I brought the v05 prototype to the Superconference and briefly showed it during my talk (then proceeded to zip-tie it to my badge and carry it around for the rest of the evening). If you missed that, here's a quick clip of the board in "neuron demo mode":

So... the finished product won't have a bunch of DIP switches on board, or a 0.1" spacing ISP header, or a bunch of varnished jumper wires everywhere. The blue LED (which is also red sometimes) won't stay on all the time; these indicators will be used to show which dendrites are excited or inhibited, probably pulsing with brightness proportional to input weighting.

Beyond the LEDs, that's the other big change--input weighting. I'm also planning to use 4-pin connectors, so we can implement a backpropagation mode that actually changes individual dendrite weightings based on sequential firing. We've got a bunch of other ideas similar to this, but operating modes really are the key bit. Another shot, including the completely useless (well, beyond its current use) USB power adapter I put together:

I used the RJ11 (phone handset) cord because it's the only cable I had on hand that (a) was multi-conductor, and (b) had an O.D. close to the JST connector spec. I tried my usual rainbow ribbon cable and it split the connectors apart. I kept the phone plug on for, uh, modularity? Seemed silly to just cut off a perfectly good connector...

Briefly, here's where the v05 schematic stands:

It's anything but finalized, so I'm not quite ready to push this to the repo as a final product. In any case, I really like having eight indicator LEDs, and we both feel that six dendrites and three axons is ideal. I want to do some fancy dimming stuff with the LEDs so I'd prefer to avoid the complications related to multi/charlie-plexing, so I dropped a few of the R/B dendrite LEDs in favor of membrane potential indicators. This will actually work pretty well--we're planning to group pairs of dendrites together for one of our operating modes, so sharing a single R/B LED between two inputs makes sense. I also switched to a bunch of resistor networks, because 25 0402 resistors seems like a super pain in the ass to deal with.

Layout is still a work in progress. As promised, the new version is going to look more neuron-like than the square boringness that was v04. Further work will likely wait until after turkey day, so stay tuned...