Presentation Information
[16a-A33-10]Robustness enhancement of spin-wave Ising machine via interference asymmetry
〇Zhiqiang Liao1, Md Shamim Sarker1, Siyi Tang1, Hiroyasu Yamahara1, Munetoshi Seki1, Hitoshi Tabata1 (1.Univ. of Tokyo)
Keywords:
Ising machine,Spin wave
Gain-dissipative Ising machines (GIM) is a faster optimizer compared to traditional computers. However, real-world noise is difficult to control precisely, which can lead to performance degradation in GIMs. To address the issue of weak noise robustness in traditional GIMs, we have designed a GIM architecture based on spin wave interferometer (SWI) with asymmetry. SWI was a Y-shaped yttrium iron garnet (YIG) with a thickness of 20 nm and a length of 400 μm. By changing the phase difference between the two input terminals in the Y-sheped YIG, spin wave interference can be triggered. By adjusting the frequency, the interference gain can be in an asymmetric state without rotational symmetry.
Based on the interference nonlinearity, we constructed a simulation framework to explore the performance of SWI-based GIMs. For benchmarking, the MAXCUT problems were employed. The results show the performance of the SWI-based GIM significantly degrades with increasing noise intensity, consistent with previous reports. In contrast, SWI-GIM with asymmetric interference maintain a good performance in solving the problems even at higher noise levels. In summary, the asymmetry effectively enhances the robustness of SWI-based GIM.
Based on the interference nonlinearity, we constructed a simulation framework to explore the performance of SWI-based GIMs. For benchmarking, the MAXCUT problems were employed. The results show the performance of the SWI-based GIM significantly degrades with increasing noise intensity, consistent with previous reports. In contrast, SWI-GIM with asymmetric interference maintain a good performance in solving the problems even at higher noise levels. In summary, the asymmetry effectively enhances the robustness of SWI-based GIM.
Comment
To browse or post comments, you must log in.Log in