Differential Pair

A differential pair is two PCB traces routed side by side, carrying complementary signals of equal magnitude and opposite polarity. The receiver looks only at the voltage difference between the two lines, so any noise that couples equally onto both traces — common-mode noise — cancels out at the receiver. This rejection is what makes differential signaling the standard for high-speed interfaces such as USB, PCIe, HDMI, and Ethernet, where a single-ended trace of the same length would drown in crosstalk, supply noise, and radiated interference. The pair also radiates less itself, since the two opposite currents create largely cancelling fields.

Routing requirements

The noise cancellation only works while the two traces remain electrically identical, so differential routing is a discipline of symmetry. The core rules: matched length, so both edges arrive together — a length mismatch converts into timing skew, and skew converts clean differential signal into common-mode noise the receiver cannot reject; constant spacing, because the gap between the traces (together with their width and the dielectric) sets the differential impedance, and any change in spacing is an impedance discontinuity that reflects energy; and a continuous reference plane beneath the pair for the return currents. Common targets are roughly 90 ohms differential for USB, 100 ohms for Ethernet, and 85 ohms for PCIe. Crossing a split in the ground plane, necking the pair through a congested area asymmetrically, or placing a via on one trace but not the other all break the symmetry the whole scheme depends on. Serpentine tuning — the little accordion wiggles you see on motherboards — exists purely to equalize lengths within a pair and between pairs.

Relevance to mining boards

Control boards in ASIC miners use differential pairs for exactly the signals that cannot tolerate corruption: DDR memory clocks and strobes between the SoC and RAM, Ethernet pairs to the PHY and magnetics, USB, and high-speed clock distribution. The hashboards themselves are more forgiving — the chip-to-chip daisy-chain signalling on a hashboard is comparatively low-speed — but the control board is genuinely a high-speed digital design, and it fails like one. At the repair bench this has a practical meaning: when a board has been dropped, flexed, or corroded, a lifted pad or a hand-soldered jumper wire on one half of a pair introduces length mismatch and an impedance discontinuity. The result is rarely a clean failure; it is flaky Ethernet that negotiates down, a network port that dies when the board warms up, or chains that drop offline under load. Treat both traces of a pair as a single unit during any rework: if one needs a jumper, match the other, keep the wires paired and short, and route them over intact copper.

Reading a board for pairs

You can usually spot differential pairs by eye: two traces of identical width running in lockstep, spaced closer to each other than to anything nearby, with matched serpentine bumps. When repairing near them — replacing a nearby connector, wicking a solder bridge — avoid scratching the soldermask over the pair and avoid leaving flux residue bridging them, since contamination between the traces changes the local dielectric and can degrade the link margin on a board that used to work.

Differential pairs are a specific case of controlled impedance routing, and they rely on a solid, unbroken reference plane for their return path. For repairs beyond bench scope, D-Central's repair service handles control-board-level faults where these constraints matter most. The habit that saves boards is simple: before cutting or jumping any trace, ask whether it has a twin — and if it does, plan the repair for the pair, not the wire.