RBF類神經網路之超大型積體電路設計

Abstract

多維放射函數被廣泛的運用在數種類神經網路模型中，並且在模糊邏輯系統中也有許多的應用。但是，在傳統查表的硬體實現法中，必須要多增加額外的硬體因此而無法配合在高速的實際運用中。在現今的科技中，用超大型積體電路去實現類神經網路與模糊邏輯系統，已經使其能夠與複雜的計算機系統溝通配合。 在本論文中，我們使用超大型積體電路技術來實現放射狀基礎網路函數（RBFNN）。其中的放射函數，我們將其包含在硬體中，使得整個架構能夠比傳統的查表法快速。在此，我們完成網路的向前傳遞部分，並且使的輸入層、中間隱藏層與輸出層能夠隨需要而調整。並在有限狀態機中，將資料路徑管線化，使其運算更快。最後將整個架構運用於一個模糊歸屬函數的映射例子中來證明其功能。

Multi-dimensional radial functions (RFs) are widely used in several neural network schemes and may have interesting applications also in fuzzy logic based systems. Unfortunately their classical look-up table hardware implementation needs an external board that does not allow high speed real world applications. At the state of art, digital VLSI implementations of Neural Networks (NN) and Fuzzy Logic based systems (FLS) easy to interface with more complex computational systems such as workstations or microprocessors are made possible. In this thesis, we propose VLSI techniques to implement the Radial Basis Function Neural Network (RBFNN), many RF generators can be integrated on a single chip giving rise to a computational system which is faster than classical look-up table implementations, this architecture is used to implement the forward step of a RBFNN. This input, hidden, and output nodes are adjustable. The Finite State Machine (FSM) method is used to pipe the datapath for higher speed. We also use this design in fuzzy membership function mapping to verify the functionality of the architecture.