DC CIRCUIT ANALYSIS OF INTEGRATED CIRCUITS FOR THE ...

bracebustlingElectronics - Devices

Oct 7, 2013 (3 years and 8 months ago)

157 views

DC CIRCUIT ANALYSIS OF
INTEGRATED CIRCUITS FOR THE
LYRIC PHYSICAL DESIGN
FRAMEWORK
Submitted by Dennis Weston
for the degree of
BSc (Hons) Computer Science
2006
DC CIRCUIT ANALYSIS OF INTEGRATED CIRCUITS FOR THE
LYRIC PHYSICAL DESIGN FRAMEWORK
Submitted by Dennis Weston
COPYRIGHT
Attention is drawn to the fact that copyright of this dissertation rests with its
author.The Intellectual Property Rights of the products produced as part of the
project belong to the University of Bath
(see http://www.bath.ac.uk/ordinances/#intelprop).
This copy of the dissertation has been supplied on condition that anyone who
consults it is understood to recognise that its copyright rests with its author and
that no quotation from the dissertation and no information derived from it may
be published without the prior written consent of the author.
Declaration
This dissertation is submitted to the University of Bath in accordance with the
requirements of the degree of Bachelor of Science in the Department of Computer
Science.No portion of the work in this dissertation has been submitted in sup-
port of an application for any other degree or quali¯cation of this or any other
university or institution of learning.Except where speci¯cally acknowledged,it
is the work of the author.
Signed...........................(Dennis Weston)
This dissertation may be made available for consultation within the University
Library and may be photocopied or lent to other libraries for the purposes of
consultation.
Signed...........................(Dennis Weston)
Abstract
In today's world,integrated circuits,or microchips,are all around us and a®ect
almost everything we do.They are found in all our electrical products from ev-
eryday systems like our computers and our washing machines to the important
anti-lock brakes in our cars.Engineers who design these microchips need to make
sure they are reliable,especially in safety critical systems such as the °ight sys-
tems of aeroplanes,before they are manufactured.Circuit analysis of microchip
designs can highlight areas which may be unreliable due to the problems of IR
drop and electromigration,which can then be corrected by the circuit designers.
This project concentrates on designing and implementing a DC circuit analysis
system to be used with a speci¯c microchip design tool - the Lyric Physical De-
sign Framework - to enable users to provide reliable chip designs before the cost
of manufacture and testing is incurred.
Acknowledgements
Dr Marina De Vos
For being my project supervisor and providing advice and guidance
Dr Peter Shepherd
For proof reading and providing guidance from an electrical engineering stand-
point
Dr Alwyn Barry
For support and advice through some di±cult times
The sta® at Pulsic Limited,especially Jeremy Birch
For giving me their time and resources to help me complete this project
My family
For their ongoing love and support
David Taylor and James Steward
For always being available to chat and trying to keep my spirits high when I was
feeling low
i
Contents
1 Introduction 1
1.1 Integrated Circuits..........................1
1.2 Electronic Design Automation....................1
1.3 Lyric Physical Design Framework..................2
1.4 Scope.................................2
1.5 Aims..................................3
1.6 Document Structure.........................3
2 Literature Review 5
2.1 Chip Reliability............................5
2.2 Circuit Analysis............................7
2.3 Circuit Analysis in Lyric.......................9
2.4 Existing Circuit Analysis Systems..................10
2.4.1 ECAP II............................10
2.4.2 SPICE.............................11
2.5 Other algorithms...........................14
2.5.1 Kron's Method of Tearing..................14
2.5.2 Cholesky Factorisation....................15
2.5.3 Hierarchical Analysis Algorithm...............15
2.5.4 Random Walk Algorithm...................16
2.5.5 Hierarchical Random Walk Algorithms...........17
2.6 Conclusions..............................17
3 Requirements Analysis and Speci¯cation 18
3.1 Requirements Analysis........................18
3.1.1 Type of Circuit........................18
3.1.2 Inputs.............................19
3.1.3 Outputs............................20
3.1.4 Speed.............................20
3.1.5 Accuracy...........................21
3.1.6 Reliability...........................22
3.1.7 Platform Versatility......................22
3.1.8 Security............................22
3.1.9 User Interface.........................22
3.1.10 Heat Map...........................23
3.2 Requirements Speci¯cation......................23
ii
3.2.1 Functional Requirements...................23
3.2.2 Non-functional Requirements................24
3.3 Constraints..............................24
3.3.1 Hardware Constraints....................24
