Antminer S9 – Control Board Not Booting

Tier 1 — Cut power at the wall or breaker (not just the miner's switch). Wait 30 seconds for PSU bulk caps to discharge. Power back on. This clears wedged driver state and transient conditions that masquerade as a dead CB. A surprising percentage of 'dead CB' tickets resolve here, especially after firmware updates or brown-outs. If the miner boots, monitor for 24 hours before trusting it — intermittent boot is usually the early stage of an underlying fault, not a fix.

Tier 1 — Power off, pop the lid, eject the microSD, blow the slot clean with canned air, and re-insert firmly until the click. Dust in the SD slot or a half-seated card is a stunning share of home-miner shipments arriving at our bench. While you are in there, verify the card is the right way up — the gold contacts face the board. Power back on, wait 3 minutes, and check for network / LED behaviour before proceeding.

Tier 1 — Re-seat each ribbon cable between the CB and the three hashboards. Power off first. Disconnect, inspect the mating surfaces for blackening, green oxidation, or bent pins, and re-connect firmly. A shorted or half-seated ribbon can hold the CB in reset or hang `bmminer` during hashboard init, which the watchdog then rebooting looks exactly like a dead CB. If any ribbon looks damaged, flag it — on-order replacement ribbons for S9 family are common.

Tier 1 — Re-seat the ethernet cable at both ends and try a different switch port. Verify link-LED behaviour on the CB RJ45. If no link LED comes up with a known-good cable into a known-good port, the CB's PHY is suspect (Tier 3+) rather than a boot issue. If link is healthy but the miner still doesn't answer DHCP, the CB boot is failing in kernel / userland — proceed down the chain.

Tier 1 — Multimeter the CR2032 coin cell on the CB. Target 2.9–3.2 V unloaded on a healthy cell; below 2.6 V is suspect. A dead cell alone does not always block boot, but combined with certain stock firmware builds the kernel can hang on `hwclock` sync when the RTC returns garbage on the I²C bus. Replace pre-emptively with a fresh CR2032 (CAD $2). Re-check voltage on the new cell — expect 3.0–3.3 V.

Tier 2 — Photograph the CB silkscreen near the SoC (BHB42631 / BHB42701 / BB-based / S9-V1-C1 variants). Download the silkscreen-matched stock image from support.bitmain.com/hc/en-us/articles/115000267373 — Bitmain indexes S9-family images by CB type, not by miner model. Format the microSD `FAT32` with SD Card Formatter (full format, not quick). Write the image with balenaEtcher, verify the write, re-insert, power on, wait 3 minutes. This alone resolves roughly half of dead-CB tickets in our bench queue.

Tier 2 — Swap to a known-good industrial-grade microSD. Even after a clean re-flash, the underlying card can be physically dead from worn TLC cells. Bench standard: SanDisk Industrial 16 GB or Transcend Industrial — avoid generic TLC consumer cards on 24/7 miners. Write the same silkscreen-matched image. If the miner boots on the new card, retire the old one permanently. This single upgrade removes the #1 cause of S9 CB failures on fleets that have been running since 2017–2018.

Tier 2 — Multimeter on DC. Probe the CB aux 12 V input connector while the PSU is powered and the CB is attempting to boot. Expect 11.8–12.4 V sustained. Below 11.5 V: PSU aux is sagging, the harness is damaged, or the PSU is tired. Swap the aux harness or the entire PSU with a known-good unit. A CB that only sometimes boots is almost always a drooping aux rail, not a dead SoC — fix the power side before chasing the firmware side.

Tier 2 — Verify the PSU's aux output separate from the main hashboard 12 V rails. On APW3 / APW3+ / APW3++ / APW5 the aux comes off a small auxiliary connector (distinct from the two 10-pin PCIe-style mains). Test continuity, inspect insulation for melted spots, swap with a known-good aux harness. Sister reference: antminer-s9-psu-not-detected walks the full PSU handshake chain in detail if the aux fault is deeper than the harness.

Tier 2 — If the SD is known-good (Tier 2 re-flash + swap complete) and the rails are healthy, check the Bitmain firmware-to-silkscreen mapping again. A BHB42701 image flashed onto a BHB42631 CB will hang boot every time. Download the exact image matching the silkscreen variant, re-flash, retry. If you are unsure which variant you have, photograph the silkscreen and cross-reference the Bitmain CB troubleshooting article — image filenames are NOT always obviously keyed to the silkscreen.

