Avalon A11&A12 Series Miner Disassembly Guide

Opening an Avalon A11/A12 chassis is nothing like opening an Antminer. No PIC, no FPGA, no shell to SSH into: the three hashboards plug straight into a single MM (Miner Manager) board, each board carries 120 ASIC chips wired as 40 groups of 3 in parallel with the groups chained in series off one board-level Vcore rail. That series topology is why careless teardowns are expensive: flex a board as you pull it and break a joint at a group boundary, and every group downstream drops out of the chain; the board comes back enumerating short. On an A1246–87 that is a third of the machine’s hashrate gone until the board is back from the bench.

Models covered

A1246–87/85/81, A1166 Pro–81/78/75/72/68, A1066 Pro–64-L/60-L/56-L — same chassis, same three-hashboard layout. The trailing digits are the factory hashrate bin, not a hardware revision: an A1166 Pro–68 is the 68 TH/s bin of the same board. One caveat: the A1066 Pro is an A10-series board — the chassis procedure is the same, the hashboard chip count is not.

The hardware you are about to expose

• Control board (MM): Canaan’s Gen-2 integrated controller — a Kendryte K210 RISC-V part running FreeRTOS out of on-chip SRAM. No Linux, no SSH, no shell: everything software can tell you comes off the CGMiner API on port 4028 or the web UI.
• MM → hashboard link: SPI at 1 MHz, 40-byte packets with a CRC16. Not UART, not I²C-to-a-PIC.
• Hashboards: three per chassis, 120 chips each (A32xx family). Rails are Vcore (variable), VTOP (0.75 V) and VDDIO (1.8 V). Voltage is commanded at the board/domain level by the MM’s P_SET_VOLTAGE packet (5.0–9.0 V, 125 mV steps) — there is no per-chip voltage control on this platform.
• Thermal sensing: two NTC 10K thermistors per hashboard (TH1/TH2). Tear a lead during a heatsink pull and the board reports garbage temps or drops out.
• Chain signals: CKin→CKout, Din→Dout, Rin→Rout, Cin→Cout, daisy-chained group to group — scope them when a board enumerates short.
• Fans: four 12 V units (Fan1/Fan2 on one panel, Fan3/Fan4 on the other) plus a temperature-control harness.

Do not open it until software has stopped talking

• Pull estats from port 4028: per-board temperatures, voltages, frequencies and fan speeds. That separates “one board missing” from “all three throttling” from “fan fault”.
• If one board is absent from estats and the other two are healthy, the fault is that board, its cable, or its MM slot — re-seat the board and its cable, and re-test it in a known-good slot on the same machine before you condemn the hashboard.
• Stop the boards hashing for a clean read with ascset|0,hashpower,0; identify the unit in a rack with ascset|0,led,1-1. Both are diagnostics, nothing more — dropping the hashboard rail does not make the machine safe to open. The controller and the whole PSU mains side stay live.
• Read the log before you count screws — AvalonMiner miner log technical guidance. If it will not power up at all, work the PSU first: Avalon miner power supply failures.

Tools and the one real hazard

A cordless driver with a PH2 hex-shank bit and a clutch you actually set — every fastener threads into thin sheet steel or an aluminium extrusion, and full torque strips them. A grounded anti-static wrist strap bonded to bench ground, ESD gloves and coat. Silicone thermal grease around 2 W/(m·K), isopropyl, lint-free wipes.

The shock hazard lives in exactly one place: the PSU. Its mains side holds ~400 V on the PFC capacitors after unplug — leave it sealed. The 12 V side and the hashboard rails are not a shock risk to you; they are a short risk to the board. A dropped screwdriver across a busbar costs you a hashboard, not your heartbeat.

Teardown, in order

1. Power down. Stop the miner, unplug mains at the PSU inlet, and let the machine sit — the PSU holds ~400 V on its PFC capacitors after unplug. No API command substitutes for pulling the plug.
2. Lid. Disconnect the Fan3 and Fan4 power leads first, remove the two lid screws, then slide the lid off. Those leads run down to the MM board underneath — lift the lid with them still seated and you rip connector housings off their headers.
3. Hashboard–to–MM connectors. Unplug every connector between the three hashboards and the MM board. Pull the housing, never the wire.
4. MM harness. Unplug the small white power connector, the large white temperature-control connector, and the black fan-control connectors. Fan1 and Fan2 latch — press and hold, then pull.
5. MM board. Two screws. Lift it straight out onto an ESD mat.
6. Aluminium strips and panels. Ten screws hold the two strips. Then take out the twelve front and rear panel screws in a diagonal pattern so the frame does not rack and bind on the boards.
7. Fan1/Fan2 panel. The temperature-control cable routes through this panel. Feed it back through the opening by hand; never let the panel hang on it.
8. Hashboards. Hold the frame, push each board in the release direction and draw it out one at a time, supported along its length. Do not lever it out by a heatsink — a 120-chip series chain has no tolerance for a flexed substrate.
9. Heatsinks (board-level work only). Eighteen screws in total release the heatsinks on both sides of the board; the board carries three heatsinks at the rear and one at the front. Back every screw out before you lift — lift, do not twist. If a sink is welded down by baked compound, warm it rather than shearing chips off their pads.

Reassembly — where boards get killed

• Front heatsink alignment. Two slots on its sides locate it. A millimetre off and the screws will still start — straight onto chip corners.
• Thermal compound. Clean every die, then lay ~2 W/(m·K) silicone grease evenly across the chips. Thicker is not cooler; an over-greased board runs hotter. Snug the screws cross-pattern, clutch low — over-torqued heatsink screws crack packages.
• PSU refit. The frame hooks mate with the openings in the PSU base. Align, press the clip down, push it home until it clicks — if it will not click, it is not aligned. Do not force it.

Verify before the lid closes

1. Power up on the bench, lid off. Query estats: all three boards must enumerate and report plausible TH1/TH2 temperatures. An absurd or missing board temperature is usually a pinched NTC lead from the heatsink refit, not a dead chip.
2. Confirm all four fans spin and report RPM, then hash 30 minutes and re-read estats: stable frequencies, no board dropping out or sitting cold.

Common mistakes

• Importing an Antminer procedure. No PIC and no I²C hashboard-management path here — voltage is commanded by the MM over SPI, not through a chain-level microcontroller. Those workflows do not transfer.
• Reusing baked compound. If you broke the thermal joint, you replace the grease. Every time.

Related

• ASIC Fault Finder — work back from the symptom first.
• How to remove the hashboard from an ASIC miner
• ASIC repair parts — fans, connectors, thermal compound.
• Start a repair — send us the board if the chain is broken.