An Antminer S21-family hashboard is a series string of BM1368 or BM1370 5nm ASICs grouped into voltage domains, fed by a boost circuit and per-domain regulators, and driven by a no-PIC Amlogic control board over a software UART. Diagnose it by mapping domain voltages and the signal chain, then deciding chip-level rework versus a bench job.

This guide is the board-level deep dive that sits behind our general Antminer S21 repair walkthrough. The general page covers symptoms and recovery steps; this page goes inside the board — domains, test points, the no-PIC signal path, and the honest line where a multimeter stops and hot air begins. D-Central has repaired ASIC hashboards in-house in Laval since 2016. We respect Bitmain’s factory test method (the AMTC PT1/PT2/PT3 line and its diode reference tables) and build on it with bench experience, never reprint it.

S21-family board architecture

Every S21-class hashboard follows the same chain: the APW17-series PSU feeds 12–15 V into the board, a boost converter raises it to ~21–25 V, and per-domain low-dropout regulators (LDOs) and buck converters step that down to the rails each chip group needs. The ASICs are wired in series for core voltage inside each domain, so the domain voltage equals the per-chip core voltage multiplied by the number of chips in series. This is the single most important fact in S21 repair: voltage is per-domain, not per-chip, and because the chips form one daisy chain, a single dead chip takes the whole board offline. There is no partial-board mode.

Model	ASIC (5nm)	Chips / board	Domains	Chips / domain	~Domain V	~Boost out
S21 / T21	BM1368	108	12	9	~1.2 V	~25 V
S21 Pro	BM1370	65	varies	varies	~1.0 V	~21 V
S21 XP	BM1370	91	13	7	~1.04 V	~21 V
S21 Hydro / Immersion	BM1368	varies	varies	~1.0 V	~21 V

On the S21, domains 1–10 are served by three LDOs each (one producing the 1.2 V IO rail, two producing the 0.8 V core rail), while the two highest domains run from MP2019-class buck converters (U166, U200) that drop the boost to ~2 V before the domain LDOs. The S21 XP uses the same idea with twelve level shifters (U1–U12) and its top domains fed from MP2019 converters at ~2.5 V. Board housekeeping is an EEPROM (U6 on the S21) that stores the board identity and calibration, a 25 MHz crystal (Y1) that seeds the CLK chain, and inlet/outlet temperature sensors on the I2C bus.

How the no-PIC architecture changes diagnosis

The biggest difference from the S19 era is that the S21 generation removed the PIC microcontroller. On S17/S19 boards a PIC16F1704-class chip gatekept the hashboard: it switched the DC-DC converters on, held a heartbeat watchdog, and a missing 3.3 V on the PIC output pin was a classic “zero chips” cause. BM1368/BM1370 boards have no PIC. Instead the control board manages voltage enable and the core-voltage setpoint directly over I2C, using a DAC the Amlogic platform repurposes from its audio silicon (the A113D started life as an audio SoC). So when you troubleshoot an S21 board, do not go hunting for a PIC rail that does not exist — check the I2C voltage-control path and the boost stage instead.

The control board itself is also different. S19-era boards used a Xilinx Zynq (dual Cortex-A9 at 667 MHz) with an FPGA handling ASIC UART at hardware speed. The S21 family uses an Amlogic quad Cortex-A53 board with software UART straight to each hashboard — one serial port and a reset GPIO per board, no FPGA in the path and no SD-card slot. For the technician this means the control board talks to the chain in software, and a board that enumerates fewer chips than expected is almost always a hashboard-side break, not a controller quirk.

Domain map and test points

Each chip exposes five signal test points plus its two power rails on the back of the PCB. Compare every reading against a known-good board of the same model — these are reference levels, not absolute pass/fail.

Test point	Signal	Expected (S21, powered)
CLK	CLKO — clock output	0.58–0.6 V
CO	Command out (UART TX forwarded)	~1.1 V
RI	Response in (UART RX from downstream)	~1.1 V
BO	Busy out (flow control)	~0 V (pulse)
NRSTO	Reset out to next chip	~1.2 V
VDD1V2	1.2 V domain IO rail	1.2 V
VDD0V8	0.8 V core rail	0.8 V

Board-level reference points: VDD_IN at the power connector is 12–15 V; the boost output is ~25 V (S21) or ~21 V (S21 Pro/XP); the 3.3 V logic rail arrives from the control board on the signal connector. One honest caveat on diode-mode testing: Bitmain’s AMTC published per-pin diode-resistance and voltage tables for the S17 and S19 families (measured on a Fluke 15B+), but there is no equivalent public per-pin table for the BM1368/BM1370 S21 boards. For S21-era work we read diode mode and compare to a reference board of the same model rather than to a fixed published number — do not borrow the S19 figures, they are different silicon.

Diagnosing a dead or weak board

Work from cheap and safe to powered and invasive.

