I2C Bus

The I2C bus (Inter-Integrated Circuit, pronounced "I-squared-C") is a two-wire serial bus that lets one controller talk to many low-speed peripheral chips over a shared pair of lines: a data line (SDA) and a clock line (SCL). Every device on the bus has a unique address, so a single controller can reach dozens of sensors and memory chips with just two signals and a couple of pull-up resistors. It is slow by modern standards and completely adequate for its job — reading a temperature, fetching a serial number — which is why it is everywhere in embedded hardware, mining machines very much included.

I2C inside an Antminer

On an Antminer, the Zynq control board is the bus master, and the I2C lines run to each hashboard through dedicated pins on the board-to-board data cable (pins 4 and 5 of the classic 18-pin connector, SDA and SCL at 3.3 V). Three kinds of devices answer on the hashboard side. First, the EEPROM — a small memory chip storing the board's identity: serial number, chip count, calibration data, voltage and frequency profiles. Second, the temperature sensors (parts like the LM75A, TMP1075, or NCT218, conventionally addressed in the 0x48–0x4F range) that feed thermal protection. Third, on models that have one, the PIC microcontroller that manages board power also communicates over I2C. When the control board boots, it reads each board's EEPROM over this bus first — if that read fails or returns invalid data, the firmware reports an EEPROM error and refuses to bring the board online, no matter how healthy its ASICs are.

Why it matters on the repair bench

A striking number of "dead hashboard" symptoms trace back to a broken I2C path rather than to the hashing silicon: a cracked solder joint on SDA or SCL, a damaged EEPROM, corroded connector pins, or a corrupted EEPROM image. The bus also has a characteristic failure mode worth knowing — because SDA and SCL are shared, open-drain lines, a single stuck device holding a line low silences every device on the bus, so one failed sensor can make EEPROM, temperature, and PIC all vanish at once. Diagnosis follows that logic: check the pull-up voltage on both lines (a line stuck at 0 V means something is clamping it), watch the traffic with an oscilloscope or logic analyzer, and isolate devices until the bus recovers. A falsely tripping thermal shutdown, likewise, is often a flaky sensor or stuck bus rather than real heat.

EEPROM repair, the routine case

The bus is not confined to hashboards, either. Modern Antminer power supplies are digitally managed over the same two wires: the control board sets output voltage and reads back status from an APW12-class PSU via an I2C-based protocol, which is how firmware adjusts voltage for different frequency targets without anyone touching a potentiometer. That makes I2C literacy doubly valuable on the bench — the same skills that recover a hashboard EEPROM let you interrogate a PSU that answers but refuses to regulate.

Because the EEPROM gates board recognition, re-writing a known-good image is one of the most routine board-level repairs. A corrupted image is fixed by reading the EEPROM from a working board of the same model and writing it to the failed one with an external programmer (the ubiquitous CH341A or a dedicated mining-repair tool); mismatched boards in one chassis are fixed by synchronizing all boards to the same configuration; a dead EEPROM IC is replaced outright and programmed with correct data. It is honest, unglamorous work — two wires, one small chip — and it revives boards that would otherwise be scrapped. I2C is one of several serial interfaces on the control board: for the boot console used alongside it in diagnostics, see UART, and for the chip mastering the bus, see Zynq SoC.