A DC to 5.8 GHz CMOS Variable-Gain Transimpedance Amplifier for Photonic Neuromorphic Hardware
Lukas Hüssen, Xun Chen, Muh‐Dey Wei, Renato Negra, Arka Dipta Das, Andrea Zazzi, Jeremy Witzens · 2024
The design of specialised neuromorphic computing hardware has gained a lot of interest in recent years. The implementation of photonic neuromorphic circuits promises to bring the speed of optical data links to the field. However, not all functions needed to build a photonic neural network are yet effectively integratable in a photonic chip. Especially, the implementation of the activation function poses a challenge in the photonic design. In this work, a transimpedance amplifier (TIA) for photonic neuromorphic hardware is presented. The design delivers a sigmoid-type current to voltage transfer characteristic with variable gain and offset from DC to several gigahertz, including a track and hold circuit enabling a sampled output at up to 10 GHz clock speed. Four TIAs with different peaking inductor distribution were manufactured in 65nm CMOS. The measurements validate a maximum 3-dB bandwidth of DC to 5.8 GHz in the lowest gain stage for the design with peaking inductors at the output stage.