Tier 3 — Hook up a UART console. Wire a CP2102 or FT232RL USB-UART adapter to the 4-pin CB header: GND to GND, adapter-TX to CB-RX, adapter-RX to CB-TX, leave VCC disconnected (CB powers itself). Open `screen /dev/ttyUSB0 115200` or PuTTY at 115200 8N1. Power on, capture the full boot log. This is the biggest diagnostic unlock on a stuck S9 CB — you stop guessing and start reading exactly where boot fails. A CP2102 adapter costs CAD $10 and is reusable for every miner you will ever own.

Tier 3 — Match UART output to the failure mode. `mmc0: error -84` = SD dead (Tier 2 SD swap covers it). `VFS: Unable to mount root` = rootfs wrong or corrupt (re-flash with correct silkscreen-matched image). U-Boot reboot loop = DDR training or aux rail issue (re-check Step 8 under load, probe with a scope if you have one). `Kernel panic - not syncing: Attempted to kill init` = /sbin/init missing from image (re-flash). Boot completes but `bmminer` never starts = check /etc/init.d/bmminer or a hashboard ribbon short is hanging hashboard polling.

Tier 3 — Cross-flash a different known-good stock image. If Bitmain's current S9 image hangs on your CB but an older one works, you have a firmware-to-hardware mismatch. Our bench maintains a library of known-good S9 stock images by silkscreen variant and year — 2018–2019-era images boot cleanly on nearly every S9-family CB. Some late-rev BHB42701 boards dislike 2020+ builds. If an older image boots cleanly, stay on it rather than chasing the latest stock.

Tier 3 — Flash DCENT_OS (D-Central's open-source Antminer firmware — our Mining Hackers option) where the build tree supports S9-family hardware (verify current support at https://github.com/DCentralTech/DCENT_OS before flashing). Alternatives if DCENT_OS does not yet cover your S9 variant: Braiins OS+ for S9, or legacy Vnish / Asicseer community builds. All of them use a saner init sequence than late-2020+ Bitmain stock, and any of them will boot a physically-healthy S9 CB that stock will not. Landing: https://d-central.tech/dcent-os/.

Tier 3 — If UART shows `rtc0: read error` or `hwclock: can't open '/dev/rtc'`, the RTC crystal (32.768 kHz, next to the RTC IC or in the SoC package) likely has a cold solder joint after years of thermal cycling. Touch each foot with a clean 340 °C iron tip for one second per pad. Let cool, retest. If the RTC IC itself is dead, that is fine-pitch SMD rework and you are into Tier 4 territory. Flux, anti-static discipline, and a loupe are not optional.

Tier 4 — Stop DIY and book a D-Central repair slot when any of these is true: UART shows zero output despite verified aux rail and SD card (SoC-level fault); eMMC is dead on a late-rev S9 SE / S9k CB and you cannot force SD-boot (BGA169 replacement is bench-only); visible physical damage — burnt components, cracked SoC or DDR, lifted pads near RJ45, water intrusion; CB boots briefly then dies after a surge event (collateral damage across rails); three known-good silkscreen-matched images on industrial SDs all fail.

Tier 4 — D-Central bench process for a dead S9 CB: chassis isolation against a known-good bench chassis + known-good hashboards + bench PSU; UART + JTAG on the SoC; eMMC read / write via fixture where applicable; PHY and passive replacement; salvaged-grade S9-family CB swap from cold-storage inventory where a repair is not economical; 24-hour burn-in at nameplate before ship-back. We still maintain an S9 rework bench because the pleb-mining and Bitcoin-Space-Heater communities keep needing them.

Tier 4 — Ship the CB in an anti-static bag. Protect the RJ45. Include a paper note with the silkscreen variant, the last firmware image flashed, any UART output you captured, and the step number in this guide you reached. That paperwork shaves 30–60 minutes off our bench time per miner, which shows up directly on your invoice. Shipping accepted Canada-wide, US, and international — Canadian bench, cold-climate-tested, Bitcoin accepted on repairs where the invoice makes sense.