Noise-Controlled Signal Transmission in a Multithread Semiconductor Neuron

We report on stochastic effects in a new class of semiconductor structures that accurately imitate the electrical activity of biological neurons. In these devices, electrons and holes play the role of ${\mathrm{K}}^{+}$ and ${\mathrm{Na}}^{+}$ ions that give the action potentials in real neurons. The structure propagates and delays electrical pulses via a web of spatially distributed transmission lines. We study the transmission of a periodic signal through a noisy semiconductor neuron. Using experimental data and a theoretical model we demonstrate that depending on the noise level and the amplitude of the useful signal, transmission is enhanced by a variety of nonlinear phenomena, such as stochastic resonance, coherence resonance, and stochastic synchronization.

Figure 1

(a) Solid state neuron excited by a sinusoidal drive of amplitude $V_D$ and a white noise with standard deviation (rms voltage) $V_N$. Depolarization waves emanating from input contacts diffuse along the $p\mathrm{\text{−}}n$ wire and interfere in the central region (soma) before reaching the output. The $p^{+}\mathrm{\text{−}}{n}^{+}$ tunnel layers are the pillars supporting the free standing $p\mathrm{\text{−}}n$ lines. (b) Sigmoid gain curve of the tunnel amplifier. (c) $I\mathrm{\text{−}}V$ characteristics of the tunnel device ($\mathsf{N}$-shaped) and the load line including circuit elements $R=0\Omega$, $L=10\mathrm{mH}$, $V_{\mathrm{soma}}=290\mathrm{mV}$ and the $p\mathrm{\text{−}}n$ wire. (d) Equivalent circuit of the artificial soma.

Figure 2

Sequence of output spikes for $V_D=0$: (a) experimental, (c) calculated. Phase portraits: (b) experimental $\Delta =1.6\mu \mathrm{s}$, (d) theoretical $\Delta =0.075\sqrt{LC}$.

Figure 3

(a)–(d) Sequences of spiking events at four noise intensities across stochastic resonance. For clarity, the noise is shown $\times {10}^{-1}$. (e)–(h) Power spectral densities of the output signal. The peak passes through a maximum at $V_N=130\mathrm{mV}$.

Figure 4

Characteristics of the neuron output as a function of noise at three amplitudes of the drive signal: $V_D=70\mathrm{mV}$ (open symbols), 40 mV (full symbols), 0 (solid line). (a),(c) Power of periodic component at $f_D=28\mathrm{kHz}$; (b),(e) mean spiking frequency; (c),(f) coherence of interspike intervals. The simulated frequency (power) was converted from dimensionless to experimental units by multiplying with $3.4/(|g|L)$ ($340\mu \mathrm{W}$.)

Figure 5

(a),(c) Power of periodic component; (b),(d) mean spiking frequency as a function of drive and noise amplitudes. Left: experimental data; right: simulation with Eqs. (2).