Novel Ferroelectric-Based Ising Machine Featuring Reconfigurable Arbitrary Ising Graph and Controllable Annealing Through Device-Algorithm Co-Optimization
Weikai Xu, Jin Luo, Zhiyuan Fu, Runze Han, Shengyu Bao, Kaifeng Wang, Qianqian Huang, Ru Huang · 2024
For the first time, a ferroelectric (FE)-based Ising machine (FE-Ising) is proposed and experimentally demonstrated with the fully-flexible spin$(\sigma)$connectivity and the lowest hardware cost through device-algorithm co-optimization. For the Ising topology, a novel interaction$(J)$-centric topology is proposed, which enables reconfigurable arbitrary Ising graph without the need of extra configuration overhead in conventional$\sigma$-centric topology. For hardware implementation of$J$-centric FE-Ising, a compact$J$unit composed of n-type and p-type 1T1C FeFET is further proposed and experimentally demonstrated with robust device reliability and small variability. Besides, by employing the computing-in-memory architecture, the FE-Ising graph is experimentally constructed to compute the Hamiltonian energy. Moreover, for$\sigma$update and controllable annealing, the novel FE-based programmable inverter and tunable random number generator are also proposed and experimentally presented. Based on the above hardware system, the solving processes of max-cut problem are demonstrated with significantly enhanced speed, success probability ($>$90%) and energy-efficiency, showing its great potential for solving the sophisticated and heterogeneous real-world combinatorial optimization problems.