3.3.2 Software Constraints.....................24
3.3.3 Time Constraints.......................24
3.4 Summary...............................24
4 Tools and Technologies 25
4.1 Programming Languages.......................25
4.1.1 C................................25
4.1.2 C++..............................27
4.1.3 Java..............................28
4.1.4 Python.............................29
4.1.5 Conclusion...........................29
4.2 Integrated Development Environments (IDEs)...........30
4.3 Compilers...............................30
4.4 Debuggers...............................31
4.5 Revision Control Systems......................31
4.6 Backing Up Files...........................31
4.7 Testing Technologies.........................32
4.8 Summary...............................32
5 High Level Design 33
5.1 System Structure...........................33
5.2 Initial Circuit Analysis Algorithm..................34
5.3 Matrix Inversion Algorithms.....................36
5.4 Iterative Circuit Analysis Algorithm.................37
5.5 Units To Be Used...........................39
5.6 Data Structures............................40
5.7 GUI..................................41
5.8 Summary...............................42
6 Low Level Design and Implementation 43
6.1 Code Files...............................43
6.2 Types.................................44
6.3 Data Structures............................45
6.3.1 Matrices............................45
6.3.2 Circuit Elements.......................47
6.4 Filling the Data Structures......................48
6.5 Modi¯ed Nodal Analysis Algorithm.................49
6.6 LU Decomposition Algorithm....................51
6.7 Iterative Circuit Analysis Algorithm.................52
6.8 Current Calculation..........................54
6.9 GUI implementation.........................54
6.10 Intermediate Testing Results.....................56
6.11 Summary...............................56
iii
7 System Testing 57
7.1 Test Strategy.............................57
7.2 System Accuracy...........................58
7.3 Speed of the Iterative Algorithm...................59
7.4 Multi-platform Ability........................60
7.5 Reliability...............................61
7.6 Usability Testing...........................62
7.7 Summary...............................62
8 Conclusion 63
8.1 Meeting Requirements........................63
8.2 Positive Aspects............................64
8.3 Things That Could Be Improved..................64
8.4 Future Work..............................65
8.4.1 Full Chip Analysis......................65
8.4.2 Accommodation of Active Circuit Elements........65
8.4.3 AC Circuit Analysis.....................66
8.4.4 Heat Maps...........................66
8.4.5 Beyond Circuit Analysis...................67
Bibliography 68
A Example ring and mesh from an IC design in Lyric 72
B Screenshots 73
C Test Plan 76
D Circuit Designs for Testing Purposes 78
D.1 Test Design 1.............................78
D.2 Test Design 2.............................79
E Test Results 80
E.1 System v PSpice using Test Design 1................80
E.2 System v PSpice using Test Design 2................80
E.3 Multiple Platform Testing using Test Design 1...........81
E.4 Multiple Platform Testing using Test Design 2...........81
F Output Files from System Testing 82
F.1 Windows XP Pro using Test Design 1................82
F.2 Windows XP Pro using Test Design 2................83
F.3 Solaris 8 using Test Design 1.....................83
F.4 Solaris 8 using Test Design 2.....................84
F.5 SuSE Linux 9.0 using Test Design 1.................85
F.6 SuSE Linux 9.0 using Test Design 2.................86
F.7 Di® of the Output Files of MNA and Iterative Algorithms....86
G Code Listings 90
iv
G.1 input.h.................................90
G.2 input.c.................................91
G.3 mna.h.................................109
G.4 mna.c.................................109
G.5 iterate.h................................116
G.6 iterate.c................................116
G.7 current.h................................123
G.8 current.c................................123
G.9 output.h................................124
G.10 output.c................................124
G.11 utilities.h...............................126
G.12 utilities.c................................127
G.13 nodalAnalysis.h............................130
G.14 nodalAnalysis.c............................131
G.15 mna
interface.h............................134
G.16 mna
interface.c............................135
G.17 Segment of code used to interface the system with Lyric.....137
v
Chapter 1
Introduction
1.1 Integrated Circuits
An integrated circuit is a thin chip consisting of at least two connected electronic
components,such as resistors,capacitors and,most commonly,transistors.Typ-
ical integrated circuits these days are smaller than a squared centimetre in size
and can contain millions of interconnected devices.
