IceRiver 1004LV100 ASIC Chip Replacement Guide

Note the per-board hashrate deficit. Compare against KS5L nameplate (~12 TH/s) or KS5M nameplate (~15 TH/s). Confirm the missing chunk lands on a chip-count integer multiple. Photograph the web UI per-board chip count and hashrate before any rework — having before-numbers on file is invaluable when you're 4 hours in and want to know if anything improved.

Visual sweep with the lid off. Photograph every hashboard from above and at 30 degrees. Look for scorched chip die, lifted heatsink, swollen tantalums adjacent to the suspect chip, or shifted thermal pads. Any of these changes the procedure from chip replacement to surrounding-damage investigation first.

Cold-boot the miner once at the breaker (30 seconds off). If hashrate climbs back to nameplate, the symptom was a boot glitch, not a chip-level failure — abandon this procedure and watch the miner for 7 days before any rework.

Thermal-camera sweep at full hashrate after 10-minute warmup. Mark cold-die positions on a printed board photo. Single-die cold spot = chip suspect. Four-die cold cluster = LDO suspect (head to the LDO voltage-domain page first). This is the most valuable Tier-2 step on this fault and resolves >50% of misdiagnosed chip-replacement calls.

Multimeter on the suspect chip's `0.45 V` rail to ground in resistance mode, miner powered off. Compare against a known-good chip on the same board. `<1 ohm` to ground = internal short = chip is dead. `OL` or `>1 kohm` = open — could be solder joint or chip, will need reflow to disambiguate. `~50-200 ohm` = chip is fine, look at upstream rail feed.

Probe the suspect chip's `0.8 V` and `1.8 V` rails for parity with neighbours. If they read low or dead, the upstream LDO is the real fault — fix that first via the LDO voltage-domain page or you're chasing the wrong failure. If rails are healthy and the chip is cold, the chip itself is the failure point.

Refresh thermal paste on the entire chip array if you're already in there for diagnostics — Arctic MX-6 or Thermal Grizzly Kryonaut, uniform thin layer. Dried paste contributes to the thermal cycling that fatigues BGA joints in the first place. Cheap insurance and the workshop is already open.

Reflow the suspect chip first before committing to replacement. Pull heatsink, clean residual paste with `99 %` IPA, flux the chip die generously with no-clean BGA flux. Bottom-side preheat to `150 C` on a hot plate or IR preheater. Once at temperature, top-side hot-air at `350-400 C` for `30-45 s` over the chip die. Watch for the chip to settle as solder reflows. Cool naturally on a 5-10 minute step-down.

Re-test rails after the board cools to room temperature. If `0.45 V` to ground now reads `~50-200 ohm` instead of `<1 ohm` or `OL`, the joints reseated and the chip is fine. Reassemble and burn in. Solder-joint fatigue is roughly 30-40% of failures and reflow recovers them. Schedule a 30-day follow-up; second failure inside 30 days is replacement territory.

Refresh thermal pads on the entire suspect domain — `1.5-2.0 mm` thickness, `~6 W/m K` thermal pad. Dried-out pads are the upstream cause of BGA-fatigue cracks. Refreshing them is the only way to prevent the next failure on the same chip. Skip this step and you're rebooking the same repair in 6 months.

If reflow doesn't recover the chip, stop here unless you have full BGA-rework gear on hand (Tier 4: bottom-side preheater, `350-400 C` hot-air station with focused BGA nozzle, stereo microscope, BGA vacuum pickup, replacement chip in stock). The chip is replacement territory, and a hot-air-only attempt at chip removal will lift surrounding components and brick the board.

Chip removal. Bottom-side preheat to `170-180 C`. Top-side hot-air at `380-400 C` over the target chip with focused BGA nozzle, dwelling `45-60 s` until the solder under the chip is fully liquid. Lift with vacuum BGA tweezers in a single smooth motion — never pry. Lifting with mechanical force tears pads off the PCB and the board is finished. Cool the area to ~80 C before any further work.

Pad cleanup. Wick all residual solder off the BGA pads with desoldering braid + flux. Inspect under microscope at `30-40x` for: lifted pads (board is dead), tombstoned passives nearby (reflow them back), or solder-mask damage (repairable with green solder mask if minor). Clean the pad array with `99 %` IPA and a soft brush.

Solder paste application. Apply a thin uniform layer of `Sn63Pb37` leaded paste — it melts at `~183 C`, easier to manage by hand than RoHS lead-free at `217 C`. Stencil application is ideal; if no stencil, a hand-paste-and-tack-reflow at `~200 C` to convert paste to balls is acceptable but needs careful microscope work. Do not mix leaded and lead-free solder anywhere on the board.

Chip placement and reflow. Drop the replacement `1004LV100` onto the prepped pads. Package self-aligns within `~0.1 mm` if the paste is correctly placed — solder surface tension does the work. Bottom-side preheat to `150 C`, top-side hot-air at `350-400 C` for `45-60 s` over the chip die. Watch for the visible settle (chip drops `~0.2 mm` as BGA reflows). Cool naturally on a 5-10 minute step-down. Inspect under microscope: every solder ball should be a flat-bottomed dome, no bridging visible at chip edge.

Rail re-test before reassembly. Multimeter on the new chip's `0.45 V` rail to ground: should read `~50-200 ohm` cold. `1.8 V` and `0.8 V` to ground: should read healthy. If `<1 ohm` short, you have either a solder bridge under the chip (microscope sweep, redo reflow) or — rarely — a DOA replacement chip (try a second chip from the spare batch you ordered).

Bring-up on a KS-series tester. Single-board bring-up off-chassis using a KS-series universal hashboard tester before reinstalling. Confirms Chip Count is full, hashrate is at nameplate, and no domain-level failures emerged during rework. Skip this step and you'll find out the hard way after chassis reassembly.

Post-repair burn-in. Run miner at nominal hashrate for 24 hours minimum. Re-sweep thermal camera at the 1h, 4h, and 24h marks. The previously cold die should be at temperature parity with neighbours. If it goes cold again within 24 hours, root cause was upstream — back to LDO / boost / PSU diagnostics, the chip is innocent.

Stop DIY: replacement chip dies within 24 hours, or a different chip on the same board fails within the same week, or you tear a pad on the PCB during chip removal, or visible PCB damage requires jumper-wire repair, or you're past 12 hours bench time on one board with no isolation progress. All ship-to-D-Central scenarios — book an ASIC Repair slot.

Ship safely if you're not doing the work. Anti-static bag per board, double-box with `>=5 cm` foam every side. Include: KS model, hardware-revision label, observed symptoms, firmware version, what you've already tried — especially any chip you reflowed or attempted to remove. A photo of the cold-die thermal sweep and a clear note saying chip-level work attempted on position X saves bench time and saves you CAD.