This work is a contribution to high level synthesis for low power systems. While device feature size decreases, interconnect power becomes a dominating factor. Thus it is important that accurate physical information is used during high-level synthesis [1]. We propose a new power optimisation algorithm for RTlevel netlists. The optimisation performs simultaneously slicingtree structure-based floorplanning and functional unit binding and allocation. Since floorplanning, binding and allocation can use the information generated by the other step, the algorithm can greatly optimise the interconnect power. Compared to interconnect unaware power optimised circuits, it shows that interconnect power can be reduced by an average of 41.2 %, while reducing overall power by 24.1 % on an average. The functional unit power remains nearly unchanged. These optimisations are not achieved at the expense of area.
When working in the well established field of register transfer (RT) level modelling, one will soon find out, that doing RT level leakage estimation is by no means a simple modification of timing, area or dynamic power modelling. There are several factors, making leakage macro modelling a unique and completely new challenge: Leakage is barely state dependent from an RT level view. Instead, dynamic parameters highly influencing leakage are temperature ϑ (dominating subthreshold leakage) and supply voltage V DD (having highest influence on gate and pn-junction leakage). Process variations also are a new challenge for RT models. All leakage currents are highly dependent on variation of certain process parameters [1, 2]. Especially the high sensibility to magnitude and distribution of lowest level technological parameters prohibits a simple function-like 'parameters-in estimation-out' model interface. When being used for design space exploration, in-depth information like spread of channel doping concentration may not be available. Instead, RT level leakage models have to integrate parameter distribution prediction and estimation of the resulting leakage currents. Low leakage power management techniques, as adaptive body biasing (ABB) or power gating have large impact on leakage currents. But in contrast to direct leakage reduction techniques as high-k oxides, they have a variable influence on the leakage being controllable from outside an RT component and thus cannot be abstracted within the model. A leakage model supporting leakage power management will thus have to regard these techniques as input parameters. Parameter variation as well as power management techniques will not just affect the leakage current of a component, but also its timing. For a single transistor, the subthreshold leakage, its body potential controlled by ABB, and its operation speed all are highly correlated. Thus a holistic leakage model regarding parameter distribution and power management techniques will have to communicate and share information with an adequate delay model. Only this way, the power-delay correlation, being characteristically for sub-100nm systems and having largest impact on yield, can be described accurately. This thesis will motivate, develop, and evaluate an RT level macro modelling methodology satisfying all these demands. The estimation framework is embedded into the commercial behavioural-to-RT synthesiser tool PowerOpt, where the model predictions guide the synthesis of a SystemC/C++ system specification. The framework consists of the model itself, a floorplan based temperature and voltage drop model, and a variation engine. This engine is translating high level metrics into parameter variation measures, which are then used for leakage and performance prediction. This way, the vii Contents high correlations between leakage and performance [3] as well as the thermo-electrical coupling and electro-electrical coupling [4] which are characteristic for today's devices can be accurately described and thus also automatica...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.