Analog circuit design is a time-consuming process, demanding full attention of an
experienced designer. For decades, various computer methods have aided circuit designers
in their everyday work. The design process was revolutionized with the apperance
of the SPICE simulator, which takes care of complex numerical calculations in
order to predict the actual circuit behaviour. Before SPICE, all the calculations had
to be carried out by the designer manually.Motivated by ever-increasing demands on
circuit performance, lack of capable engineers, and tightening deadlines, computer science
has developed an aiding tool called parameter optimization,which has been used
eectively in a designer’s workow. This automated procedure is capable of nding
optimal numerical parameters (such as resistance, capacitance, transistor gate width
and length) for given conguration of elements (i. e. , topology). However, parameter
optimization alone is often not enough to meet all the design criteria. In this case, a designer
has to choose another topology or design a new one to full the requirements.
This calls for the next logical step in developing methods for computer-aided circuit
design.
In this doctoral thesis,we developed a method capable of automated synthesis of both,
the topology aswell as parameters of analog electrical circuit. Based on a high-level circuit
behaviour description (such as desired gain, damping, or bandwidth), we employ
an evolutionary process to explore the topology and parameter space. The method
nds the solution within the boundaries set at the beginning. In the dissertation, we
rst discuss existing solutions for automated circuit synthesis, highlighting their advantages
and drawbacks. One of the known problems of most existing techniques is
bloat, an uncontrolled growth of the evolving circuit. One of the main challengeswhen
designing a tool for automated circuit synthesis is therefore developing a circuit topology
representation that will allow large enough search space while limiting bloat
as much as possible. In this thesis, we propose a circuit representation in a form of a
connection matrix. We developed operators for topology modication, based on the
idea of genetic material exchange. Our topology description allows topology buildingblocks
to be either simple two-pole electrical elements or elaborate subcircuits with
multiple connection terminals (or both).Abrute-force exploration of all possible topologies
combining even the smallest of sets of building-blocks is infeasible even using
the most state-of-the-art computer hardware. We developed an algorithm based on
the principles of natural evolution in order to nd the solution within a limited time
budget.
We utilize two methods to discriminate the individuals among each other, single- and
multi-objective sorting. Single-objective sorting proved to be successful for synthesis
of simple circuits, while multi-objective sorting was necessary to successfully develop
more complex topologies. Using multi-objective sorting, population diversity is
maintained automatically, with much less chance of the method to stuck in a local minimum.
When we design topologies with a large number of numerical parameters, we
use an additional parameter optimization method PSADE to nd even more optimal
values of these parameters.
We developed a novel combined technique, based on theNSGA-II multi-objective sorting
method and the PSADE parameter optimizer. This combination proved successful
in maintaining a good balance between the population diversity on one hand and
convergence speed at the other. Our hybrid algorithm is able to improve an existing
topology or it can invent a circuit from scratch.
We also had to develop appropriate cost functions capable of steering the evolution
towards a successful topology as well as helping to tune the numerical parameters.
We successfully tested our approach by synthesizing the lters (both passive and active)
and a voltage reference for BiCMOS technology. Although the proposed method
does not entirely replace an experienced human designer, it can signicantly speed up
his/her work and helps to explore new design ideas.
|
