Utilize the Shift-and-Add Architecture to Approximate Multiple Nonlinear functions
Utilize the Shift-and-Add Architecture to Approximate Multiple Nonlinear functions
复制标题
利用移位加法架构来逼近多个非线性函数
DOI:
10.1002/cta.2994
复制
发表时间:
2021
期刊:
影响因子:
--
通讯作者:
Yuchun Chang
中科院分区:
文献类型:
--
作者:
Botao Xiong;Yuanfeng Sui;Zhi Jia;Sicun Li;Yuchun Chang
The piecewise linear approximation method is used to estimate the non‐linear functions, which is implemented by the shift‐and‐add architecture consisting of three blocks: “adder tree,” “multiplexer,” and “segment index encoder.” Compared with state of the art, the design flow in this letter enables the “adder tree” and “multiplexer” blocks can be shared by multiple nonlinear functions. In other words, the shift‐and‐add architecture can approximate multiple nonlinear functions with a slight overhead. The fused Log and Antilog converter demonstrates the feasibility of the proposed method. The Antilog function is implemented by the Log converter at the cost of 14% cell area and 6% delay.