Scalable orthogonal delay-division multiplexed OEO artificial neural network trained for TI-ADC equalization

• DOI: 10.48550/arxiv.2305.06040
• Publication Date: 2023-01-01
• Authors: Andrea Zazzi; Arka Dipta Das; Jeremy Witzens
• Keywords: Power budget

We propose a new signaling scheme for on-chip optical-electrical-optical artificial neural networks that utilizes orthogonal delay-division multiplexing and pilot-tone based self-homodyne detection. This offers more efficient scaling of the optical power budget with increasing network complexity. Our simulations, on 220 nm SOI silicon photonics technology, suggest can support 31 x neurons, 961 links freely programmable weights, using single 500 mW comb an SNR 21.3 dB per neuron. Moreover, it features low sensitivity to temperature fluctuations, ensuring be operated outside laboratory environment. demonstrate network's effectiveness in nonlinear equalization tasks by training equalize time-interleaved ADC architecture, achieving ENOB over 4 entire 75 GHz bandwidth. anticipate this architecture will enable broadband latency signal processing practical settings such as ultra-broadband data converters real-time control systems.