Topical References

System Specification, Representation and Language Models

A complete representation of an embedded system should include its functionality as well as its behavior over time (temporal properties). Often this functionality is specified at several levels of abstractions such as an structural connection of blocks, or in terms of its (algorithmic/RTL) behavior. Further, a system specification may contain (intentionally or inadvertently) information or directives for choices to be used in its implementation. You may start by looking at popular languages used for system specifications, particularly reactive real-time systems.

On specification of reactive systems, the following references will be useful [BL90, BW90]. Synchronous programming languages have been proposed for modeling reactive systems under the assumption that computation time is much less than communication and event interval times. A good introduction to the subject is provided by Berry in [BB91] and Esterel language is described in [Ber89]. Nicholas Halbawachs discusses use of synchronous programming using Argos in his Kluwer book [Hal93]. Specification and Description Language (SDL) [Se87, SSR89] was adopted by CCITT and has been used in the telecommunications industry, primarily for protocol specifications.

Abstract models for synthesis based on control and data.

Abstract models for embedded systems broadly fall into two classes based on their orientation towards data or control flow. There is some debate in the community about finding a model that can be used to efficiently perform both data and control-oriented transformation. In addition, appropriate representation of timing behavior is needed. The following references will provide a starting point on this subject [CH86, HLN 90, Har92, CGH 93a, LVBA93]. Gajski in his latest book [GVNG94] discusses use of SpecCharts as an effective specification model which are automatically translated into VHDL code.

Use of transformations in architectural synthesis.

Transformations have been formalized and successfully applied in case of signal-processing (or data-oriented) algorithms. Parhi in  [Par89] provides a good review of these techniques. Schoofs et al in  [SGM94] describe signal type optimization again for DSP applications. Recent work on code optimization can be found in [Let al95]. Control transformations using control simplification and behavioral don't cares are discussed respectively in [CM94] and [Ber92a, Ber91].

System performance analysis using abstract models

Performance analysis consists of either analysis using models such as Petri Nets or others [BK90, Kob78, Sha79, Mol82]. For embedded systems, the determination of timing properties of software has been a challenging problem and is studied in by Shaw and others in [WS82, Mea89, Par92, PS90, PK89, Sha89, Sha91, YEBH93]. Recent work on this subject can be found in presentations [LM95] on software analysis (session 27) in this conference.

Constraint modeling and feasibility analysis

Constraint modeling refers to specification, capture and analysis of constraints inorder to deliver desired timing behavior or to observe bounds on resource utilization. Feasibility in time-constrained systems has often been addressed in the context of operation schedulability under given timing assumptions and processor characteristics. The following references point to literature in capturing constraints, particularly timing constraints and analysis techniques [Das85, CK86, GM94b, Lau89, Lei80, LW82, LL73].

Compilation issues for hardware/software systems

When compiling for hardware and software from a single specification, the optimizations required use different metrics and cost criteria. The following discuss the anatomy of compilation and issues [Haa91, KW88].

Hardware/software interface specification and analysis

Modeling,synthesis and analysis of interface between system components (particularly, hardware and software) has been a subject of research for quite some time [BK87, COB92, FKCD93, KLM93, AB91]. In recent years, the focus seems to have shifted to asynchronous hardware elements to effectively represent and synthesize interface [MD92].

Hardware-software system architectures

The choice of a suitable architecture for hardware-software systems is an open issue. The following publications list some of the architectures being pursued [SS93, BRV89, CvSRM91, CGH 93b, GK89, HHW89, HHR 91, HD92, PNRK93, SB91, SBB92, Wal90].

System partitioning into hardware and software

Choice of some architectures requires a suitable division of system functionality to be implemented in either hardware or software. Work in this area is still in infancy. Some techniques are proposed in [BRX93, GM92, HK93, IOJ94, MABC92]. A survey of these techniques and partitioning based on SpecCharts are presented in [GVNG94].

Hardware-software co-synthesis

Co-synthesis, that is, combined synthesis of both hardware and software has been proposed as a viable alternative to ad hoc design methods for embedded systems [CGH 93b, EHB93, GM93, KL93, WWD92, RWH 93, BT93, Wol94].

Software synthesis for embedded systems

Software synthesis requires not only observance of constraints on physical memory size but also satisfaction of constraints on timing performance. Often software generation for a given target architecture requires linearization of operations. Such linearization may adversely affect the satisfiability of the imposed timing constraints. How can you determine if a valid serialization for a given set of timing constraints exists? If so, how do you define and choose an optimum/optimal serialization? The following references detail the issue and point to some answers [CB94, GRVD87, GCM94, CGH 95].

Optimization of machine code for embedded systems is a more general problem than code optimization for general-purpose computing systems, due to the additional dimension of detailed timing constraints and response time requirements [Cha82, FOW87, GG90, GM94a].

Hardware synthesis for embedded systems

Hardware synthesis from language descriptions has come a long way in recent years due to the understanding of and advances in behavioral and RTL synthesis techniques. Recent book by De Micheli [Mic94] provides a good description of the tasks and approaches for digital hardware synthesis. The following are some pointers to the literature on the subject [TLW 90, RWB92, CW91, MKMT90, JOI93, KM90b, KM90a, KM91, Nag79, Ber92b].

Simulation of hardware/software systems

Simulation of systems containing hardware and software components presents special problem due to the need to coordinate events in two very different computation domains and yet maintain efficiency in simulations. These references point to the recent advances in this subject [BHLM91, Fis91, GCM92, tHM93, OH93, Sch90, FLLO95].