Publications

PhD Thesis


< back to PhD thesis

« High level modeling of heterogeneous systems, analog/digital interfacing ».

Author: F. Cenni
Advisor: E. Simeu
President of jury: P. Girard
thesis reviewer(s): Ch. Grimm, B. Granado,
These de Doctorat Université de Grenoble
Speciality: Micro et Nano Electronique
Defense: April 06 2012
ISBN: 978-2-84813-185-6

Abstract

The tremendous development of technology and methodology for electronic circuit design is leading to systems with high levels of complexity. Both the ever-shrinking electronic systems and improving efficiency of computer aided design (CAD) tools enable the design of extremely complex multi-tasking systems that are heterogeneous mixed-signal Systems on Chip (SoCs). The cohabitation of many physical domains such as mechanical, chemical, optical or magnetic in recent SoCs justifies the big effort that takes place in CAD tools for designing such systems. Nowadays, the design of the individual components is usually well understood and optimized through the usage of a diversity of CAD or Electronic Design Automation (EDA) tools, design languages, and data formats. These are based on applying specific modeling/abstraction concepts, description formalisms (also called Models of Computation (MoCs)) and analysis/simulation methods. The designer has to bridge the gaps between tools and methodologies using manual conversion of models and proprietary tool couplings/integrations, which is error-prone and time-consuming. The interaction among the huge quantity of individual components in recent systems is of vital interest for the overall system to function in compliancy with the specifications requested by the customer. At early stages of the design phase the interaction among the variety of Intellectual Properties (IPs) integrated on the system can only be validated through the simulation of a system model. Different levels of abstraction can be defined when modeling some IPs, a high abstraction means capturing only the gross behavior of the system and keeping a reduced accuracy of the model with respect to the real device. Analog and digital parts coexist in one system, it is necessary to model both of them for performing an overall simulation of the system. Furthermore digital parts involve processors with their embedded softwares. A crucial issue is nowadays becoming the validation of the interaction among AMS parts together with the digital and the embedded software. For such a target a recent C++ based common design and simulation platform is establishing itself. Such a platform is based upon the SystemC (IEEE 1666) kernel and called SystemC AMS, it allows creating and refining a virtual prototype of the overall system on a high level of abstraction. This is done by supporting different description formalisms, also called Models of Computation (MoCs), for arbitrarily describing many types of behavior and interaction. This makes possible the exploration of different architecture options, estimation of the performance, validation of re-used parts, verification of the interfaces between heterogeneous components and inter-operability with other systems. This thesis deals with the modeling of analog and mixed-signal physically heterogeneous systems. In particular it is presented a study on different techniques for extracting behavioral models at different levels of abstraction and computational weights. Although the tools developed for behavioral model extraction are mostly based on the AMS extension of the SystemC kernel, the methodology can be applied to other Analog Hardware Description Languages (AHDL) such as Verilog-AMS and VHDL-AMS. These techniques are regrouped in three branches. Firstly, a behavioral modeling technique from a schematic netlist entry description is studied and automatized in order to extract additional desired information under the guise of state space equations. Secondly, techniques based on analytical fittings of frequency responses are explored for either reducing the model order or for identifying analytical simulateable models starting from analysis carried out with other application-specific CAD tools. Finally, system identification techniques are examined for the extraction of black-box models from empirically obtained data, either from simulations of accurate models or from measure data. A proof-of-concept library implemented using SystemC AMS shows the applicability of the methodology and an LNA case study is developed for a power-minimization specific application. The mixed nature of the Ph.D., both academic and industrial due to the collaboration with the STMicroelectronics enterprise, led to concentrate the modeling efforts on a CMOS image sensor case-study. This case-study aims at an overall simulation of an industrial platform for mobile application based on transaction-level modeling using the SystemC kernel. In order to do so, the analog/digital interfacing between the SystemC AMS MoCs and different types of SystemC-TLM coding styles is studied and adapted to the STMicroelectronics' proprietary TLM protocols as a proof of the applicability of the SystemC AMS based methodology to the industrial design flow. The image sensor model is abstracted at different levels using different SystemC AMS Models of Computation showing impressive simulation time performances with respect to the low level models (notably the VHDL-AMS based model). The virtual platform is currently being used for an early validation of the image correction algorithms and embedded software, this will result in an improved reliability of the product.

pdf pdf

Other localisation