定寬乘法器(fixed-width multiplier)本質上屬於近似乘法器(approximate multiplier),可用於降低乘法運算所需之功率消耗與電路面積。因此定寬乘法器被廣泛應用於數位信號處理應用中,例如多媒體(multi-media)、數位訊號處理(digital signal processing)、類神經網路(neural network)等相關應用領域,在乘法器之輸出入資料格式本身便適用定寬乘法器。 在本文中,我們提出了一種新穎的定寬乘法器設計。所提出的體系結構的主要優點是它使用簡單而準確的方法進行糾正。如同實驗結果,可以同時實現較低的功率以及較高的精確度。實驗結果顯示,此定寬乘法器設計架構確實可達到低功率及高精確度。
A fixed-width multiplier is essentially an approximate multiplier, which can be used to reduce power consumption and circuit area required for a multiplication operation. Fixed-width multipliers have been widely used in digital signal processing applications, such as multimedia, digital signal processing, neural network, and other related application fields, since the input/output data formats of multipliers in these application are suitable for fixed-width multipliers. In this thesis, we propose a novel fixed-width multiplier design. The main advantage of the proposed architecture is that it uses a simple yet accurate method for error correction. As a consequence, both low-power and high-accuracy can be achieved at the same time. Experimental result consistently show that proposed fixed-width multiplier architecture can achieve both low power and high accuracy.