An Antminer S19-family hash board is one long series string of 76 to 126 ASICs grouped into voltage domains — so a single dead chip or one collapsed domain takes the whole board offline, never just part of it. This guide walks the S19, S19 Pro, S19j Pro and S19 XP board architecture, the domain voltage map, the PIC’s role, and how to tell a reflowable fault from a chip-replacement job before you commit a soldering iron to it.

It is meant to complement, not replace, D-Central’s broader Ultimate Guide to Troubleshooting the Antminer S19 Series — that page covers the whole machine and its symptoms; this one goes down to the board, the domains, and the chips.

How an S19-family hash board is built

A complete S19-class miner is a control board, three identical hash boards (Chain 0/1/2), an APW12 power supply, and fans. Each hash board is a self-contained SHA-256 engine: a daisy chain of ASICs that share a 25 MHz clock and a UART command bus, powered by a boost converter feeding a ladder of voltage domains. Lose any link in that chain and the controller stops seeing chips past the break.

Chips, domains and chip generation

The defining fact of these boards is that core voltage is regulated per domain, not per chip. The chips inside a domain are wired in series, so the domain voltage equals the per-chip voltage multiplied by the number of chips in that domain. There is no “partial board” mode — a board either enumerates its full chip count or it has a fault.

Board	ASIC (process)	Chips/board	Domains	Chips/domain	Domain V	Boost out
S19	BM1398 (7 nm)	76	38	2	~0.36 V	~19 V
S19 Pro	BM1398 (7 nm)	114	38	3	~0.32 V	~20 V
S19j Pro	BM1362 (5 nm)	126	42	3	~0.30 V	~19 V
S19 XP	BM1366 (5 nm)	110	11	10	~0.4 V	~19 V

Note the generational split: the base S19 and S19 Pro run the 7 nm BM1398, while the S19j Pro (BM1362) and S19 XP (BM1366) moved to 5 nm silicon with very different domain counts. That matters when you source replacement chips — a BM1398 die will not stand in for a BM1366.

The voltage path

Power flows the same way on every board: the APW12 PSU delivers 12–15 V into the board, a MOSFET switch (enabled by the PIC/controller) gates it, and a boost converter raises it to roughly 19–20 V. From there, per-domain LDOs step the boosted rail down to the 1.8 V I/O supply and the ~0.8 V core supply each chip group needs. The APW12 (12–15 V, ~233 A, 3600 W) is the standard S19 supply; the S19j XP variant ships with the higher-current APW17. If the boost output is absent, every domain downstream is dead and the board reports zero chips.

The PIC’s role — and why a dead PIC looks like a dead board

The S19 era is a PIC-managed generation. A PIC16F1704 (U3 on the S19, U6 on the S19 Pro) sits on the board as an I2C slave — addressed by chain number on the shared bus — and does three jobs: it enables the domain DC-DC converters, runs a heartbeat watchdog with the control board, and holds per-board calibration (frequency and voltage parameters). A crucial diagnostic consequence: the board’s temperature sensors (LM75A/TMP75) do not appear on the I2C bus until the PIC has enabled domain voltage. So a corrupted or unpowered PIC presents as “no temps, board won’t power up” — easily mistaken for a power-stage failure. Measure 3.3 V at the PIC’s output pin; if it is missing, the PIC needs reprogramming (PICkit3) before you chase anything downstream. (Later S19j Pro board revisions on Amlogic/Cvitek control electronics use a dsPIC variant, but the principle is the same.)

Control board variants

Early S19/S19 Pro use the Xilinx Zynq 7010 control board — dual ARM Cortex-A9 at 667 MHz with an Artix-7 FPGA that handles the time-critical ASIC UART, plus 256 MB NAND and DDR3. Later S19j/S19j Pro and S19 XP units shifted to BeagleBone (TI AM335x) and Amlogic (A113D) control boards that bit-bang the UART in software with no FPGA. The hash board itself is the same series-string design regardless of which brain drives it — but knowing which control board you have matters when you build or borrow a test fixture.

Diagnosing a dead or weak board

Read the controller first

Before opening anything, ask the miner what it sees. Over the web UI or the cgminer/bmminer API (devs / stats), each chain reports a detected ASIC count. A healthy S19 reads 76/76 per chain; a board that reads, say, 29/76 has a chain break immediately after the last detected chip (chip 30). A flat 0 on a chain points at the boost stage, the PIC, the crystal, or the edge level shifters — not a single mid-chain chip. The kernel log (dmesg) and per-chain error counts fill in the rest.

