Canaan Avalon hashboards repair differently from Antminer boards, and the difference is structural. Avalon centres each board on a module-manager (MM) controller, talks to it over I²C or UART instead of a Zynq daisy-chain, addresses each ASIC individually, and exposes the deepest network-readable telemetry in mining. Diagnose over the API before you open the case; BGA-level work belongs on a bench.

How an Avalon hash board is built (A12xx–A15xx)

Where a Bitmain board is a long UART daisy-chain of identical chips hung off a Zynq control board, the Avalon family is organised around the MM (mining-module) controller. The host — historically a Raspberry Pi paired with an AUC (AvalonMiner Userspace Console / USB-to-I²C converter), and on the modern industrial units an integrated RISC-V SoC controller — does the network and Stratum plumbing. The MM near the boards keeps the job, rolls the extranonce, feeds work to the ASIC matrix, and reports health back up. The notes below are grounded in Canaan’s own open-source cgminer fork and public API manuals, not guesswork.

The silicon and controller generation differ across the lineup:

• 12-series (AvalonMiner 1246, 1166 Pro, 1146 Pro, 1126 Pro, 1066) — Canaan’s earlier 7 nm-class ASICs on the MM3-generation controller.
• 13-series (A1346 / A1366) — the A3200C-Plus-class die, MM317-generation firmware, air-cooled.
• 14-series (A14x, A1466, A1466HS, A14xI) — the refined 7 nm A3198S, MM318-generation firmware, in both air and liquid-cooled SKUs.
• 15-series (A15x, A1566HA/I, A1566HS) — the A3197S (5 nm-class; Canaan does not officially publish the exact node), MM319-generation firmware, air and liquid-cooled.

Three architectural facts matter for repair. First, the ASIC bus is per-chip addressable — the MM selects a chip through a hardware mux (broadcast included) rather than passing every signal through every chip in series. Second, the chips are still wired in series voltage strings electrically: core voltage is regulated per string / per board domain, not per individual chip core, with only a small per-chip voltage trim baked into each die’s eFuse at the factory. Third, every chip carries an on-die PVT (Process-Voltage-Temperature) sensor, so the board can report temperature and voltage for each chip individually — telemetry an Antminer simply does not produce.

Diagnosing a dead or weak board — software first

The Avalon way is to interrogate the board over the network before touching a probe. The controller runs a cgminer-style API on TCP port 4028; the estats report (or the slim litestats on the home line) is the richest diagnostic surface in the industry. Read these fields:

• TA — total active chips. Compare against the board’s nameplate matrix. A shortfall means dead or skipped chips.
• GHSmm vs GHSavg — theoretical versus measured hashrate. A board whose measured rate trails its theoretical rate is carrying weak chips or thermal throttling.
• MGHS — per-board hashrate. On a three-board miner this isolates which board is down.
• PVT_T / PVT_V arrays — per-chip temperature and voltage. This is how you find the exact cold (dead) or hot (shorting) chip without a thermal camera.
• ECHU and ECMM — per-board UART error counts and MM error counts. Climbing error counters point at a marginal bus, connector, or MM, not a dead chip.
• SF0–SF3 — the smart-speed frequency targets the MM is currently running.
• PS[…] — PSU status. Sending ascset|0,hashpower returns the supply’s error flag, control voltage, hash voltage, output current, output power, and set voltage in one line.

The single biggest behavioural difference from Antminer lives here. Because the bus is per-chip addressable, a chip that merely stops responding on the data bus is skipped — its neighbours keep being polled and keep mining. You lose that chip’s share of the hashrate, not the whole board. That is a real reliability advantage of Canaan’s design. But be honest about the limit: a chip that fails shorted, or a failed string regulator, still collapses the series voltage domain it sits in, and that can take a board offline electrically. So “one bad chip kills the board” is half-true for Avalon — false for a data-dead chip, true for a power-fault chip.

At the bench the physical checks are the universal ones: measure PSU input and the board’s incoming DC rail, measure string/domain voltages against a known-good reference board, and use thermal imaging under a short powered test to spot a cold (dead) or hot (shorting) device. One honest caveat: the per-pin diode and voltage pass/fail tables in our bench library are Bitmain AMTC reference data — Canaan does not publish an equivalent per-pin Avalon table, and the deep per-model tuning data is encrypted and eFuse-locked. So Avalon component diagnosis leans harder on comparative measurement against a reference board, thermal imaging, and the chip’s own PVT readout than on a published number for every pad.

