Several weeks ago, a friend e-mailed me asking for help building a bitcoin miner based on Bitmine A1 ASIC – a mighty chip capable of 40GH/sec and also very DIY friendly. Last Sunday my friend showed up in the morning carrying a box of parts and in the evening we had semi-functioning board and zero casualties. In this article I’m writing my notes hoping that other builders following the same path may find them useful. As soon as we get working board I’ll build another one and post the real build log.
Note 1: The board we were building is a reference design by Bitmine.ch, a company that designed A1 ASIC. The reference board documentation is inconsistent; the rev.1.0.A schematic is different from rev.1.0.B Gerbers. Several part designators won’t match the PCB silkscreen, and the 500 ohm R12 resistor, likely added to improve stability of 2-phase buck converter, was not present on the schematic and/or BOM; we finally managed to figure out what it is by studying the board’s Pick-and-Place job file.
Note 2: Gerber file of a paste layer (the one used to cut a stencil) has openings for the thermal pads and power rails that are too large, using them as-is caused too much paste to be dispensed. Ohararp, the stencil shop, suggested shrinking some of the openings in half. This has helped, especially for pads with many thermal vias but the amount of solder on A1’s power bars was still excessive – take a look at the title picture, left side of the A1. On the subsequent builds about 3/4 of the paste from the power bars would have to be removed manually.
Note 3: The power supply was populated first and then tested. It turned out that PCIe power connectors exist in both polarities and the one we installed is of the wrong variety. On the the first turn on of the converter my bench supply promptly went into current limit at 1A showing ~1.4V across the terminals. The input polarity was reversed and as a result the power MOSFET body diodes started conducting. Once input wires has been swapped the converter started outputting ~0.92V, as expected.
Populating the rest of the board went uneventful. Even though A1 package has rather strange footprint it reflows and self-centers well. The ASIC part of the board was reflowed on a hot plate. Strictly speaking, this was not necessary; I’m sure the power supply wouldn’t mind another reflow cycle. However, I like my hot plate so much that I never skip an opportunity to use it.
Note 4: On the next power-on the ASICs started drawing current, however, we soon realized that we can’t talk to the board. The data interface of the ASIC requires 1.8V signals and none of the MCU we had on hand (Arduino, RPi, PIC24) can be run from such a low voltage. All I could do was to order parts to make a level shifter. According to tracking, they will arrive tomorrow – stay tuned!