Intelligent Retrieval for Component Reuse in System-On-Chip Design Andrea Freßmann, Rainer Maximini, Martin Schaaf University of Hildesheim, Data- and Knowledge Management Group PO Box 101363, 31113 Hildesheim, Germany {fressmann, r_maximi, schaaf}@dwm.uni-hildesheim.de Jasmin Franz, Ralph Traphöner empolis GmbH Europaallee 10, 67657 Kaiserslautern, Germany {jasmin.franz, ralph.traphoener}@empolis.com 1 Introduction and Motivation Systems on Chip (SoC) are electronic circuits that combine multiple functionalities on single chip and are the building blocks for a wide range of consumer electronics such as mobile phones or DVD players. In contrast to traditional integrated circuits, which are individually designed and highly optimized, the focus of SoCs lies on high integration and cost reduction. Nowadays, technology allows SoCs to become very complex and, due to the immense functionality aggregated on one chip, verification is time consuming. Very similar to newer advancements in software engineering, a possible solution for designing SoCs is to reuse already existing and fully verified components so called Virtual Components or Intellectual Properties (s). Within the project Q: Qualification for Efficient Design Reuse (12/2000 11/2003, see https://www.ip-qualifikation.de) funded by the German Ministry of Education and Research (BMBF), the partners AMD, Fraunhofer Institut für Integrierte Schaltungen, FZI Karlsruhe, Infineon Technologies, Siemens, sci-worx, Empolis, Thomson Multi Media, TU Chemnitz, University of Hildesheim, University of Kaiserslautern, and University of Paderborn worked together in searching and developing new design methodologies, standardized description techniques, and tools specialized for the design of SoCs.
Retrieval Entry Check Purchase Library Entry check failed Increasing costs for dealing with s Figure 1: Selection Process Guided by the overall selection process depicted in Figure 1, the main objective of Q is to provide a comprehensive support for designers in selecting and integrating s into their actual design. The applied scenario is a Virtual Marketplace for s that offers a variety of web services for finding, purchasing, and transferring between different libraries. As the amount of s offered in the World Wide Web grows steadily, the retrieval of potential s suitable for a given design situation is a substantial part of the Q project and will be in the focus of this article. In Q structural Case-Based Reasoning (CBR) technology is improved to cope with the three main requirements of the retrieval application. The first requirement was to develop a standardized characterization format that is sufficient for the automatic exchange and retrieval of s in a distributed web-based environment. Here, it was essential to formalize knowledge about an itself using a well-founded semantic that can be used by the CBR retrieval engine. The developed Characterization Language (CHL) will be briefly presented in section 2. Second, the handling of highly parameterized s with dependencies between their parameters was required (see section 3). Furthermore, users need more confidence in the results proposed by the retrieval system. Hence, an explanation component is incorporated to make the selection process more transparent (see section 4). In section 5 the developed overall Broker Tool Suite is presented that offers a comprehensive solution concerning the application scenario. The Broker Tool Suite has been completely implemented on top of the commercial retrieval engine orenge. 2 Language for Characterizations Structural CBR requires s to be described by a characterization constructed from a previously developed domain vocabulary. As depicted in Figure 2, the Characterization Language (CHL), an instance of XML Schema, is a meta-model that can be tailored by Providers toward CHL Profiles representing a format that each characterization must follow. CHL Profile, Content Format and corresponding instances build the Q Format for transferring the descriptions about s as well as the itself (files in hardware design languages) in web-based settings.
