PHI Lab Research
Rethinking the computer from principles of critical phenomena in neural networks.
We aim to develop special purpose computers, accelerators to modern digital computers, for combinational/continuous optimization problems, quantum simulation of many body systems and deep machine learning. A coherent Ising Machine based on degenerate optical parametric oscillator network solves various combinational (discrete) optimization problems by mapping them onto the Ising model, while a coherent Kuramoto machine based on non-degenerate optical parametric oscillator network solves various continuous optimization problems by mapping them onto the Kuramoto model. A coherent SAT (Satisfiability Problem) solver based on recurrent neural network solves some other optimization problems by mapping them onto the k-satisfiability problem. A coherent mixing solver based on optical homodyning performs larger-scale vector-vector multiplication.