Common Avalon failure modes

• Weak or dead ASIC. A data-dead chip drops out of the TA count and you lose its hashrate; a shorted chip pulls its voltage string down and shows as a hot spot or a sagging PVT_V. The PVT arrays usually name the offender before you open the case.
• String / regulator fault. The per-board core-voltage regulation for a string fails open (chips in that string go dark) or shorted (string voltage collapses, current spikes). This is the classic “whole board dead” electrical failure.
• MM / controller and AUC link fault. If an entire board goes missing rather than losing chips, suspect the MM, its power, or the I²C/AUC link and ribbon — not the ASICs. Climbing ECMM with no chip loss points the same way.
• PSU and connectors. Avalon’s high-current bolt-on DC connectors and the supply itself are common culprits; the PSU error bitfield flags input under-voltage, over-temperature (OT1/2/3), over-current, output under-voltage, and fan failure. A loose or arc-pitted power lug mimics a board fault.
• Water and corrosion (hydro / immersion). The A1466HS, A1566HS and immersion “I” variants add coolant paths and cold plates. Coolant ingress, galvanic corrosion at the cold-plate interface, a weeping fitting, or contaminated immersion fluid all produce intermittent or dead boards — and corrosion repair starts with a full clean and dry before any electrical conclusion is trustworthy.
• Thermal. Dried thermal interface, a clogged dust filter, or a failed fan throttles the board long before it dies — visible as elevated PVT_T and falling frequency.

Component-level repair reality — and where DIY ends

Reflow is the right first move when the symptom says joint, not silicon: a board that hashes cold and drops chips as it warms, or whose fault moves under gentle pressure, often has a cracked solder joint that a controlled hot-air reflow with proper flux — and fresh thermal interface on reassembly — can recover cleanly. That is achievable bench work if you have hot air, a preheater, and magnification.

Chip replacement is a different commitment. Avalon’s dense small-die matrix and the liquid-cooled cold-plate construction make these boards unforgiving to rework. Replacing a device means desoldering a fine-pitch ASIC, cleaning and re-tinning pads, fitting a replacement verified good on a chip tester first, and reflowing to a controlled profile — on a board you must then bring back up and re-validate. And here is the Avalon-specific wrinkle: the per-model voltage and frequency tables are AES-encrypted and eFuse-locked, with per-chip calibration read from each die’s OTP at boot. You cannot hand-tune a repaired Avalon board the way you might tinker elsewhere — it has to re-pass the MM’s own start-up validation and read back clean PVT and error counters, much as it would on Canaan’s factory test flow. That disciplined retest-from-the-top approach is how Canaan’s line keeps these boards consistent, and it is the standard a real repair should meet.

Stop and send it in when the fault is an MM, a string regulator, a BGA-level chip, or any liquid-cooled board with suspected corrosion — and when you don’t have hot air, a preheater, a microscope, a reference board, a chip tester, and a way to re-run the validation. For most owners that is the honest line: diagnose confidently over the API, recover the cheap stuff (supply, connectors, filters, thermal interface, a clean reflow), and bench the rest.

Get Avalon hashboard repair done right

D-Central has done component- and chip-level ASIC repair in-house in Laval since 2016, with the reflow stations, chip testers, reference boards, and parts to bring a dead or weak Avalon board back rather than scrapping it. Start a repair here: D-Central ASIC repair. If you’d rather replace a board outright or keep a known-good spare on the shelf while yours is on the bench, D-Central stocks replacement hashboards, ASIC chips, and control boards.

Before you order anything, confirm the symptom with the ASIC Fault Finder & error-code database and weigh the economics with our repair vs. replace guide. For the model you’re holding, the specs live on the profile pages — AvalonMiner 1246 (12-series), A1346 (13-series), A1466 (14-series), and A1566HA (15-series). This guide complements, and does not replace, the model-specific walkthroughs: AvalonMiner 1246 repair, Avalon A1346 repair, and Avalon A1466 repair — start there for board-fitment and teardown specifics, and come back here for the cross-family diagnostic logic.