measurable and fixable, so rule them out first with voltage-rail checks and thermal inspection before concluding the silicon itself has aged. A pattern worth recognizing: a machine that spent years overclocked in a hot container, now unstable at stock settings even after a competent board-level service, is the classic electromigration profile. The honest disposition for such boards is downward, not out: silicon too degraded for full-power stability will often run for years more at reduced frequency and voltage, which suits heat-reuse duty perfectly — a dignified retirement rather than e-waste.

Diagnose board faults with the diode & voltage reference.