ASIC Hashboard Repair — How Bitcoin Miner Hashboards Fail and Get Fixed

Every Bitcoin miner hash board fails in the same handful of ways. The ASIC chips sit in series-strung voltage domains, so power is regulated per domain — not per chip — and a single dead chip blacks out its whole board. This hub explains that universal architecture, the diagnostic ladder behind every repair, the four failure modes, and exactly where DIY ends and the bench begins.

D-Central has repaired ASIC mining hardware in-house in Laval, Quebec since 2016. This page is the parent of our hashboard-repair series: read it to understand how a hash board works and fails, then jump to the per-family guide for your miner.

The universal hashboard architecture

A complete Antminer-class miner is three hash boards, one control board, one PSU and fans. Each hash board carries dozens to a hundred-plus SHA-256 ASICs daisy-chained together. Those chips are grouped into voltage domains, and within a domain the chips are wired in series for core voltage. That single design fact drives almost everything about diagnosis:

• Voltage is per-domain, not per-chip. A domain’s regulated rail equals the per-chip core voltage multiplied by the number of chips in the string. Measuring “the voltage on chip 7” is meaningless in isolation — you measure the domain.
• One dead chip kills the whole board. There is no partial-board or limp-home mode. An open chip breaks the series string and the signal chain at the same time, so every chip downstream goes dark and the board reports far fewer ASICs than it should.
• Domain voltage is the master health signal. A domain that sags ~100 mV below its neighbours points to a partial short (a failed chip or blown cap); a domain reading high points to an open (cracked joint or broken trace).

The power path is the same across generations: the PSU feeds 12–15 V in, a boost stage lifts it to roughly 19–25 V, and per-domain LDOs (and on newer boards, MP2019-class buck converters on the top domains) step that down to the precise rail each chip string needs. Chips-per-board and domain voltage vary by chip generation:

The S21 Pro (BM1370, 5 nm) runs 65 chips per board. Whatsminer and Avalon use the same series-string-domain principle but different domain counts, boost topologies and connectors — the per-family guides below cover those specifics. The takeaway is universal: more, smaller domains on the S19 era (38 domains of 2–3 chips); fewer, larger domains on the S21 era (12–13 domains of 7–9 chips).

The control-board-to-board signal path

The control board talks to each hash board over an 18-pin flat ribbon carrying 3.3 V logic, ground, an I2C bus (EEPROM + temperature sensors), and the signal chain. Five signals run the chain: a 25 MHz CLK from the on-board crystal (Y1), CI/CO (commands, flowing chip 1→N), RI/RO (nonce responses, flowing N→1), RST, and BI/BO busy flow-control. Commands ripple forward down the daisy chain; responses ripple back.

On Bitmain’s common control board, a Xilinx Zynq 7010 (dual ARM Cortex-A9 at 667 MHz) pairs with an Artix-7 FPGA — and it is the FPGA, not the CPU, that drives the UART chain at hardware speed. S17/S19-era boards also carry a small PIC microcontroller (PIC16F1704) holding per-board calibration; the S21/T21 generation dropped the PIC entirely (no-PIC). Newer boards add 11–12 level shifters at the domain boundaries to translate signal levels as the chain crosses domains.

Chain-break behaviour is the single most useful diagnostic tell. When the controller enumerates only 29 of 108 chips, the break is right after chip 29 — a dead, shorted, or signal-forwarding chip there makes everything downstream invisible. The reported chip count per chain is your starting coordinate for every fault hunt.

The universal diagnostic approach

Good hashboard repair is a ladder. You climb it in order, cheapest and safest first:

1. Visual inspection. Under a loupe or USB microscope, look for burnt or discoloured components, a cracked crystal, corrosion or water staining, stray solder balls bridging pads, and heatsink/thermal-paste condition. Many “dead” boards are solved here.
2. Unpowered multimeter and diode checks. Measure domain impedance (consistent across domains is healthy; very low = short, very high = open) and power-input resistance (near-zero = a short on the power bus). Then use diode mode on the per-chip signal test points and compare to factory reference values. Never let the black probe touch the heatsink — that shorts the board.
3. Powered diagnostics on a test fixture. A fixture brings the board up safely so you can sweep every domain voltage (all within ±50 mV of each other = healthy; 100 mV+ low = partial short; high = open), verify the boost output, probe the signal chain, and run chip enumeration.
4. Thermal imaging. A cold chip among hot neighbours is a dead chip; a hot spot on one chip or a passive is a short or a failed regulator. See our thermal-imaging short-circuit guide.
5. Binary-search fault isolation. The dichotomy method injects a valid signal at the chain midpoint to halve the search space each pass until the exact failed chip is pinned.

