Spikeling: A low-cost hardware implementation of a spiking neuron for neuroscience teaching and outreach

Abstract

Understanding how neurons encode and compute information is fundamental to our study of the brain, but opportunities for hands-on experience with neurophysiological techniques on live neurons are scarce in science education. Here, we present Spikeling, an open source in silico implementation of a spiking neuron that costs £25 and mimics a wide range of neuronal behaviours for classroom education and public neuroscience outreach. Spikeling is based on an Arduino microcontroller running the computationally efficient Izhikevich model of a spiking neuron. The microcontroller is connected to input ports that simulate synaptic excitation or inhibition, to dials controlling current injection and noise levels, to a photodiode that makes Spikeling light sensitive, and to a light-emitting diode (LED) and speaker that allows spikes to be seen and heard. Output ports provide access to variables such as membrane potential for recording in experiments or digital signals that can be used to excite other connected Spikelings. These features allow for the intuitive exploration of the function of neurons and networks mimicking electrophysiological experiments. We also report our experience of using Spikeling as a teaching tool for undergraduate and graduate neuroscience education in Nigeria and the United Kingdom.

Fig 1. Basic hardware and software.

A. Fully assembled Spikeling board. B. Screenshots of the Serial Oscilloscope software used, displaying Spikeling activity of the network in (C). C. Three Spikelings connected into a simple network.

Fig 2. Manual exploration of Spikeling functions.

A. Example recording of Spikeling membrane potential (top) and current (bottom) during manual manipulations of the input current dial (4) to depolarise the neuron (left), following the addition of a noise current (dial 3, right). B. Example light responses in modes 1–3 (left to right, toggled by the button) to manual PD stimulation with a torch. The grey horizontal lines indicate I_total = 0. PD, photoiode.

Fig 3. Basic stimulus-driven functions.

A. Example recording of Spikeling in mode 1 driven by the internal stimulator (port 1) via the analog-in connector (port 3), as indicated. Gain and stimulus rate are controlled on dials 2 and 1, respectively. Right: stimulus aligned response segments (grey) and average (black) as well as spike raster plot. B. As (A, right), with varying input gain to probe amplitude tuning. Note systematic effects on spike number, rate, time latency, and time precision. C. As (A), but this time driving Spikeling via an LED attached to the stimulus port stimulating the photodiode. Note different waveforms of input current and consequences on the elicited spike pattern compared to (A). D. As (C), with addition of current noise (dial 3). Note distortion of spike timings, while the number of spikes triggered remains approximately constant. LED, light-emitting diode.

Fig 4. Volley coding and stochastic resonance.

A. By varying the stimulus rate, Spikeling can be setup to ‘miss’ individual stimulus cycles at the level of the spike output (left). However, when elicited, spikes remain phase locked to the stimulus (right). B. Example of stochastic resonance. As (A), with neuron hyperpolarised just enough to prevent all spikes (left). Now, addition of membrane noise occasionally elicits spikes (middle), which again are phase locked to the stimulus (right). Dotted line indicates approximate spike threshold.

Fig 5. Synaptic networks.

A. Two or more Spikelings can be connected to form synaptic connections, as indicated. Left: excitatory synaptic connection with synaptic gain gradually increased by hand over time (dial 2). Right: inhibitory connection at two different depolarisation states (dial 4). B. Example of a 2-neuron CPG. The two Spikelings are set to mode 2 and wired to mutually excite each other. In each case, all traces display the activity and incoming spikes of the top-most Spikeling. CPG, central pattern generator.

Fig 6. Estimating linear filters by reverse correlation.

A. Via the Arduino code, the stimulus port can be set to deliver 50 Hz binary noise, here used to drive the photodiode via an LED (confer Fig 3C). Current and spike pattern elicited by this stimulus. B. Linear filters of a slow (black) and a fast (red) photo-adapting mode 1 neuron estimated at the level of spikes (left) and subthreshold membrane potential (right). LED, light-emitting diode.

Fig 7. Spikeling in the classroom.

A. ‘Bag of parts’ disassembled Spikeling, as used in our summer school in Gombe, Nigeria. B. Students soldering Spikelings as part of an in-class exercise on do-it-yourself equipment building. C, D. Students exploring Spikeling functions based on an exercise sheet provided (see Spikeling manual). E, F. In-class use of Spikeling as part of a computer lab for third year neuroscience undergraduates at the University of Sussex, UK.