使用倒傳遞類神經網路自動合成IPD製程上之微波頻電感

Abstract

本篇論文提出利用倒傳遞類神經網路自動合成出使用者需要的微波頻電感。由於RF被動電路設計過程中，電感在眾多被動元件中影響整體效能最多的，而電感的設計往往也是最耗時最難處理的部份。 為了解決以上的問題，我們利用類神經網路提出了一套自動化合成電感的方法，由於類神經網路具有良好的預測能力，能夠由電感尺寸精確估算電感值及品質因素。根據模擬或量測的資料，作為倒傳遞類神經網路學習的依據，可以訓練出精確且計算快速的類神經模型，利用這些類神經模型可以加快電感設計的時程，依使用者的需求自動合成出矩形螺旋電感，同時保有電磁模擬軟體的準確性。

This thesis brings forward an automatic inductor synthesis tool at microwave band according to the user’s specification by using back-propagation neural networks (BPNNs). Inductor is one of critical component that dominate the overall performance of RF devices in RF design, and inductor design is often one of most difficult and time consuming process. In order to solve the above problem, we propose the method of automatic inductor synthesis using neural networks. The results of EM simulation or measurement of inductors served as a basis for neural network training with suitable training procedure. The neural networks then can predict inductance and Q-value for a given inductor layout. These neural models can speed up the inductor design by automatically generating the layout of planar spiral inductors according to the user’s specification.