{ Characterization Schema (Domain Vocabulary, CHL Profile) CHL VSIA + OpenMORE Attributes Constructs for defining class hierarchies Conceptualization of units FCT + MSC Content Format instanceof Q Transfer Format (shortly Q Format) Characterization Content Figure 2: CHL Container Overview CHL provides a flexible based representation language, whose set of potential properties is constituted of two different attribute types [1; 2]: Application attributes that refer to properties important to decide about the applicability of an in a given design situation Quality criteria that characterize the and its deliverables according to its quality (e.g. code coverage, guidelines, simulations). Both types are subjects to current standardization efforts of the VSIA (Virtual Socket Alliance). For the application attributes, the document of Virtual Component Attributes (VCA) [5] presents a variety of attributes together with proposals for their syntactical representation. For the quality criteria, the OpenMORE Assessment Program from Synopsis provides properties for quality assessment [4]. An instance of CHL is called profile and contains meta data of one specific. Furthermore, the Characterization Schema facilitates definitions of the domain vocabulary, e.g. conceptualizations of units or taxonomies [3]. For example, taxonomical properties can be defined as XML constructs that can be references by attribute types of profiles for classifying the corresponding. 3 Retrieval of Parameterized s s are design objects that have been particularly developed for the purpose of flexible reuse. The dominant characteristic of such a design is the fact that it consists of a large number of different parameters by which its concrete function can be determined when the design is compiled to a specific layout on the chip. Examples of such parameters are the width of buses or sizes of memories and pipelines. Further, a design usually imposes several dependencies among its parameters. Also characteristic attributes (e.g., power consumption of the implemented design) depend on the specific assignment of the parameters. The important challenge for retrieval is
to consider these dependencies during case-based retrieval. As a major result of the project, the traditional case representation in CBR has been extended towards so called generalized cases [12, 7]. Generalized cases allow variables in the case representation as well as constraints among these variables. Hence, a generalized case does not only cover a point of the representation space of cases but whole subspace of it. This subspace describes the range of situations in which an is applicable through proper assignment of its parameters. The important capability of CBR that makes its use attractive for retrieval in general and for retrieval in particular is its use of similarity measures. Similarity measures provide a means to approximate the utility of a certain object to be retrieved if it does not exactly match the stated query. In Q, particular similarity measures for s have been acquired from an provider. However, the extension of case representations to generalized cases also required developing new methods similaritybased retrieval. Within Q, a converter has been developed that samples the subspace spawned by the generalized cases and creates a number of point cases that are distributed reasonably well within the subspace [10]. A commercial CBR retrieval engine such as orenge can then be used for generalized cases at the cost of accepting a certain retrieval error that depends on the sampling density. An alternative approach that emerged from this application consists of transforming the similarity measure and a generalized case to a mathematical optimization problem that can be solved with common optimization algorithms [8, 9]. This asks for a tighter integration of optimization and CBR for handling more complex retrieval and approximation problems and will be an important line of future research. 4 Selection Support by Explanations For highly complex domains, such as electronic designs, presenting only an ordered list of retrieval results is not sufficient. users have a demand for additional information and explanations making the proposed results more transparent. By presenting additional explanations the confidence in the result set increases and possible deficiencies can be revealed and corrected. Generating a report for each proposed with respect to the user s query is a fundamental demand for explanation support. Graphical rendering of the attributes with highest impact on the assessment and textual explanations aids the designer in getting a quick overview of the. Such explanations may be design rationales of the domain expert or best practices associated to attributes specified by other users and are part of the knowledge representation [11]. But before s are considered as candidates the result set is analyzed. If the coverage of attributes specified in the query and in the case is low a deficiency in the case base is indicated, which is traced back to conflicting attributes posed in the query. For aiding the users to improve and refine the query, these attributes or attribute combinations are analyzed and presented. But also the query can have deficiencies, which is the case if a high number of suitable matching cases are retrieved. Here, appropriate analysis techniques determine
