Week 3 - FabISP

Week 3 - Make the FabISP!

This week our assignment was to mill our very own FabISP from FR1 copper plated boards with the Modela desktop milling machine. This is an exercise in introducing everyone to electronics while simultaneously building their own low cost ICSP programmer for future projects. I have some previous experience with electronics, so I thought I would try to mod the boards a little with some LEDs. I started off trying to decipher the circuit by looking at the .png files we were provided. Most things made sense, though the zener diode voltage clamps on the USB data lines and the requirement of the jumper for VCC to the ISP connector weren’t obvious. Was power from that ISP connector going to power whatever the FabISP would be programing? Or would that connection power the FabISP when it was initially being programmed? On the USB side, I knew that USB was differential signaling, and that the power from USB was 5V, but beyond that didn’t know much about the details of USB data transfer. This link from the history of the FabISP provided some useful background on the nitty gritty of USB data transfer the FabISP would be doing. Notably it mentioned that the USB digital data lines work at 3.3V (explaining the 3.3V zener voltage clamps), and that FabISP would be using the USB low-speed protocol. This is known as USB 1.x, and the low-speed protocol specifically has a data rate of 1.5Mbit/s. This is a 1.5bit/us, meaning the 20MHz Attiny44a doesn’t have a ton of head-room for other processing while dealing with USB data. This is in contrast to really simple UART style serial data protocols found with Arduinos or the Raspberry Pi, which run (as a fast setting) at 115200bit/s (0.1Mbit/s). Next I tracked down an old version of the USB 1.0 specifications, which sadly made me realize that I couldn't drive LEDs directly off the USB data lines. Determined to add an LED somewhere, I decided to go for a simple power indicator.

Takeaways:

Modified the designs by tracing the .png in Inkscape and working with the vectors.

Added some LEDs across power rails and responding to MOSI activity

Some toolpath issues, smaller tool diameter in fab modules fixed

Make sure to double check you have the right resistor values. I misread the labels and picked the wrong ones twice

Source files at bottom of page

Rendered Inkscape overlay - "Valentin" FabISP modified for a power indicator LED

We were pointed to a few choices on FabISP designs: Neil's FabISP, the FabISPkey, and the Valentin. I chose the Valentin, which had the combined features of small size, an on board USB connector, a key ring, and a snap off design. I also wanted to modify it to add a power indicator LED. I first imported the board .pngs into Inkscape, and used the trace bitmap tool with multiple scans without smoothing to make a vector representation of the png. With all the traces as objects in Inkscape, I could easily select subsets to move around in order to modify the board. In the image above, the left is Valentin's original board, and the right is my version. I extended the board by altering the exterior cutout, and moved some traces around so I could add a few additional pads (in yellow) for a resistor and an LED. These pads bridge the power and ground rails, making for a nice indicator. I found the LED datasheet by following the part number from the fab inventory page. The green LED had a pretty typical (worst case) voltage drop of 2.1V, meaning a 499ohm resistor from the 5V rail would be more than sufficient to keep the current under the 20mA maximum current of the part.

Programming the LED modified Valentin

Unfortunately I don't have images of the initial construction, but here is an image of programming the FabISP Valentin. A few comments on the assembly: Milling the FabISP wasn't too difficult, but there were some tricks. My first attempt didn't work due to the 1/64th endmill not being correctly zeroed and the board not being properly taped down. It only polished the copper a bit and cut on one side. After retaping, rezeroing, and hitting down on the modela 2 times after zeroing but before cutting, it made clean cuts for the traces. However, the toolpaths being created by fab modules weren't quite right and some shorts existed. By decreasing the tool diameter parameter (0.3mm 1/64 endmill, 0.5mm 1/32 endmill), the path was generated correctly and everything cut like I wanted. Overall, it took 2 1/64 runs and 2 1/32 runs. This was fine as the Modela is high resolution enough that as long as everything is zeroed properly it will cut in the exact same places. Programming was a bit tricky as well. First off, this resource on what the AVRISP mkII's lights mean was quite helpful. I guess I didn't read the instructions properly as it took me awhile to figure out that the FabISP needed to be externally powered. In the image above I'm powering it off the bench power supply. I also installed some 499k resistors instead of 499ohm, which took a while to figure out. Next, I tried to compile and install the firmware on my mac. I installed the avr toolchain using MacPorts, but could never get the thing to program properly. The error log is here. In short, it seemed to program successfully, and then failed with "avr_read(): error reading address 0x0000 read operation not supported for memory "signature"". No idea what the issue is, but I tried again from the Ubuntu computer running the Modela and it programmed perfectly.

Modifying Neil's FabISP for more blinkage

The Valentin worked fine, but it seemed a bit little fragile, and having to plug it directly into my laptop's USB port made it harder to work with. I also still wanted more LEDs, in particular I wanted LEDs that would flash with activity. The USB data lines seemed off limits due to their low current sourcing capability and expecting certain impedances. On the other hand, the MOSI / MISO, SPI lines were being driven by robust microcontroller pins that could source/sink enough current to light an LED. I went back to Neil's FabISP, moved some things around, and took advantage of the extra space. The left side in the image above shows Neil's original design, while the middle and right images show my modifications. I added two LEDs: one green power indicator like the last design, and a blue LED going from 5V to the MOSI trace for the ICSP connector as an activity indicator. I found this helpful document on Atmel's ICSP protocol. When there is no data transfer, the MOSI pin is in a high impedance state. However, when data is being transfered, the MOSI pin on the attiny44a is asserting either 5V or 0V. The 0V state sinks current through the blue LED, thereby lighting it up only when a data transfer is being attempted through the ICSP connector. According to the attiny44a datasheet, the I/O pins on the micro can sink/source ~40mA. In theory that is more than enough headroom to drive an LED while simultaneously asserting digital logic. I put a 1k ohm resistor to current limit the LED, by eye the blue LED flashes as bright or brighter than the green LED. In actuality I'm unsure if the addition of this LED on the MOSI trace will affect the programming process, as I haven't yet tested it directly. That being said, the LED can easily be removed or the resistor changed to a (less disruptive) high resistance value if it significantly affects the robustness of the programming process.

FabISP-LED and FabISPKey-LED

End product turned out pretty well! If you're wondering how I took a picture of the blue LED (purportedly only active with activity): I entered sudo make fuse on my computer while simultaneously taking the picture. This makes the FabISP briefly ping for things to program on its ICSP connector. Overall I really enjoyed making circuits on the Modela, really neat way to rapidly prototype circuit boards. Source files below: