Capacitance Control of Spiking Modes and Spike Characteristics in Neuristor Devices

By adjusting the capacitance value, we have the ability to manipulate the inter-spike intervals (ISI), spike widths (Δt) and spiking modes while keeping other circuit parameters constant. The spike width is defined as the full width at half maximum. In Fig. 4(a)-(c), we observe regular spiking mode where only one spike is generated per period. In Fig. 4(a), with C1=C2=4.7 nF, the ISI is around 43 μs and the spike width is approximately 3 μs. In Fig. 4(b), with C1=2.2 nF and C2=4.7 nF, the ISI is approximately 19 μs and the spike width is roughly 1 μs. In Fig. 4(c), where C1=C2=2.2 nF, the peak interval time is around 26 μs and the peak width is approximately 2 μs. In Fig. 4(d), we observe a new mode of spiking called chattering mode, where two spikes are generated in one period. The capacitance values are C1=2 nF and C2=4.7 nF, the ISI is around 15 μs, and the width of the two spikes are around 1.6 and 1.3 μs respectively. Our experimental results are consistent with simulation results, as shown in Fig. 4.

In regular spiking mode, we found that C1 has a significant impact on both ISI and spike width when C2 is fixed. Specifically, a smaller C1 leads to a smaller ISI and spike width, and the reduction ratio of spike width is greater than that of ISI. When C1 and C2 are scaled up or down proportionally, the ISI and spike width also scale up or down proportionally. Therefore, reducing capacitance can effectively shorten the interval between spikes, increase spike frequency, and improve the computing speed in spiking neural networks, with the only limiting factor being the parasite capacitance of the S-NDR devices.

Moreover, by adjusting the ratio of C1 and C2, different spiking modes can be generated, such as the chattering mode in Fig. 4(d). These different spiking modes not only mimic the behaviors of biological neurons but also play an important role in network design and neural coding in spiking neural networks.