An Optical Processor for Matrix-by-Vector Multiplication: an Application to the Distance Geometry Problem in 1D

Simon Hengeveld (1) , N Rubiano da Silva (2) , D Gonçalves (2) , P Souto Ribeiro (2) , Antonio Mucherino (3)

1 IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires
2 UFSC - Universidade Federal de Santa Catarina = Federal University of Santa Catarina [Florianópolis]
3 Inria Rennes – Bretagne Atlantique

We present the architecture of a new optical processor specialized in matrix-by-vector multiplication via the manipulation of the light wavefront. This processor can reach up to 1.2 Giga MAC (multiply-accumulate) operations per second using commercially available devices. Moreover, this architecture is compatible with a hardware upgrade with the potential to achieve a processing speed of above Tera MAC per second. We initially present the optical processor and then discuss the use of such a processor for tackling a special class of the one-dimensional Distance Geometry Problem (DGP), which is a well-known NP-hard problem.

Domaines: Complexité [cs.CC]; Géométrie algorithmique [cs.CG]; Recherche opérationnelle [math.OC]; Optique [physics.optics]

Simon Hengeveld, N Rubiano da Silva, D Gonçalves, P Souto Ribeiro, Antonio Mucherino. An Optical Processor for Matrix-by-Vector Multiplication: an Application to the Distance Geometry Problem in 1D. Journal of Optics, 2022, 24 (1), pp.1-11. ⟨10.1088/2040-8986/ac3a9e⟩. ⟨hal-03636293⟩