In today's world integrated circuits (IC's) or'chips'are all around us and a®ect
almost everything we do.We ¯nd them in our computers,our washing machines,
our mobile phones - in all electrical equipment.The engineers who design such
chips want to ensure that they will be reliable,especially when they are involved
in such critical processes such as the °ying of an aeroplane or the operation of
ABS brakes in a car.A chip failure in one of these applications could cause
fatalities and expensive lawsuits plus the cost of recalling/repairing the a®ected
chips.
In order to avoid failures,it used to be the case that chips were produced and
tested to verify their reliability before going into full production,however this
is an expensive and time consuming solution.Therefore chip designers prefer a
software tool which can check the reliability of a chip design before it even gets
put into production.
1.2 Electronic Design Automation
Electronic Design Automation (EDA) is a category of tools for designing and
producing electronic systems ranging from printed circuit boards (PCBs) to in-
1
tegrated circuits [38].This basically means using computer programs to design,
lay out,verify and simulate the performance of electronic circuits on chips and
PCBs [9].Due to the continuous scaling of semiconductor technology as governed
by Moore's Law,EDA tools have become increasingly important in the design
of microchips;without EDA tools there would be nowhere near the number of
electronic devices on the market as there are today.
1.3 Lyric Physical Design Framework
The Lyric Physical Design Framework,from here on shortened to Lyric,is an
EDA tool produced by a company called Pulsic Limited and is the software
which this project seeks to improve.Lyric is de¯ned as:
"a high performance auto-interactive IC physical design solution for
analog,custom,digital,mixed-signal,embedded DRAM/SRAM and
system-on-chip (SoC) design - incorporating solutions for ECO place-
ment,routing,timing closure,signal integrity and power routing"[17]
Lyric is a complicated tool,but in simple terms it allows chip designers to lay-
out the electrical components required for a particular chip and connect all the
components together with interconnects in a process called'routing'in order to
produce a complete chip design.
To ensure Lyric produces reliable chip designs,it needs to incorporate some
software tools that will analyse the design,looking for the issues that can cause
chip failure.The results of the analysis should then be reported to the design
engineers who can make the necessary corrections.This encapsulates the focus
of this project.
1.4 Scope
The domain of this project is a large one as the reliability of a microchip can be
in°uenced by many factors including the correctness of the logic of the chip and
errors in the chip introduced during the manufacturing process.However,in order
to limit the scope to a manageable level,this project will only be concerned with
the important errors which are caused by an excess of current (electromigration)
or a reduction of voltage (IR drop) across the circuitry.These errors will be
described in more detail in Chapter 2,as will the process of circuit analysis
which is a technique to ¯nd these errors and the basis of the ¯nal system.
2
This still leaves a large domain however,for instance there are two types of
circuit to think about,alternating current (AC) and direct current (DC).The
circuit analysis for DC is done in the time domain while AC analysis is done in
the frequency domain.For the purposes of limiting the broadness of the project
and to ensure the goal are achievable,the project will only be concerned with
DC circuits.
1.5 Aims
The aim of this project is to design a library of functions that will analyse the
circuitry of an integrated circuit design to discover any issues that may cause
the chip to become unreliable.This library should be able to be integrated into
the Lyric Physical Design Framework software as an analysis tool and be able to
show graphically the results of its analysis in the Lyric graphical interface.
Another major aim,garnered from preliminary discussions with Pulsic Limited,
is to ensure the tools provides the ability to quickly re-analyse a chip design after
it has initially been analysed and the design edited.This is important as design
engineers quickly want to see the e®ects of any alterations they make to a design
as Lyric is a real-time interactive tool.
It is aimed that the analysis tool will produce accurate results and produce said
results in a timely fashion comparable to any similar tools on the market.It is
also intended that the tool be easy to use for a professional IC design engineer
and that the results produced be displayed in a clear and obvious manner.