Unpowered checks: diode and resistance

With power disconnected and capacitors discharged, the multimeter tells you a lot. First, measure resistance across the board’s main power input — a near-zero reading means a short on the power bus (failed MOSFET, shorted boost, or a blown filter cap), and you must not power the board until you find it. Then use diode mode on the per-chip test points. Bitmain’s AMTC (Authorized Maintenance Training Center) publishes factory reference values for exactly this; here are the documented figures for the BM1398-based S19 and S19 Pro, measured with a Fluke 15B+:

Test point	Diode (Ω)	Voltage when powered (V)
BI/BO (bus)	1220 ±20	0
RST (reset)	980 ±20	1.7 ±0.1
RX/RI (UART in)	390 ±20	1.7 ±0.1
TX/CO (UART out)	1220 ±20	1.7 ±0.1
CLK	1220 ±20	0.7–0.9
LDO 1.8 V	440 ±20	—
LDO 0.8 V	20 ±5	—

A pin reading well outside these ranges flags a damaged chip or a broken signal at that point. Treat these as a reference, not an absolute pass/fail line — board batch variation and a different meter will shift the numbers. Safety rule: never let the black probe touch the heatsink; you will short a domain.

Powered domain voltage sweep

On a test fixture, let the board stabilize for a few minutes, then sweep each domain’s 1.8 V and 0.8 V test points and compare them against each other and a known-good board. Domains within roughly ±50 mV of one another are healthy. A domain reading ~100 mV low is a partial short — a failed chip or a leaky capacitor pulling the rail down. A domain reading high with no current is an open — a cracked joint or a broken trace. “If the voltage between voltage domains is abnormal, the entire hash board will not work,” and the cause is almost always a chip or the domain’s power-management part.

Locating the exact bad chip

Once the controller tells you the break is between chip M and M+1, the classic dichotomy (binary-search) method narrows it to the single chip: inject a valid signal at the chain midpoint and watch whether the detected count rises, then halve again. A thermal camera shortcut is just as telling — under a brief powered run, a cold chip among hot neighbors is dead (not drawing current), while a localized hot spot on a chip or a passive is a short. The S19 board, unlike the S21, has no cross-domain level shifters — only the U1/U2 shifters at the board edge translate the 3.3 V control signals to the chain’s 1.8 V — so a mid-chain break is the chip, its solder, or a domain-boundary resistor/coupling cap, not a shifter.

Common failure modes

• Dead chip (open): the chip stops forwarding clock/UART/reset; every chip downstream goes invisible. The most common chain-break cause, from ESD, thermal cycling, or a power event.
• Shorted chip: pulls a domain rail toward zero, the boost current-limits, and the whole board can read 0 chips. Shows as a low domain on the sweep and a hot spot on thermal.
• Broken domain / LDO fault: a shorted LDO kills its domain’s supply (and can damage chips); an open LDO or cracked solder leaves the domain unpowered. Check the domain’s filter caps for shorts first — they fail more often than the regulator.
• Boost / power-stage failure: no boost output means no domains; suspect the gating MOSFET, the boost IC, or the inductor. Replacement boost modules exist for exactly this.
• PSU / connector / cable: a worn 6-pin power connector, a damaged 18-pin signal ribbon, or a sagging APW12 produces intermittent or no communication. Rule these out before condemning silicon.
• Water and corrosion (hydro): the S19 XP Hydro and S19 Pro+ Hydro use water blocks; a coolant leak leaves green/white residue, eats pads and traces, and drives electrolytic corrosion under power. Inspect under magnification, clean with isopropyl, and ohm out every corroded trace for continuity before any power-on — corrosion damage is frequently worse beneath the solder mask than it looks on top.

Component-level repair reality

The honest dividing line is reflow versus chip replacement. Reflow is the high-reward, lower-risk fix: when symptoms point at a cracked or cold joint — a board that hashes cold and drops chips warm, or a count that changes under gentle pressure — a controlled hot-air reflow with proper flux can recover it. Chip replacement is real BGA-class work: desolder the failed QFN die at 350–380 °C, wick and re-tin the pads, fit a verified replacement (a salvaged or new chip should be proven on a matching BM1398/BM1362/BM1366 chip tester first), and reflow to a controlled profile. Reapply Fujipoly SPG-30B thermal gel on reassembly. Then re-run the full factory sequence: PT1 chip enumeration, PT2 pattern test, PT3 frequency sweep — a repaired board earns its way back from PT1, the same discipline Bitmain’s own repair line uses.

That tooling list — hot-air rework station, microscope, the correct chip tester, reballing supplies, a fixture and a 50 A bench supply for hash testing — is the line most owners should not cross. Stop and send it in if you do not have all of it, or if the fault is BGA-level, a short you cannot localize, multilayer trace damage, or hydro corrosion. There is no shame in it; this is bench work.

Where DIY ends — and where we pick it up

D-Central has done component- and chip-level ASIC repair in-house in Laval since 2016. If you have isolated the fault and just need a part, or you have reached the BGA line, we can take it from there — start a repair at D-Central ASIC repair. To confirm a symptom before you order anything, run it through the ASIC Fault Finder.

For parts, we stock the pieces these repairs need: replacement S19 hash boards, individual ASIC chips, control boards, APW12 power supplies, and the testers and tools that make this work possible. For the machine-level picture and per-model specs, see the profiles for the Antminer S19, S19 Pro, S19j Pro, S19 XP, and S19 XP Hydro.