Mainz 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
Q: Fachverband Quantenoptik und Photonik
Q 17: Photonics and Biophotonics I
Q 17.5: Talk
Tuesday, March 3, 2026, 12:00–12:15, P 3
Loss-Minimized Incoherent Photonic Computing on a Mach-Zehnder interferometer network — •Konrad Tschernig1,2, Mingwei Yang1,2, Felix Kübler1, Okan Akyüz1, Lennart Mannteuffel1, Enrico Stoll1, and Janik Wolters1,2 — 1Technical University of Berlin, Berlin, Germany. — 2German Aerospace Center (DLR), Berlin, Germany.
We present an algorithm for loss-minimized incoherent photonic multiplication of N-dimensional vectors with N× N matrices on standard Clements Mach-Zehnder interferometer (MZI) meshes [1]. By applying arbitrary unitary transformations to incoherent light states, our method avoids phase control and additional MZI blocks required by the singular value decomposition in coherent schemes [2]. In comparison, the crossbar architecture [3] requires N times more optical energy than our scheme to perform the same N × N matrix-vector multiplication (MVM). Experimentally, we implement a 4×4 photonic MVM and demonstrate an optical convolutional neural network for MNIST classification. Utilizing our loss-minimized architecture, we aim to reduce the input intensity to the single photon regime to explore the limitations from shot noise.
[1] Clements, William R., et al., Optica 3.12 (2016): 1460-1465.
[2] Shen, Yichen, et al., Nature photonics 11.7 (2017): 441-446.
[3] Feldmann, Johannes, et al., Nature 589.7840 (2021): 52-58.
Keywords: Matrix Vector Multiplication; Mach-Zehnder Interferometer; Clements decomposition; Incoherent light sources; Artificial Intelligence