v0.4

v0.4 requires a bit of extra consideration; in addition to scaling manufacturing by more than an order of magnitude, this generation was current for more than a year and was the basis for Zach and Joe's collaboration. During its lifespan, v0.4 took us to The Hackaday Prize 2015 Best Product Finals and helped NeuroTinker, LLC land a Phase I SBIR grant from the National Science Foundation. This overview consolidates a lot of scattered information from Zach's Hackaday Projects site (1, 2, 3, 4, 5, 6, 7) along with numerous other sources.

A pile of recently manufactured NeuroBytes v0.4 boards, showing the pre-NeuroTinker Salfred Labs text.

A pile of recently manufactured NeuroBytes v0.4 boards, showing the pre-NeuroTinker Salfred Labs text.

The primary goal of NeuroBytes v0.4 (called "Neuron" at the time) was to prove out the core concept of the product at pilot scale quantities. We wanted this generation to be low-cost, physically small, flexible, and durable enough to start getting in front of potential users. Zach made a brief overview video in late 2014 that dives into more details:

Key Components

NeuroBytes v0.4 was the first generation based around a discrete microcontroller rather than a more expensive (but easier to use) Arduino board. We selected the ATtiny44A for a few reasons: low cost, I/O capability, and--most importantly--AVRDUDE, the friendly command-line utility that gives the ATtiny series an entirely open-source development toolchain.

ATtiny44A microcontrollers waiting to be soldered onto NeuroBytes v0.4 boards.

ATtiny44A microcontrollers waiting to be soldered onto NeuroBytes v0.4 boards.

We also decided that the v0.2 indication method--an RGB LED--was the way to go, as it closely mirrored Andrew's original software design and clearly showed when neurons were excited, inhibited, firing, or at rest. Buying the cheapest LEDs on Digi-Key probably wasn't the best choice as the 605nm "red" ended up being more of an orange (my fault), but they worked well enough for prototyping purposes.

v0.4 RGB LEDs. In addition to being orange rather than red, they were a pain to solder by hand since we used the pad design intended for reflow soldering. 

v0.4 RGB LEDs. In addition to being orange rather than red, they were a pain to solder by hand since we used the pad design intended for reflow soldering. 

NeuroBytes v0.4 needed to be modular; we wanted to put the platform in front of students and educators alike. As such, the connector design was crucial to the success of the prototype. We ended up with insertion-mounted TE Connectivity 2mm headers which worked well enough for this generation:

Board mounted shrouded headers, connector housings, and crimp terminals.

Board mounted shrouded headers, connector housings, and crimp terminals.

This project was Zach's first foray into real circuit design. He picked his way through a few KiCad tutorials and ended up with a simple schematic and layout:

v0.4 schematic, showing six dendrite connectors with pulldown resistors, ATtiny44A microcontroler, bypass capacitors, RGB LED with current limiting resistors, and single axon terminal.

v0.4 schematic, showing six dendrite connectors with pulldown resistors, ATtiny44A microcontroler, bypass capacitors, RGB LED with current limiting resistors, and single axon terminal.

v0.4 final board layout with dimensions. Zach was unnecessarily proud of the lack of vias.

v0.4 final board layout with dimensions. Zach was unnecessarily proud of the lack of vias.

Accessories

The previous physical NeuroBytes prototype iterations only existed on a breadboard and interacted with the real world through a couple of pushbutton switches. The new boards eventually spawned a range of accessory modules that improved usability, some of which are shown here.

EXCITERS: These modules plug into one of the six v0.4 dendrite sockets. Exciters simply connect the VCC and SIG pins, causing the resting membrane potential value to increase (in excitatory sockets) or decrease (in inhibitory sockets). Since the standard v0.4 firmware scales each input at 70%, Exciters can be used to "sensitize" NeuroBytes so that a single excitatory input will cause an action potential.

EXCITERS: These modules plug into one of the six v0.4 dendrite sockets. Exciters simply connect the VCC and SIG pins, causing the resting membrane potential value to increase (in excitatory sockets) or decrease (in inhibitory sockets). Since the standard v0.4 firmware scales each input at 70%, Exciters can be used to "sensitize" NeuroBytes so that a single excitatory input will cause an action potential.

SWITCHES: Like Exciters, these modules connect VCC to SIG, only they do so on a momentary basis. The tiny PCB-style switches are good for injecting signals into loops, while the larger snap-action switches are excellent environmental sensors.

SWITCHES: Like Exciters, these modules connect VCC to SIG, only they do so on a momentary basis. The tiny PCB-style switches are good for injecting signals into loops, while the larger snap-action switches are excellent environmental sensors.

AXONS: These cables connect two v0.4 boards together, either Axon-to-Dendrite (for signaling) or in other configurations to simply share power. 

AXONS: These cables connect two v0.4 boards together, either Axon-to-Dendrite (for signaling) or in other configurations to simply share power. 

AXON TERMINALS: The v0.4 design only featured a single axon connector, making it a bit tough to form complex networks. These accessories allow the user to signal multiple downstream NeuroBytes from a single board.

AXON TERMINALS: The v0.4 design only featured a single axon connector, making it a bit tough to form complex networks. These accessories allow the user to signal multiple downstream NeuroBytes from a single board.

Pilot Scale Production

In a somewhat risky move, Zach skipped prototyping the circuit design and jumped right into pilot-scale production; in this case, that meant ordering four panels of 32 boards each. The boards were solder pasted and populated by hand, reflow soldered in a modified toaster oven, then hand soldered for the seven connectors. 

Eight early v0.4 boards prior to depanelizing, running an LED test program.

Eight early v0.4 boards prior to depanelizing, running an LED test program.

First pass yield stood at roughly 85% for the entire run, with half of the remaining boards getting successfully reworked. The most common non-recoverable defect: backwards microcontroller installation.

First pass yield stood at roughly 85% for the entire run, with half of the remaining boards getting successfully reworked. The most common non-recoverable defect: backwards microcontroller installation.

If you want to learn a bit more about the pilot scale process we followed, Zach gave a talk at The Hackaday Superconference 2015 highlighting a few details: