IceRiver Hashboard LDO Voltage Fail: Domain Diagnosis

Note the deficit. Compare per-board hashrate against nameplate via the web UI. Write down per-board number and the absolute deficit. Confirm it is a near-integer multiple of `Total / N` (one chip's throughput). Saves diagnostic time later — you'll know up front whether you're looking for one dead domain, two dead domains, or a different fault entirely.

Visual sweep with the chassis lid off. Power off at the breaker, ESD strap on, lid off. Photograph every hashboard. Look for scorched LDO packages, blackened tantalum capacitors, swollen / cracked passives, lifted solder joints, or thermal pads visibly slid out of place. Half the time you spot the dead LDO before pulling a multimeter out.

Confirm the failure is not transient. Cold-boot the miner once at the breaker (`30 s` off, then on). Watch hashrate for 15 minutes. If it climbs back to full nameplate, you did not have an LDO fault — you had a boot-enumeration glitch. Stop and re-evaluate against the [IceRiver 0 ASIC Chips Detected](https://d-central.tech/asic-troubleshooting/iceriver-zero-hashrate-0-asic-chips-detected/) page or the [IceRiver Web Hashrate vs Pool Hashrate Gap](https://d-central.tech/asic-troubleshooting/iceriver-web-hashrate-vs-pool-hashrate-gap/) page.

Thermal-camera sweep at full load. A `FLIR ONE Pro` (~$300 CAD) plugs into a phone. Run miner at nominal hashrate for 10 minutes. Sweep each hashboard end-to-end at `30-50 cm`. Mark the position of any cold cluster. This is the single most valuable Tier-2 diagnostic on this fault. Most KS LDO repairs start and finish with this step alone.

Multimeter probe the cold cluster's `1.8 V` and `0.8 V` rails under load. Compare against a healthy cluster on the same board. Note rail status: dead (`~0 V`), low (`1.3-1.5 V`, drift), or healthy. Then probe the upstream rail — `2.2 V` for low domains, or boost output `TP872` for high domains (`24-28` on `KS3`, `12-14` on `KS3L`). Confirm whether the deficit is downstream-only (LDO fault) or upstream (boost / PSU fault).

If both LDO output and upstream input rail are pulled down: short downstream of the LDO. Visually inspect tantalums on that domain for scorching / swelling / cracks. Use a `200 mA` current-limited bench supply on the affected rail to inject current — the shorted component heats first under your finger or thermal camera. On `KS0`: power-on showing current limit at `1A` with voltage below `19V` indicates short circuit (Zeus Mining documented procedure). Disconnect each phase to isolate stage.

Refresh thermal pads on the cold cluster — even before reflow. Pull the heatsink, peel and replace thermal pads on the LDO and its 4 chips. Use `1.5-2.0 mm` pad stock with `~6 W/m·K` rating. If the LDO was drifting from heat (most common KS LDO failure mode), the pad refresh alone often brings it back into spec without further intervention. Re-test rails before deciding to reflow.

Boost-converter check (`KS3` / `KS3L` / `KS5L` / `KS5M` only). On boards with a high-domain boost stage, locate `TP872` and `TP873`. Probe `TP872` for boost output: expect `15.5 V` (KS3) or `17 V` (KS3L). Probe `TP873` for gate-drive: expect `22 V` (KS3) or `26 V` (KS3L). Boost output dead = `MP1517` or its driver MOSFETs failed. Gate drive missing = controller-side or feedback-loop fault.

Reflow the suspect LDO. Preheat board bottom to `~150 °C` (hot plate or PCB preheater). Top-side hot air at `310-330 °C` for `~30 s` directly on the LDO package. Do not blast cold air after — let the board cool naturally to avoid thermal-shock cracks on adjacent ceramics. Re-test `1.8 V` and `0.8 V` rails after cooling. About `40-60 %` of out-of-spec LDOs recover with a clean reflow.

