Antminer – Hashboard Short Circuit

Kill power at the breaker — not the PSU switch. A tripped miner under fault can re-latch if the PSU's over-current timer resets before you unplug it. Leave the breaker off for at least 60 seconds and unplug the PSU from the wall. Label the breaker so a second person can't flip it back on while you're working. This is non-negotiable: every second under short-circuit current kills more copper and more adjacent components.

Photograph the scene before you touch anything — hashboards in their slots, PSU, control board, any visible damage. If this escalates to D-Central, those photos are the first thing a bench tech wants. Phone camera is fine. At least one photo must show each hashboard face-up with the slot number visible. Photos also protect you if the miner is under warranty and you need to document pre-existing condition.

Remove and isolate every hashboard. Unscrew the chassis, unplug data ribbons and power cables at the board side (not the PSU side), and pull each board out onto an anti-static mat or clean cardboard. Label each with its slot position (0/1/2) using tape. Inspect under bright light for burnt components, cratered MOSFETs, bulging caps, split MLCCs, scorched PCB, or dark discolouration. Record which board shows damage and in which physical region.

Verify the chassis is safe to re-energize without hashboards. With boards out, re-check the control board and PSU for visible damage. Any melted insulation, blackened connector, or persistent burnt-plastic smell means stop and ship to D-Central. If the chassis looks clean, you can proceed to Tier 2 testing. Do not re-power with hashboards installed until you've completed the Tier 2 DMM check on each board.

Check your electrical environment. Inspect outlet, breaker, and PSU plug for heat damage. If the breaker tripped on power-on, audit the circuit — S17-and-newer miners need dedicated 240 V circuits, not shared 120 V household circuits. A re-tripped breaker with a known-good miner means a circuit problem, not a hashboard problem. Running a 3250 W S19 on a 120 V / 15 A circuit is a standing invitation to future shorts.

DMM short-check each hashboard, board-out, power-off. Multimeter on ohms, lowest non-continuity range. Probe `+12V_IN` to `GND` on the board-side connector. Healthy S19-class reads > 500 Ω after caps settle (the reading ramps up from low hundreds over 5–10 seconds as decoupling caps charge). Below 20 Ω that doesn't rise = input-side short. Repeat for the `VDD_MID` (19 V boost) rail to GND, and for the 0.4 V domain rails to GND. Record which rail is shorted on which board.

PSU-only bring-up test. Disconnect every hashboard; leave control board and fans connected. Power on from a different outlet on a different circuit than where the fault occurred. If the PSU still clicks or trips, the PSU is damaged — swap with a known-good APW9 / APW12 matched to your miner (verify against Bitmain's compatibility table; wrong-model PSUs have shipped shorts). If the PSU is now stable, one of the hashboards was the fault — proceed to Step 8.

Swap hashboards between slots (miner powered off throughout). If you have a known-good board, install it in the suspect slot and power on with only that board. If `ERR_SHORT` still appears with a known-good board in the slot, the backplane / control-board power path is the fault — go to Step 16 (Tier 4). If the fault follows the original board to a new slot, the board itself is the fault — proceed to component-level isolation.

Inspect every power connector for heat damage — PCIe-6 and PCIe-8 power connectors on S19-class boards melt under sustained high current, especially with used or low-grade PSUs. Look for discoloured plastic, darkened copper pins, pulled-back retention clips. A melted connector is a replace-the-cable job at minimum, and the board-side header sometimes needs replacement too. Melted connectors can mimic a short by dropping enough voltage to trip protection circuits.

Check line voltage under load at the miner's outlet. Plug a known-good miner or a dummy load (a space heater works). Measure outlet voltage with a DMM. 240 V split-phase should hold ≥ 235 V; 208 V commercial ≥ 202 V; 120 V residential ≥ 115 V (and on 120 V you're undersized for any S17-or-newer regardless). Low line voltage accelerates PSU wear and makes future shorts more likely — fix the upstream circuit before the repaired board goes back in.