1. Visual, under magnification. Look for burnt or discolored LDOs and buck converters, lifted or cracked domain-boundary resistors and CLK coupling caps, a cracked crystal, corrosion, and foreign solder balls bridging pads. On hydro and immersion boards, inspect for coolant residue and white corrosion around the cold plate and connectors.
2. Unpowered multimeter checks. Disconnect power, wait 30 seconds for the caps to discharge, then measure resistance from board ground to each domain rail and across the power input. Near-zero across the input bus means a short on the power MOS, boost, or a filter cap. A single domain reading far lower than its neighbors is a partial short — a failed chip or a blown decoupling cap. Never let the black probe touch the heatsink; you will short a domain.
3. Powered domain sweep (test fixture). On a fixture, power up and let it stabilize, then read each domain. All domains within ~50 mV of each other is healthy; a domain 100 mV+ low is a short, and a domain reading high is an open (broken trace or cracked joint). Abnormal voltage between domains is the classic signature of a damaged chip or a failed domain regulator, and it will stop the whole board.
4. Chip enumeration (PT1). Run the fixture’s enumeration test. If it reports, say, 29 of 108 chips, the break is at the chip immediately after the last detected one. Bitmain’s official S21 jig needs the B047 FPGA image and a V2.1/V2.3 control board; a salvaged S21 control board makes a serviceable DIY fixture. The fault display reads “ASIC NG: X.”
5. Binary-search isolation. To pinpoint the exact bad chip, inject a valid response by shorting the RO test point to the domain rail at the chain midpoint and re-enumerate: more chips counted means the fault is downstream, the same or fewer means it is upstream. Halve again until you land on the chip.
6. Thermal imaging. Under brief powered load, a cold chip among hot neighbors is dead; a single hot spot is a short drawing excess current; a hot passive is a failed LDO or cap; a whole cold domain is a domain that never came up.

For a guided, symptom-first version of this same logic, point owners at our ASIC fault finder.

Common failure modes

• Dead or open chip. The chain breaks at chip N; everything downstream goes invisible because CLK, CO, BO and reset stop forwarding. A chip can also be alive but fail to forward signals on a cold CO/CLKO joint — it presents identically.
• Shorted chip / bad domain. Pulls the domain rail down, the boost current-limits, and the board can read zero chips. Usually traced to one chip or that domain’s regulator.
• Broken trace or lifted passive. A cracked domain-boundary resistor or a failed 100 nF CLK coupling cap breaks the chain at a domain edge — the binary search walks you to it.
• Boost / DC-DC failure. No boost output means no domain voltages and zero chips; suspect the boost MOSFET, inductor, or IC. A dead top-domain buck (MP2019-class) kills only the high-voltage domains.
• PSU and connector. The S21 uses the APW17 (171215-series) PSU — ~12–15 V, up to ~267 A, with a different output connector than the S19’s APW12, so they are not interchangeable. Check the input voltage at the board before condemning the board itself; worn 18-pin signal cables and bent connector pins cause intermittent or no communication.
• Water and corrosion (hydro/immersion). Coolant ingress, leaking cold plates, and dielectric-fluid residue cause creeping corrosion and intermittent domain faults. These boards demand careful cleaning and full inspection before any rework.

Component-level repair reality

Some S21 repairs are squarely DIY for a hardware-fluent owner: re-seating or replacing the 18-pin signal cable, swapping a worn PSU, reflowing an obvious cold joint on a domain-boundary resistor, replacing a shorted filter cap, or cleaning corrosion. A failed LDO or boost stage is replaceable if you can drive a hot-air station — pre-built boost modules exist for the S19/S21 family, and individual 0.8 V/1.2 V LDOs are stock parts.

Where DIY ends is the BGA. Replacing a BM1368 or BM1370 means hot air at a controlled profile (roughly 350–380 °C), reballing salvaged chips with 0.4 mm solder, a microscope, and ideally a standalone BM1368/BM1370 chip tester to verify the replacement before it goes on the board. After any chip work the board must re-run the full PT1→PT2→PT3 sequence from the start, with the heatsink and fan fitted for the frequency sweep. Reballing 5nm silicon, chasing an intermittent fault under load, or rebuilding a shorted power stage is bench work — without the right tooling you will spend more in time and risk than the board is worth. The honest rule: if you are reaching for hot air on the ASICs, that is usually the moment to send it in.

Parts, repair, and where to go next

D-Central repairs S21-family hashboards in-house at our Laval shop — if you are past the easy fixes, start a repair at d-central.tech/asic-repair and we will diagnose it properly. If you are doing your own bench work, we stock the parts: replacement hashboards, ASIC chips, control boards, and the wider repair parts catalog. For diagnosis, our hashboard testers — the ARC Antminer tester and the STASIC MultiTester Pro — take the guesswork out of chip-level fault location.

Match this guide to your exact unit with the model profiles for the Antminer S21, S21 Pro, and S21 XP, and pair it with the symptom-led Antminer S21 repair guide. We stand on the shoulders of Bitmain’s factory test method and the open-source mining community — our job is to keep these boards hashing honestly, and to tell you plainly when a board is better sent in than soldiered on.