Replace tantalum capacitors that show damage. A scorched / cracked tantalum next to a dead LDO is almost certainly the root cause — tantalums fail short. Replace with same value / same voltage rating, or step the voltage rating up one notch for margin. Soldering iron + flux + tweezers is sufficient for most KS tantalum packages. Re-test the rail after the replacement before condemning the LDO.

Replace the LDO if reflow + cap replacement does not recover the rail. Use the original silkscreen designation: `MP2019`, `SGM2036`, or `SGM2036-ADJ` are the common parts on retail KS hardware. Source from LCSC, Digi-Key, or pull from a salvage board. Apply flux to pads, place new part with tweezers, hot air at `310 °C` for `~25 s`. Watch the package self-align as solder reflows. Probe rails before reassembling the heatsink.

Reflow the boost converter (`MP1517` on `KS3` / `KS3L`) if `TP872` diagnostics fingered it. Same procedure as LDO reflow, slightly larger thermal mass — `~35 s` of hot-air time. After reflow, probe `TP872` for `15.5 V` / `17 V` output before reassembling. If reflow does not recover the boost, replace the part using the same hot-air technique.

Replace boost-stage MOSFETs (`U9-U13` on `KS3` / `KS3L`) if `TP873` gate-drive voltage is missing. Match part numbers from silkscreen. Hot-air rework, tweezers placement. This is the deepest non-chip-level repair in the KS fault tree but still standard SMT work — no microscope required for the FET packages used here, though a magnifying lamp helps.

Refresh thermal paste on the entire chip array on that board. Once the chassis is open, refresh paste on every chip with Arctic MX-6 or Thermal Grizzly Kryonaut. Cheap insurance against the next domain failing the same way. Don't skip this just because the chips you're replacing pads on are the ones that died — the chips that were *still working* are the ones now overdue for paste service.

Verify on a single-board test bench before reinstalling. A KS-series universal hashboard tester (Zeus Mining sells one) lets you bring a single board up outside the chassis and confirm hashrate, then sweep with the thermal camera again to verify the previously-cold cluster is now warm. Catches re-failures *before* full chassis reassembly.

Post-repair burn-in. Run the miner at nominal hashrate for `24 h` minimum. Watch web UI hashrate at the `1 h`, `4 h`, and `24 h` marks. Re-sweep with the thermal camera. The previously cold cluster should now be at parity with neighbours. If it goes cold again within `24 h`, root cause was upstream (boost / chip-side short / cap drift) and you missed it — back to step 6 or step 8.

Stop DIY when: multiple non-adjacent LDOs have failed and reflow does not recover any of them; replacement LDOs die within 30 days (something upstream is cooking them); board has visible PCB damage requiring jumper-wire repair; chip-side internal short is suspected without bench instrumentation to find it; capacitor bulging or scorched-component odor; or you are past `8 h` total bench time on a single board and not closer to isolation than when you started. Book a D-Central repair.

Ship safely. Each hashboard in an anti-static bag. Double-box with `≥5 cm` of foam every side. Include a note: model, hardware revision label off the chassis sticker, observed symptoms, firmware version, what you tried. **A photo of the cold cluster from your thermal camera, attached to the shipment, cuts our diagnostic time by hours** — saved hours = saved CAD on your invoice. Saves us bench time, saves you money.

D-Central bench process: programmable bench supply with current logging, Zeus-class KS-universal hashboard tester for isolated bring-up, microscope + hot-air rework, salvage-grade LDO and tantalum stock, post-repair `24 h` burn-in at nameplate hashrate. We diagnose and repair every LDO / boost / passive failure on the board. We don't replace ASIC chips on KS hardware (chip + IceRiver-flashed firmware blob requirement makes that uneconomical for retail volumes) — but we will quote a salvage-grade hashboard swap if the chip array itself is the failure point.

If chip-side damage is the failure point, we offer a salvage-grade hashboard swap rather than chip-level repair. We will be transparent at quote time about which side of the line your board sits on. Salvage-grade boards are stock-dependent — confirm availability at quote. Replacement boards come with the same `24 h` post-install burn-in process before they ship back.