Here are notes on the AV3 PCI-104 Atom-based flight computer stackup.
Board 1: Advantech PCM-3363N-1GS6A1E
The Advantech PCM-3363N-1GS6A1E is a PCI-104 CPU board with an Intel Atom N455 1.66 GHz Single Core processor, 1GB DDR3 800 MHz Memory soldered on-board, CompactFlash (CF) slot, 4x USB ports, 4x COM ports, and a whole slew of standard PC ports (VGA, SATA, etc). Note that we got the commercial temperature range model, not the industrial range, since that wasn't available when we had to order the boards.
Note while this board keeps the X,Y PCI-104 form factor, the above-board heatsink and the below-board CF slot violate the above and below-board heights, respectively.
We have a Lexar Professional 300x 8 GB CompactFlash that claims UDMA support.
Things we hacked:
- Jumpered J2 pin 2/3 to make it RS422 (because RS484 is insane)
- J3 jumpered pin 2/3 to make PCI I/O 3.3 V
- J1 jumpered to always turn on ATX power
- Removed CN4 (front panel) because it has a reset and power button lines (we don't want accidentally tripped)
- Removed CN17 GPIO connector in prep for directly soldered cable
- Removed 8 pull up resistors from CN17 that were attached to PCA 9555 chip
We need to add BH1 which is the RTC battery power.
Board 2: Advantech PCM-3680I
The Adventech PCM-3680I is a PCI-104 2-port CAN-bus module with isolation protection. It's basically an FPGA that interfaces two SJA1000 controllers to the PCI bus, along with with very nicely isolated CAN driver sections.
The PCM-3680I also violated the above-board height because of two isolated DC-DC converters that were 10 mm high. These converters were in the space that the CF slot takes up when stacked below the CPU module, which is the only sane way to stack the boards. So, we got out the soldering irons, removed the supplies, laid them on their sides, siliconed them down to the PCB, and soldered mod-wires back into the original holes.
In more detail: The space between the boards is 14.95 mm, and the CF slot is 8.35 mm high giving a total space left of 6.6 mm. There's nothing we could do to the MinMax MAU250 (3kv isolation, 1 W, 4-200mA out, 100KHz switiching, 5V in, 5 V out) DC-DC converters to make them fit under the CF slot - even swapping them out for a smaller DC-DC converter wouldn't work since they're all just about as huge as the current ones (which are 19.5 x 10.2 x 7.3 mm). So we removed the converters, laid them on their sides, and moved them out from underneath the CF slot.
The 120 ohm termination resistor jumper also was in the way, so we removed those jumpers and just bridged it with a mod wire so that the 120 ohm resistor is permanently there, which makes sense for the FC.
Board 3: Tri-M HE104+DX-V1
The Tri-M HE104+DX-V1 is a 108 W high efficiency (<= 90%)switching power supply that takes 6V to 40V DC input and outputs +3.3V, +5V, +12V, -12C DC output, all on the PCI connector. Putting at the bottom of the stack is weird; it really ought to go on the top of the stack so it can vent heat, but the FC needs to go there due to its giant heatsink.
On AV3-L10, we used the SD ("shutdown") signal to turn on and off the FC from the APS GFE. Previous to L10, the SD was bypassed on the HE104+DX-V1 by a solder bridge on pins 7-8 of the Vishay ILD256T opto-isolator chip (IC3) right next to the HE104's output connector. We removed that for L10. Note that the SD pin goes through two 10K resistors in paralell and then hits pin 1 of the ILD256T, and then pin 2 is grounded. For AV4-L11, we added the solder bridge back to bypass SD since the FC is powered up directly from the RocketNet Hub.