11:45 〜 12:00
▲ [21a-C201-11] A Simple Nonbistable Architecture for Realizing Ising Machine with Stochastic Resonance Phenomenon
キーワード：Ising machine, Combinatorial optimization, Stochastic resonance
Gain-dissipative Ising machine (GIM) is a promising quantum analog application which can efficiently solve combinatorial optimization problems (COPs) by mapping them to the Ising model. Because of the symmetric bistability of the electron spin in the Ising model, traditional GIM is always limited to be implemented based on symmetric bistable physical systems. However, due to the restriction of symmetric bistability, many nonbistable physical systems have never been considered as GIMs, which limits the construction scheme of GIM.
To broaden the potential construction scheme of GIM, we proposed a simple non-bistable structure for realizing a GIM in simulation environment. We achieve nonlinearity that approximates a rectified linear unit by a series of a fixed resistance, a tunable bias voltage source, and a diode. When the nonlinear mapping process is completed, the FPGA records the voltage as the time-multiplexing analog spin amplitude. Then the spin exchange interaction is proceeded. Finally, the output feedback voltage will be applied to the resistance again for the next iteration.
In this work, by analyzing the dynamics of the proposed system in the decoupled state, we proved that the system is non bistable under any condition. Then, we utilized the popular COP benchmarks, i.e., the MAXCUT problem, to test the performance of the proposed system. By analyzing the system dynamics when solving certain COPs, the stochastic resonance phenomenon, which is an effect that the noise energy is used to enhance the system performance, is found in the proposed system. In addition, by adjusting the bias voltage , we can artificially adjust the optimal operating noise window of the proposed method. This indicates that the proposed system can adapt to different larger noise environments, which is impossible for the known traditional GIMs.
To broaden the potential construction scheme of GIM, we proposed a simple non-bistable structure for realizing a GIM in simulation environment. We achieve nonlinearity that approximates a rectified linear unit by a series of a fixed resistance, a tunable bias voltage source, and a diode. When the nonlinear mapping process is completed, the FPGA records the voltage as the time-multiplexing analog spin amplitude. Then the spin exchange interaction is proceeded. Finally, the output feedback voltage will be applied to the resistance again for the next iteration.
In this work, by analyzing the dynamics of the proposed system in the decoupled state, we proved that the system is non bistable under any condition. Then, we utilized the popular COP benchmarks, i.e., the MAXCUT problem, to test the performance of the proposed system. By analyzing the system dynamics when solving certain COPs, the stochastic resonance phenomenon, which is an effect that the noise energy is used to enhance the system performance, is found in the proposed system. In addition, by adjusting the bias voltage , we can artificially adjust the optimal operating noise window of the proposed method. This indicates that the proposed system can adapt to different larger noise environments, which is impossible for the known traditional GIMs.