We propose and experimentally demonstrate an automated wavelength seeking scheme for an O-band 1 × 2 silicon photonic switch based on an add-drop structure with two cascaded resonators. In the scheme, two three-layer neural networks are employed to learn the relationship between the monitored optical powers and the heating voltages of the two coupled ring resonators. Then, the two neural networks can quickly predict the required heating voltages according to the monitored optical powers. Therefore, the wavelength seeking can be realized through a training process, which is carried out only once, and a seeking process. By using this scheme, the dual-ring switch can be locked to the wavelength of the input light with a constant duration of 860 μs. The maximum seeking errors for the two rings are 0.08 and 0.09 nm, respectively.

中文翻译：

---

基于人工神经网络的硅双环开关快速寻波长

我们提出并实验证明了一种基于具有两个级联谐振器的分插结构的 O 波段 1×2 硅光子开关的自动波长搜索方案。在该方案中，采用两个三层神经网络来学习监测到的光功率与两个耦合环形谐振器的加热电压之间的关系。然后，两个神经网络可以根据监测到的光功率快速预测所需的加热电压。因此，波长搜索可以通过只进行一次的训练过程和搜索过程来实现。通过使用该方案，双环开关可以以860 μs的恒定持续时间锁定输入光的波长。两个环的最大寻道误差分别为 0.08 和 0.09 nm。