Integrated multi-dimensional photonic tensor accelerators (PTA)
Guifang Li, Fatemeh Ghaedi Vanani, Alireza Fardoost, Zheyuan Zhu, Sean Pang, Christopher R. Doerr · 2024
We present a photonic tensor accelerator (PTA) employing nearly all dimensions (complex amplitude, polarization, wavelength and spatial mode) of light for scalable tensor multiplication. In this architecture, multiplications are performed through coherent mixing and accumulations are performed via simultaneous photodetection of multiple degrees of freedom within one dimension of light. We have implemented the PTA on micro-optics and integrated-photonics platforms. Experimental results of the PTA as well as its applications in artificial neural networks and computing will be presented.