Regeneration of 200 Gbit/s PAM4 Signal Produced by Silicon Microring Modulator (SiMRM) Using Mach–Zehnder Interferometer (MZI)-Based Optical Neural Network (ONN)

Abstract

We propose and demonstrate a Mach–Zehnder Interferometer (MZI)-based optical neural network (ONN) to classify and regenerate a four-level pulse-amplitude modulation (PAM4) signal with high inter-symbol interference (ISI) generated experimentally by a silicon microing modulator (SiMRM). The proposed ONN has a multiple MZI configuration achieving a transmission matrix that resembles a fully connected (FC) layer in a neural network. The PAM4 signals at data rates from 160 Gbit/s to 240 Gbit/s (i.e., 80 GBaud to 120 GBaud) were experimentally generated by a SiMRM. As the SiMRM has a limited 3-dB modulation bandwidth of ~67 GHz, the generated PAM4 optical signal suffers from severe ISI. The results show that soft-decision (SD) forward-error-correction (FEC) requirement (i.e., bit error rate, BER −2) can be achieved at 200 Gbit/s transmission, and the proposed ONN has nearly the same performance as an artificial neural network (ANN) implemented using traditional computer simulation.

Keywords

• silicon photonics (SiPh)
• silicon-on-insulator (SOI)
• pulse amplitude modulation (PAM)
• silicon microring modulator (SiMRM)
• optical neural network (ONN)