Princeton University

School of Engineering & Applied Science

Ultrafast Photonic Neuromorphic Processing and Nonlinearity Mitigation in Long-haul Transmissions by Nonlinear Optical Signal Processing

Speaker: 
Yue Tian
Location: 
Engineering Quadrangle B327
Date/Time: 
Monday, May 19, 2014 - 3:30pm to 5:00pm

<strong>Abstract:</strong>
 As digital electronics and integrated circuits have been dramatically progressed in the past decades, it is increasingly difficult to keep up with Moore’s Law due to the physical limitations of electronics. Recently optical devices are receiving an increased interest as a promising alternative because of the low latency and vast bandwidth enabled by nonlinear optical signal processing. The emerging nonlinear material and devices with higher efficiency and smaller footprint have paved the way for on-chip integration and practical deployment.
 Meanwhile, as an outcome of the marriage of the neuro-ethology drawn from biological neurons with modern engineering techniques, neuromorphic processing opens up a wide range of applications such as adaptive control, learning, perception, sensory processing and autonomous robots. Among all the mathematical models drawn from nervous systems, the leaky-integrate-and-fire neuron model is the most fundamental and widely used model in theoretical neuroscience. Its spiking coding and processing mechanism is computationally efficient and scalable, adopting the best features of both analog and digital computing. Therefore mimicking spike processing with photonics can result in bandwidths that are billions of times higher than biological neurons and substantially faster than electronics. By utilizing a number of nonlinear effects in semiconductor optoelectronic devices and nonlinear fibers, fully functioning photonic neuron prototypes are demonstrated with capability to process optical spikes at the picosecond level. Based on bench-top prototypes, two lightwave neuromorphic circuits are presented as well.
 Furthermore, to mimic the learning process in neurons, an optical spiking time dependent plasticity (STDP) device is invented using nonlinear properties in semiconductor components. With the optical STDP device, for the first time the supervised learning of a photonic neuron is demonstrated, potentially laying the foundation for ultrafast learning.
 In addition, the huge transmission capacity and ultra-long distance in long-haul transmission systems requires ultrafast signal processing, which is extremely hard for electronics to accomplish real-time processing at such high speeds. With the help of nonlinear optical signal processing, a phase-sensitive boosting scheme is proposed using all-optical phase conjugation to mitigate nonlinear impairments and greatly extend the system reach.