Optimization of Imprecise Circuits Represented by Taylor Series and Real-Valued Polynomials
Optimization of Imprecise Circuits Represented by Taylor Series and Real-Valued Polynomials
复制标题
泰勒级数和实值多项式表示的不精确电路的优化
DOI：
10.1109/tcad.2010.2049154
复制
发表时间：
2010
影响因子：
2.9
通讯作者：
Z. Zilic
中科院分区：
文献类型：
--
作者：
Yu Pang;K. Radecka;Z. Zilic
算术电路通常与规格不完全匹配，导致在允许的不精确度内实现不同的实现。我们提出了一种技术来搜索给定误差范围的最便宜的定点实现。该方法在实际应用中很实用，克服了传统精度分析的悲观主义，因为它允许同时选择多个字长，甚至主要基于泰勒级数的一些函数逼近。从实值表示开始，例如泰勒级数，我们依靠算术变换来探索最大不精确性，通过分支定界搜索算法来研究不精确性。我们还采用了一种新的紧界区间方案，并推导了一种精度优化算法，该算法探索多个精度参数以获得最小面积成本的实现。
Arithmetic circuits in general do not match specifications exactly, leading to different implementations within allowed imprecision. We present a technique to search for the least expensive fixed-point implementations for a given error bound. The method is practical in real applications and overcomes traditional precision analysis pessimism, as it allows simultaneous selection of multiple word lengths and even some function approximation, primarily based on Taylor series. Starting from real-valued representation, such as Taylor series, we rely on arithmetic transform to explore maximum imprecision by a branch-and-bound search algorithm to investigate imprecision. We also adopt a new tight-bound interval scheme, and derive a precision optimization algorithm that explores multiple precision parameters to get an implementation with smallest area cost.