Theory and Design Method of Clock-less Datapath for Next Generation VLSIs

点击翻译按钮获取中文摘要

In the current and future nanometer-technologies for VLSIs, clock distribution is chip-area consuming and power consuming, and also it makes the timing problem more serious. The clock-less data-path architecture, which excludes clock signal from the data-path part, would be one of the solutions for those problems relevant to clock distribution. The purpose of this research is to show the possibility and to develop design methodologies of the clock-less data-path. One clock-less data-path architecture considered in our study is the one which consists of a controller part driven by a clock signal and a data-path part driven by only control signals from the controller (a clock signal is not fed to the data-path part). The other is the one which consists of an asynchronous controller part and an asynchronous data-path part. The results obtained in this research are summarized as follows.1.We have proposed the assignment constraint control signal scheduling considering both the data signal … More delay in the data-path part and control signal delay from the controller to each component in the data-path part. The scheduler has been incorporated into a new synthesis framework called "assignment-centric high-level synthesis" to form a complete synthesis system.2.The difference of arrival timing of control signals is called "timing skew", and it has the potential to improve the speed of the data-path. We have formulated several skew optimization problems, studied there properties and developed optimization algorithms.3.We have proposed an assignment-centric approach to the high-level synthesis for asynchronous data-paths. The major advantages of our method over conventional ones are (1) the efficiency in optimization and (2) statistical performance analysis.4.To improve hardware utilization in the clock-less asynchronous data-path, a novel implementation of asynchronous register has been proposed. The synthesis system has been modified to meet a new register sharing model provided by the new register. We have verified the effectiveness of the new register through synthesis experiments. Less