Neuro-morphic Circuit Architectures Employing Temporal Noises and Device Fluctuations to Improve Signal-to-noise Ratio in a Single-electron Pulse-density Modulator
Andrew Kilinga Kikombo, Tetsuya Asai and Yoshihto Amemiya
We investigated the implications of static noises in a pulse-density modulator based on the Vestibulo-ocular Reflex model. Based on this model, we constructed a simple neuromorphic circuit consisting of an ensemble of single-electron devices and confirmed that static noises (heterogeneity in circuit parameters) and dynamic noises (random noises as a result of spontaneous tunneling events) introduced into the network indeed played an important role in improving the fidelity with which neurons could encode signals whose input frequencies are higher than the intrinsic response frequencies of single neurons. Through Monte-Carlo based computer simulations, we demonstrated that the heterogeneous network could correctly encode signals with input frequencies as high as 1 GHz, twice the range for single (or a network of homogeneous) neurons.
Keywords: single-electron devices, neuromorphic LSIs, noise driven LSIs, neural networks.