Impact of Capacitance on Spike Interval, Width, and Pattern in a Spiking Neural Network

Keeping other circuit parameters constant, only changing the capacitance parameter can alter the inter-spike interval (ISI), peak width (Δt), and pattern of spike generation. In Fig4.a,b,c, a single spike pattern is generated during each spike formation process. In Fig4.a, with capacitance parameters C1=C2=4.7 nF, the spike ISI is approximately 43 μs and the peak width is approximately 3 μs. In Fig4.b, with capacitance parameters C1=2.2 nF and C2=4.7 nF, the spike ISI is approximately 19 μs and the peak width is approximately 1 μs. In Fig4.c, with capacitance parameters C1=C2=2.2 nF, the spike ISI is approximately 26 μs and the peak width is approximately 2 μs. In Fig4.d, with capacitance parameters C1=2 nF and C2=4.7 nF, the spike generation pattern changes, and two spikes are generated with an ISI of approximately 15 μs and peak widths of approximately 1.6 and 1.3 μs, respectively. We conducted both simulations and experiments, and the results were consistent.

Comparing the above data, it is evident that, under the condition of an unchanged spike generation pattern, the size of C1 has a greater impact on the ISI and peak width. If the size of C2 is fixed, as C1 decreases, the ISI and peak width decrease, with the peak width decreasing more than the ISI. If the sizes of C1 and C2 are changed proportionally, the ISI and peak width change proportionally. This indicates that by reducing the capacitance, we can shorten the ISI and increase the spike frequency, which can improve the processing speed in spike neural networks. Theoretically, the processing speed will only be limited by the parasitic capacitance of the S-NDR device.

When C1 and C2 are not well-matched, other spike patterns may be generated, such as two consecutive spikes in one spike generation period, as seen in Fig4.d. Similarly, artificial neurons used in biological neurons and spike neural networks also require the formation of multiple spike patterns [16].