Those factory-calibrated pass/fail numbers are real bench knowledge. As an example, the diode-mode reference values Bitmain’s authorized training centers (AMTC) use for the S19/BM1398 chip:

Values are board-batch and meter dependent (these were taken on a Fluke 15B+), so treat them as reference, not gospel. The factory PT1 (chip enumeration), PT2 (pattern) and PT3 (frequency-sweep) test ladder is the standard every repaired board should pass before it ships — credit to Bitmain’s AMTC method, which remains the benchmark for verifying a fixed board hashes correctly across its full frequency range. For deeper technique, see our companion deep-dives on voltage-domain measurement and hashboard testers, plus the chip-level rework deep-dive.

The four universal failure modes

Strip away the model differences and almost every hash board that lands on our bench fits one of four buckets:

1. Dead or shorted ASIC chip. An open chip breaks the series string and the signal chain — downstream chips vanish and the board reports a short count. A shorted chip pulls its domain voltage down and can read as 0 chips. Either way, there is no partial-board mode: one bad chip = the whole board offline until it is replaced.
2. Broken trace or cold solder joint. Thermal cycling and vibration crack BGA solder balls and the tiny domain-boundary resistors (CLK, CI, BO). Symptoms are intermittent dropouts, a chain that breaks at a domain boundary, or a board that works cold and fails hot.
3. Bad voltage domain / power stage. A shorted or open LDO, a failed MP2019-class buck on a top domain, a blown filter capacitor, or a dead boost stage (MOSFET/inductor) takes a domain — and usually the whole board — out of regulation. Abnormal domain voltage is the fingerprint.
4. PSU, connector, or corrosion. A failing APW PSU, a worn 18-pin signal connector, or oxidised power lugs mimic board faults. On hydro boards add water ingress, coolant leaks and corrosion under the cold plate — always inspect for moisture before powering a wet-cooled board.

What you can do at home vs. what needs the bench

Be honest about the toolset each job needs. Diagnosis is widely DIY-able; component-level repair mostly is not.

DIY-fixable

• Visual inspection, reseating or swapping the signal ribbon and power cables, and testing a suspect PSU against a known-good one.
• Reading the controller’s per-chain chip count to confirm which board is bad and roughly where the chain breaks.
• Unpowered domain-impedance and power-input resistance checks with a multimeter to spot a hard short.
• Cleaning dust, checking fans and thermal paste, and re-running the miner’s self-test.

Send it to the bench

• ASIC chip replacement — a BGA part needing a reball, hot-air rework on a controlled profile, alignment, and a known-good reference board to verify against.
• Boost, LDO or buck-converter replacement and short-finding that current-limits the board (thermal camera + milliohm meter work, not multimeter guessing).
• EEPROM/PIC reprogramming, trace repair, and full PT1→PT3 re-verification after any fix.

Not sure whether a board is worth fixing at all? Our repair-vs-replace guide and power-profiles database help you weigh the call before you spend on parts.

Per-family hashboard repair guides

Pick your miner family for the board-specific domain map, test points and failure patterns. Each guide complements the model’s general repair service page rather than repeating it:

To match an exact error state or fault code to a board area, cross-check the ASIC Fault Finder database.

Parts, testers, and when to send it in

We are Bitcoin mining hackers, not a swap-and-bill shop — we find the failed component, replace it, and re-test the board through the full PT ladder before it ships back. When a repair calls for parts, we stock the real things: replacement hash boards, ASIC chips, control boards, hashboard testers, and the wider mining hardware parts catalog.

If your diagnosis stops at “a chip is dead” or “a domain won’t regulate,” that is the line where DIY ends and microsoldering begins. D-Central repairs hash boards in-house — send us the symptom and the chip count, and we will tell you whether it is a connector reseat or a bench job.