1.6 Document Structure
The next chapter is the literature review which outlines the issues that can cause
unreliability in microchip designs and then goes on to investigate the methods of
analysing electrical circuits,any current technology that provides similar func-
tionality and existing algorithms which may be useful in the designing of the new
tool.
Following on from the literature review is the requirements chapter which studies
the task at hand in detail and outlines the exact aims of the project.Further
on in the document comes the high-level design chapter and then the chapter
detailing the detailed design and implementation of the system.
Chapter 6 looks at the testing of the produced system to assertain how well the
¯nished product meets up with the requirements.The document closes with an
3
evaluation of the project as a whole and provides recommendations for future
work.
4
Chapter 2
Literature Review
2.1 Chip Reliability
As outlined in the scope of the project in Section 1.4 on page 2,this project will
be concerned with the important reliability issues caused by IR (voltage) drop
and electromigration.Here these issues will be described in detail,looking ¯rstly
at IR drop.
The voltage supplied to a chip needs to be carried around the chip e®ectively in
order to provide the power for the electronic components to function.IR drop
is due to the resistance of the metal lines of the power network [24] resulting in
lower voltages reaching parts of the chip than intended.If there is an excessive
drop in the value of the voltage at some node on the chip compared to that which
was supplied then there is a problem in the design of the chip as the power is
being wasted.With the size of chips getting smaller and the amount of voltage
being supplied being lowered proportionally,even a small drop in voltage across
a chip can be signi¯cant [25].Drop in voltage can lead to delays in the circuit
causing slower performance and also cause logic failures and the inability of some
components to function properly since they aren't receiving enough power.
Traditional methods for detecting this voltage drop cannot handle today's large
designs [43] and so are useless for ensuring designs are reliable.An alternative
solution is to use electron beam and focused-ion beam studies [43] but the costs
involved in these processes can be debilitating in today's market environment.
Therefore an e®ective software tool which can detect voltage drop during the
design phase is invaluable.
Second,is the problem of electromigration in the interconnects (microscopic alu-
miniumwires) that join up the electrical components on a chip.Electromigration
research was pioneered by James R Black.He de¯ned the phenomenon as the
5
Figure 2.1:a) A void forming.b) hillocking - short circuits forming [22]
transport of mass in metals when the metals are stressed at high current densities
[1],where current density is the ratio of current to the cross-sectional area of a
conductor.In integrated circuits where the interconnects are now usually less
than 130 nanometres in width and carry a relatively high current,the current
density is high.Inside the interconnects,atoms of the metal are being bombarded
as an'electron wind'(current - a °ow of electons) [1] tries to rush through it.
This results in collisions in which electrons pass on their momentumto the atoms
and pushes them out of position.If this happens enough over a period of time,a
su±cient amount of atoms will be moved far enough to cause the wire to physi-
cally snap,creating what is known as a void [37] and breaking the circuit.(See
Figure 2.1.a )
Another outcome of electromigration is that atoms of a connecting wire could be
forced into piling up and drifting outward from the metal and possibly join up
with a nearby interconnect.This is known as hillock failure or whisker failure
(short circuit) [37] and is likely to cause the chip to malfunction.(See Figure 2.1.
b )
In order for the two outlined possible causes for chip failure to be avoided,the
chip designer needs to know the voltage at every node in the circuit and also
the current that °ows through each of the interconnecting wires.Knowledge of
the voltage across the circuit will allow the designer to see where power is being
leaked and so amend the chip design accordingly.Knowledge of the currents at
all points across the circuit will show where the current is too high and enable
the designer to fatten the wires to reduce the current density and so reduce the
risk of electromigration.
The solution to avoiding these problems is to perform circuit analysis on the
electrical circuits in a chip design in order to spot where such problems might
occur and alter the design accordingly.
6
2.2 Circuit Analysis
Circuit analysis is the study of electrical circuits in order to predict the electrical
behaviour of physical circuits [7].Before explaining circuit analysis,here are a
few useful de¯nitions:a node is a point in a circuit where three or more wires
meet [20],a branch is any path in the circuit that has a node at each end and
contains at least one voltage source or resistor but contains no other nodes [20],
a loop is where nodes and branches form a ring in a circuit.