Bench-PSU current-limit bring-up. Set a lab PSU to 13.8 V, current limit 2 A. Connect only the suspect hashboard's `+12V_IN` and `GND` — no ribbon cables, no control-board signal. A healthy board draws < 0.2 A in pre-bringup, settling within 2–3 seconds. A shorted board clamps at the 2 A limit and voltage sags to < 5 V. When the voltage sags, immediately run your thermal camera or IR thermometer over the board. The shorted component will be the first to heat up — typically within 5–10 seconds. Cut power as soon as you identify it.

Thermal-image the fault. A FLIR ONE Pro, or even a decent IR thermometer, catches the hot component during bench-PSU bring-up. Input boost FET, voltage-domain PMICs, shorted chip positions, and cratered MLCCs all heat faster than surrounding PCB. Mark the hot component with a silver marker and photograph it — that photo is the blueprint for the rework step. If nothing heats within 30 seconds at 2 A limit, the fault is probably inside a chip package and needs chip-level isolation.

Replace a shorted boost / buck FET. Desolder the flagged FET with hot air — bottom-side preheat ~150 °C, top-side 310–330 °C for ~30 seconds. Clean pads with braid + flux. Drop a new FET of the same or equivalent part number (Bitmain's private-marked FETs are re-marked Infineon / Alpha & Omega parts; Zeus Mining's S19 service guide has the best public cross-reference). Reflow, clean with IPA 99%, re-test with bench PSU at 2 A limit before reinstall.

Replace a shorted MLCC. Drag-solder or hot-air off the cracked cap (usually 0402 or 0603, 10 µF or 22 µF depending on domain). Replace with a same-voltage-rating part — do not substitute a 6.3 V cap into a 25 V rail. Clean with IPA, re-test. MLCC replacement is the cheapest component-level repair on the board and also the most finicky — a solder bridge between the 0402 pads will short the same rail you just unshorted, so inspect under a loupe before re-powering.

Cross-flash DCENT_OS for post-repair validation. Once the repaired board passes the bench-PSU short test, reinstall in the miner and cross-flash DCENT_OS — D-Central's own open-source Antminer firmware — with per-chip HW%, autotuning, per-domain voltage/temp diagnostics, and stratum v2 support. Alternatives: Braiins OS+, LuxOS, or Vnish. The per-chip view tells you within 20 minutes whether the repaired domain is behaving. Stock Bitmain firmware only shows chain-level aggregates and will hide a marginal chip until it re-fails.

Know when to stop DIY. Book D-Central when: (a) visible fire damage, bulging caps, or PCB delamination; (b) two hashboards in the same rig failed with ERR_SHORT inside 30 days — upstream issue, not coincident failure; (c) the control-board power rail is damaged and you don't have a known-good replacement; (d) you lack a bench PSU with current limiting, hot-air rework station, or BGA / fine-pitch solder experience. Pushing past any of these lines is how a $75 MLCC repair turns into a $900 full-board replacement.

What D-Central does on the bench: test fixture with programmable current-limited load; per-domain voltage-rail validation against Zeus / D-Central reference maps; full component-level isolation via thermal imaging and schematic-level probing; component replacement using graded FETs, PMICs, MLCCs, and salvaged-or-new BM1398 / BM1362 / BM1368 chips; full reflow + reseal; 24-hour burn-in at nameplate with DCENT_OS per-chip logging to catch marginal domains before the miner ships back. Estimated turnaround: 5–10 business days from receipt.

Ship safely. Anti-static bag per hashboard. Double-box with ≥ 5 cm foam on every side. Ship the control board separately if suspect. Include a printed note with: observed symptoms, kern.log excerpt if available, firmware version, date of first fault, electrical environment (120/208/240 V, single-phase / split-phase / 3-phase), and contact info. That note saves roughly 30 minutes of diagnostic time per board, and diagnostic time is what you pay for.