Hybrid Photonic Systems for Neuromorphic Processing

Optical computing methods are seeing a resurgence of popularity due to recent advances in integrated photonics and neuromorphic engineering. Photonic systems are very suitable for analog computing approaches with moderate accuracy, yet very high processing speed. These features hold promise for implementing photonic accelerator systems for computationally expensive tasks such as matrix vector multiplications (MVM). Here I will introduce a nanophotonic approach for realizing chip-scale MVM-units by making use of hybrid integration. Using phase-change photonic devices allows for creating parallel processing circuits in which analog multiplications can be carried out at high speed. Hybrid integration of active and passive integrate photonic chips via 3D printed optical interconnects enables the realization of compact neuromorphic architectures for ultrafast information processing. Applications lie in the implementation of photonic convolution processers, as well as the realization of associative photonic memories and ultrafast correlation systems. In combination with the development of novel light sources, such photonic approaches offer new opportunities for creating brain-inspired hardware for artificial intelligence applications.