the most decisive unspecified attributes in order to propose a refinement of the query. Alternatively, a variant of the CobWeb algorithm is used to cluster the result set hierarchically to facilitate the users navigation through the set. Figure 3 gives an overview of the developed techniques as they are implemented in the retrieval system. Basically, the strategy which technique to choose depends on the number of suitable matches. No Suitable Matches Few Suitable Matches Many Suitable Matches Deficiencies in the Case Base Deficiencies in the Query Analyzing the Query for Conflict Attributes Analyzing the Similarity Distribution Analyzing the Most Decisive, Unspecified Attributes Cluster Analysis Presentation of the Attributes Presentation of the Result Set Presentation of the Attributes Presentation of the Clusters Figure 3: Explanation of the Result Set 5 The Broker Tool Suite The Broker Tool Suite is a set of tools an provider can use for developing, deploying, and tailoring a retrieval service to his/her specific needs. The suite contains tools for specifying and maintaining assets and for capturing or maintaining domain knowledge. The core is the CBR-based Open Retrieval Engine orenge [6] due to its flexibility and its modular concept. The architecture of the suite is depicted in Figure 4. Provider Consumer Marketing Developer Broker Information Service Retrieval Service Explanation Service Case Base CHL Web Services Creator Case Base Manager Constraint Manager Retrieval Manager Case Manager Consumer Suite Developer/Provider Suite Figure 4: The architecture of the Broker Tool Suite
The Consumer Suite is the web-based user front end for retrieval (see http://demo.dwm.uni-hildesheim.de:8080/ipq2/demo). It includes three services: Retrieval, Explanation and Information Service. The Retrieval Service enables the user to formulate queries, consisting of detailed technical requirements entered into a structural form. Further it allows to specify individual preferences on the level of the different application and quality criteria. The queries are sent to the CBR retrieval kernel and the results are analyzed and presented by the Explanation Service, described in the previous section. The Information Service is less for the user but more for the provider because it informs the providers about users needs. The Information Service captures the queries posed by the users and transmits them to the respective providers to enable the provider to better focus their future product development. Further, if the user has asked for a particular, the affected provider automatically gets a respective notification. The Developer/Provider Suite contains the case base and several tools to manage the vocabulary and the similarity measures. For defining the object-oriented vocabulary, designers utilize the orenge Creator to define the domain model, which contains all of the relevant attributes of CHL. Furthermore, the Creator can be used to define similarity measures via functions, tables, or taxonomies. The Case Manager uses the vocabulary for capturing and managing the case base. Therefore, three tools are integrated: The Case Base Manager allows developers to capture case knowledge as well as to define constraints among attributes. Such defined generalized cases can be converted into point cases by the Constraint Manager. This is a client for the converter process that is described in section 3. Finally, the Retrieval Manager is a front end for connecting local or remote case bases and for performing retrieval activities on the part of the developers. 6 Conclusion In this article, a brief overview about the achievements of the Q Project with focus on retrieval is given. The Broker Tool Suite offers a comprehensive framework for capturing, maintaining, and searching s with the aim of reuse. It integrates advancements that contribute to two different research areas. The representation of generalized cases and the implementation of the explanation component exceed the current state-of-the-art of CBR systems and will be integrated by empolis into their next orenge release. For the development of SoCs, CHL provides a standardized format that makes use of current web technology like XML/XML Schema. It enables providers to characterize and publish their s in web-based environments. Together with the highquality retrieval the basics for efficient reuse and exchange have been established.
