Multiphase Buck Converter

A multiphase buck converter is several buck (step-down) converter stages wired in parallel and switched deliberately out of phase with one another to feed a single, demanding load. Modern CPU, GPU, and mining-ASIC core supplies routinely use six to twelve or more phases to deliver hundreds of amps at around one volt or less while staying above 90% efficiency. The voltage regulator stage on a hashboard — the row of inductors and paired MOSFETs feeding the chip domains — is almost always a multiphase buck, and understanding it is foundational for anyone who diagnoses mining boards.

Why interleave the phases

Splitting the load solves several problems at once. Each phase carries only a fraction of the total current, so its inductor and switches run cooler and can be physically smaller and cheaper than one heroic stage. The deeper trick is timing: with n phases fired evenly across the switching period (360°/n apart), each phase's current ripple arrives at a different moment, and the ripples partially cancel at the shared output node. The converter behaves as if it switched n times faster, so output voltage ripple collapses and far less output capacitor bulk is needed to hold the rail steady. Interleaving smooths the input side too, easing stress on upstream filtering. Spreading the dissipation across the board area instead of one hot spot improves thermal margins and long-term reliability — no small matter in hardware that runs flat-out for years.

How it feeds an ASIC hashboard

A hashing ASIC's core logic wants enormous current at well under a volt — territory where no single buck stage delivers cleanly. On a mining board the multiphase stage steps the PSU's distribution voltage down to feed the chips, which are organized electrically into series hash domains: voltage regulation happens per domain, with groups of chips sharing a supply, never per individual chip. Each phase typically uses synchronous rectification — a MOSFET in place of the freewheeling diode — because at these currents a diode's fixed voltage drop would burn intolerable power. A controller IC watches the output and balances current between phases, and better controllers shed phases at light load to keep efficiency high across the operating range, then bring them back as an autotuner pushes the chips harder.

The repair-bench view

Multiphase design is also a diagnosis map. Because phases share the load, one failed phase does not always kill the rail outright — the survivors pick up the slack, run hotter, and age faster, producing the classic pattern of a board that hashes but throttles, or fails only under full load or warm ambient. A shorted high-side MOSFET, by contrast, can drag the input rail down and take the board offline instantly. Thermal imaging finds a struggling phase quickly (one inductor conspicuously hotter or cooler than its siblings), and comparing per-phase temperatures is a standard first pass when a hashboard misbehaves — the kind of systematic fault isolation our repair service applies before any component leaves the board.

Position in the power chain

Layout is the unsung half of the design: each phase's high-current loop must be short and symmetric, sense lines must read the output at the point of load, and the copper planes themselves act as both conductor and heatsink. This is why hashboard VRM sections look so strikingly regular — repeated phase tiles marching down the board — and why liquid or immersion cooling helps the regulators almost as much as it helps the hashing chips.

The multiphase buck is the final stage of a long conversion chain: mains power passes through the PSU's power factor correction and isolated DC-DC stages before the board-level multiphase stage performs the last, hardest step down to core voltage. It is where the current is highest, the tolerances tightest, and — not coincidentally — where a large share of real-world hashboard faults live.