The AVR Neuron is an AVR micro controller which accepts 0-10kHz analog input signals and returns the result of Izhikevich's neuron modeling function (signals of electric potential very similar to those observed in many types of mammalian neurons) on an analog output. Using serial communication, the Neuron can be configured to mimic any of the 20 most common spiking behaviors of biological neurons. Serial communication can also be used to capture digital representations of the model's important calibration and output values.
The version built and programmed by Charles Hart is a prototype. Networking has been tested by connecting the output to the input in ib mode, where periodic bursts occurred. Future improvements are already apparent, yet some would be simple to implement if anyone wanted to continue with this work. The project is in need of a neuro-scientist who can recognize practically useful features which the neuron could emulate if it does not already. I would also like to know what barriers yet remain to providing humans with new sensation given any analog or digital device and a network of <10 neurons such as these.
Using a "follower" op-amp and a properly calibrated power supply, the 0-5V output could easily be converted to the voltage and current ranges used by neurons in our own brains. Multiple analog input channels would be possible if the ADCMUX register were to be updated after every sample. Software-calibrated input weights are simple to implement. Software-calibrated output delay should not be too hard either. I would like to investigate the possibilities offered by USB: possibly faster on the microcontroller's end, maybe one pc can address multiple neurons in accord with the USB 2.0 spec (used only for midi devices afaik).
The prototype is much larger than it could be. I imagine that a 1x1" device could be manufactured using 7-layer board techniques. Using a 2-layer board and a number of admittedly non-ideal hardware components, a device measuring approximately 2x2" has been designed. Components would cost about 10USD per neuron for a small-scale run, but other manufacturing costs remain unknown.
