Self‑configuring silicon chip sorts and controls eight light beams by spatial pattern and polarization
This paper reports an experiment that builds a compact silicon photonic chip that can separate and control multiple light beams at once. The device works on both the shape of the beam in space and its polarization (the direction the electric field points). The authors demonstrate that the chip can take a messy, random input of overlapping beams and turn it into clean, independently controlled output beams.
The key idea is to combine on‑chip optical antennas with a reconfigurable interferometer network that performs an optical singular‑value decomposition (SVD). The antennas are diffraction grating couplers designed to support eight orthogonal modes that differ in spatial form and polarization. The interferometer network is built from Mach–Zehnder interferometers (MZIs) and thermal phase shifters. In plain terms, the chip first finds the independent ‘‘eigen’’ beams hidden inside a random speckle input, then acts on each independent channel, and finally remaps them to the desired output beams.
Because the chip implements SVD optically, it can sort and address all eight orthogonal channels without converting light to electrical signals. The middle part of the SVD (called Σ) is a diagonal set of weights. Those weights can be changed on chip using modulators or switches to do different operations. Using this architecture the authors show in situ self‑programming that enables different functions on the same hardware, including beam shaping, optical switching (routing signals between channels), and a reconfigurable optical add‑drop multiplexer (ROADM) that can add, drop, or pass signals on selected channels.
This approach matters because it brings powerful, flexible control of many light channels into a compact, silicon‑photonic platform that is compatible with CMOS manufacturing. Multidimensional control of light can increase data capacity and enable applications in optical interconnects, computing, imaging, sensing, and quantum networks. Doing the decomposition and reassembly directly in the optical domain keeps the process fast and avoids extra electrical conversion steps.