本論文中，我們提出了一種改良式粒子群演算法，名為自適應解分享粒子群演算法(Adaptive Solution-Sharing Particle Swarm Optimization, ASSPSO)，並應用於射頻積體電路(Radio Frequency Integrated Circuits, RFIC)之螺旋(Spiral)電感最佳化設計。在標準粒子群演算法中每一個體使用共同的慣性權重，且如果適應函數維度較大時，前期收斂速度會十分緩慢，為了解決這個問題，本論文使用唯一認知模型(Cognition Only Model)，以提升前期的收斂速度，並且針對慣性權重提出改良方法，利用粒子與粒子之間的差異度去更新移動方程式，可以更有效的從局部搜索開始，使粒子具有比較好的搜尋能力往全域最佳解移動，並且和幾個已經提出的改良式粒子群演算法做性能的比較，同時也顯示本文所提出的改良式粒子群演算法能有效地改善標準粒子群演算法容易陷入局部最佳解的缺點，最後應用於螺旋電感最佳化設計。在無線通訊系統中，射頻積體電路的特性好壞與電感的品質因數有密切的關係，以低雜訊放大器舉例，現階段以CMOS製程出來的螺旋(Spiral)電感之品質因素較低，使低雜訊放大器指數(NF)增加、增益(Av)下降造成電路效能不佳，所以本論文利用提出的改良式粒子群演算法有效提升電感之品質因數，以提高整個射頻積體電路(RFIC)的效能。;In this thesis, we propose a variant algorithm for Particle Swarm Optimization (PSO) which is called Adapted Solution-Sharing Particle Swarm Optimization (ASSPSO), and applied to optimization of spiral inductor of Radio Frequency Integrated Circuits (RFIC). In standard PSO algorithm, each particle of SPSO using a same equation to update a particle’s velocity. If the dimension is large, the convergence rate will be very slow and get local easily. In order to solve this problem, the proposed ASSPSO uses a cognition only model to enhance the particle’s velocity and according to the distance between particle and the best particle with best fitness to adjust the inertia weight adaptively. For the particle with a better fitness, the inertia weight is decreased; otherwise the inertia weight is increased for particles with inferior fitness. The methods cause the fast convergence ability in pre-convergence and have less time on computing. The performance of ASSPSO is fairly demonstrated by applying sixteen benchmark problems and compared it with several popular PSO algorithms. Finally, the modified PSO algorithm will be applied to optimization of spiral inductor. In wireless communication systems, spiral inductors are the essential component of radio frequency integrated circuits (RFIC). The performance of radio frequency integrated circuit is decided by the quality factor (Q). Taking Low-Noise Amplifier (LNA) for example, the quality factor of LNA is lower in CMOS process. It will lead to a bad performance of circuits. So we will take advantage of modified PSO algorithm on optimization of spiral inductor to make it better.
