Richard Hamming.
I would like to thank all my friends at the Computer and Control Laboratory for providing encouragement and a very exciting work environment, especially to Evandro, André, Germano, French and Mamdouh.
I am very thankful to all my good friends that supported me throughout this work, especially Simona, Cezar, Ana, Isadora, Eduardo, Ana Raquel, Claudio, Luis, Mayrá, Edmundo and Bani.
I also would like to thank my supervisor, Dr. Les Walczowski, for his supervision and for the freedom he allowed me in conducting my work.
Finally, I would like to express my gratitude to my sponsor, the CNPq - National Council for Research an agency of the Brazilian Federal Government, for the financial support for this work.
The Agents system generates the mask level layout of full custom CMOS, BICMOS, bipolar and mixed digital/analogue leaf cells. Leaf cells are subcircuits of a complexity comparable with SSI (Small Scale Integration) components such as small adders, counters or multiplexers. The system is formed by four server programs: the Placer, Router, Database and Broker.
The Placer places components in a cell, the Router wires the circuits sent to it, the Database keeps all the information that is dependent upon the fabrication process, such as the design rules, and the Broker makes the services of the other servers available.
These servers communicate over a computer network using the TCP/IP Internet Protocol. The Placer server receives from its client the description and netlist of the circuit to be generated using EDIF (Electronic Design Interchange Format). The output to its client is the layout of the circuit (no virtual grid is used), again codified in EDIF.
The concept of agents as software components which have the ability to communicate and cooperate with each other is at the heart of the Agents system. This concept is not only used at the higher level, for the four servers Placer, Router, Broker and Database, but as well at a lower level, inside the Router and Placer servers, where small relatively simple agents work together to accomplish complex tasks. These small agents are responsible for all the reasoning carried out by the two servers as they hold the basic inference routines and the knowledge needed by the servers. The key concept is that competence emerges out of the collective behaviour of a large number of relatively simple agents. In addition and integrated with these small agents, the system uses a genetic algorithm to improve components' placement before routing.
Next
Contents
Talk to me