Reference 1. Schaaf, M., Visarius, M., Bergmann, R., Maximini, R., Spinelli, M., Lessmann, J., Hardt, W., Ihmor, S., Thronicke, W., Franz, J., Tautz, C., Traphöner, R.: CHL - A Description Language for Semantic Characterization, Forum on Specification & Design Languages, September (2002). 2. Schaaf, M., Maximini, R., Bergmann, R., Tautz, C., Traphöner, R.: Supporting Electronic Design Reuse by Integrating Quality-Criteria into CBR-Based Selection, Proceedings 6th European Conference on Case Based Reasoning, September (2002). 3. Schaaf, M., Freßmann, A., Spinelli, M., Maximini, R., Bergmann, R. (2003). A Knowledge Representation Format for Virtual Marketplaces, 5th International Conference on Case-Based Reasoning 2003 (ICCBR 2003). 4. Synopsis Inc., Mentor Graphics Corporation: OpenMORE Assessment Program for Hard/Soft Version 1.0, http://www.openmore.com (2001). 5. VSI AllianceTM, Virtual Component Transfer Development Working Group: Virtual Component Attributes (VCA) With Formats for Profiling, Selection, and Transfer. Standard Version 2.2 (VCT 2 2.2) (2001). 6. empolis knowledge management GmbH: The orenge Framework - A Platform for Intelligent Industry Solutions Based on XML and Java. Whitepaper, empolis knowledge management GmbH, Kaiserslautern (2001) 7. K. Maximini, R. Maximini, and R. Bergmann. An investigation of generalized cases. In Proceedings of the Fifth International Conference on Case-Based Reasoning (ICCBR 03), Lecture Notes in Artificial Intelligence, 2689. Springer Verlag (2003), pages 261 275. 8. B. Mougouie and R. Bergmann. Similarity assessment for generalized cases by optimization methods. In Proceedings of the European Conference on Case-Based Reasoning (ECCBR-02). Springer, 2002. 9. A. Tartakovski and R. Maximini: Similarity Assessment and Retrieval of Generalized Cases. In Proceedings of Workshop Wissens- und Erfahrungsmanagement (FGWM 2003), Karlsruhe, 6. - 8. Oktober 2003, http://km.aifb.uni-karlsruhe.de/ws/llwa/fgwm 10. R. Maximini and A. Tartakovski, Approximative Retrieval of Attribute Dependent Generalized Cases. In Proceedings of Workshop Wissens- und Erfahrungsmanagement (FGWM 2003), Karlsruhe, 6. - 8. Oktober 2003, http://km.aifb.unikarlsruhe.de/ws/llwa/fgwm 11. M. Spinelli and M. Schaaf, Towards Explanations for CBR-based Applications, In Proceedings of Workshop Wissens- und Erfahrungsmanagement (FGWM 2003), Karlsruhe, 6. - 8. Oktober 2003, http://km.aifb.uni-karlsruhe.de/ws/llwa/fgwm 12. Bergmann, R. and Vollrath, I. (1999). Generalized Cases: Representation and Steps Towards Efficient Similarity Assessment. In W. Burgard, Th. Christaller & A. B. Cremers (Eds.) KI-99: Advances in Artificial Intelligence, Lecture Notes in Artificial Intelligence, 1701, Springer, 195-206.
Andrea Freßmann studied Mathematics and Sports at the University of Augsburg, Germany, where she took her lectureship examination at 1999. Moreover, she studied Information Technology at University of Hildesheim, Germany, and received her M.Sc. in 2002. Since then she is working in the Data and Knowledge Management Group of Prof. Dr. Ralph Bergmann at the University of Hildesheim. Rainer Maximini studied Computer Science at the University of Kaiserslautern, Germany, where he received his Diploma in 2001. Subsequently, he had a full time researcher position in the group "Knowledge based Systems and Expert Systems" of Prof. Dr. Michael M. Richter at the University of Kaiserslautern. Since the end of 2001 he is working in the group "Data and Knowledge Management" of Prof. Dr. Ralph Bergmann at the University of Hildesheim, Germany, where he is currently finishing his PhD. Martin Schaaf studied Computer Science and Business Management at the University of Kaiserslautern, Germany, where he received the Diploma in 1999. From 1999 to 2002 he had a full time researcher position in the group "Knowledge based Systems and Expert Systems" of Prof. Dr. Michael M. Richter at the University of Kaiserslautern. Since 2002 he is working in the group "Data and Knowledge Management" of Prof. Dr. Ralph Bergmann at the University of Hildesheim, Germany, where he is currently finishing his PhD. For the winter term 2001, he had a sesional instructor position at the University of Calgary, Canada, reading two lectures on Software Engineering. Dr. Jasmin Franz coordinates and manages European and national Research Projects on behalf of empolis GmbH. Her primary interests lie in the areas of innovative business applications and standards for knowledge representation. Jasmin Franz holds a Ph.D. in Linguistics from the University of Würzburg in Germany where her studies concentrated on Linguistic Information Processing. She has many years of experience in the field of semantics and text comprehensibility. Prior to her work at empolis GmbH, she held the position of representative for the managing director in an international translation company, in addition to her responsibilities as project manager for technical documentation and software localization.
Ralph Traphöner studied Computer Science in Kaiserslautern and was a co-founder of tecinno GmbH in 1991. Today, the company is part of empolis GmbH, a part of Bertelsmann. As Head of Research, he is responsible for all research activities of the company. The focus of his work is on Case-Based Reasoning, Knowledge Management and the Semantic Web.