There are many di®erent forms of circuit analysis,but as detailed in the project
proposal,this project has limited its scope to deal with dc circuits only,so only
dc circuit analysis will be described in this report.The circuits are assumed to
comprise of one or more voltage sources,zero or more current sources and one
or more resistors.The resistors are assumed to be the branches (wires) of the
circuits themselves.
In order to conduct circuit analysis on an electrical circuit,there are some basic
electrical laws which need to be used.The most fundamental of these is Ohm's
law.
Ohm's law states that the voltage across a resistance is directly proportional to
the current °owing through it.The resistance,measured in ohms,is the con-
stant of proportionality between the voltage and current [13].The mathematical
relationship of Ohm's law is illustrated by Equation 2.1.
V (t) = R£I(t);whereR ¸ 0 (2.1)
which is usually simpli¯ed to Equation 2.2.
V = I £R (2.2)
Ohm's law provides the tools for analyzing simple circuits,but for more compli-
cated circuits,such as those you would ¯nd in a microchip design,there are two
further fundamentals called Kirchho®'s laws.The ¯rst law is Kirchho®'s current
law (KCL),which states that the algebraic sum of the currents entering any node
is zero [13].For example,applying Kirchho®'s law to the node in Figure 2.2 yields
Equation 2.3:
i
1
+i
4
¡i
2
¡i
3
= 0 (2.3)
The second law is Kirchho®'s voltage law (KVL) which states that the algebraic
7
Figure 2.2:Four branches connected to a node
sum of the voltages around any loop is zero [13].This means that the sum of the
voltage drops through the resistors equals the sum of the voltage rises through
the voltage sources [21].
Kirchho®'s laws in conjunction with Ohm's law can be applied to a complex
electrical circuit with multiple loops and nodes in order to determine the voltages
and currents at all nodes and branches of the circuit.There are two predominant
methods of doing this - nodal analysis and mesh (or loop) analysis.
In nodal analysis,the variables in the circuit are selected to be the node voltages.
One node is selected as the reference node and all other node voltages are de¯ned
with respect to that node [13].This node is called the ground node and is assigned
a voltage of zero [19].
The process then continues by iterating through the nodes in the circuit and
applying Kirchho®'s current law to each of them.The result is a system of m
linear equations in the munknown voltages (where mis one less than the number
of nodes).The equations are of the form:
where G
11
,G
12
,...,G
mm
and I
1
,I
2
,...,I
m
are constants [19].The equations
can be considered a matrix and then solved with Gaussian elimination or some
other method giving the voltages for all the nodes.With the node voltages known,
the currents along each branch can easily be determined using Ohm's law.
In mesh analysis,the unknown parameters are the current values and Kirchho®'s
voltage law is used to determine them [13].The method starts by identifying all
the loops in the circuit (see Figure 2.3 for a basic diagram).
The loops are then taken one at a time and KVL equations formed for each,
8
producing a system of equations of the form:
where R
11
,R
12
,...,R
mm
and V
1
,V
2
,...,V
m
are constants [18].As with
nodal analysis,these equations can then be solved as a matrix and the voltage
values determined using Ohm's law.
The choice of which method to use is often dependent upon the circuit to be
analysed.If the circuit had fewer nodes than loops,then nodal analysis would
be preferable as it produces fewer equations to solve.For the same reason,mesh
analysis would be preferable if there are fewer loops than nodes.One other
consideration to make is that mesh analysis can only be applied to circuits which
are planar,i.e circuits which can be drawn in two dimensions [34].What this
means is that wires in a circuit must not"jump over"one another in order to
apply mesh analysis.For this reason nodal analysis is most commonly used in
circuit analysis software,since it is always guaranteed to work.
2.3 Circuit Analysis in Lyric
Currently,Lyric does not include a mechanism with which to analyse circuits
in microchip designs which contain loops.Since the circuits in chip designs are
usually very complicated with lots of loops,this is quite a problem.At the
moment,Lyric only has a very basic tool with which to analyse only parts of the
circuits where no loops occur.
According to Jeremy Birch,Chief Technology O±cer at Pulsic Limited,Lyric
currently employs a basic dendritic method for circuit analysis.It is called den-
dritic as it follows a simple tree/branching strategy and takes as input nodes at
Figure 2.3:Loops in a basic circuit [18]
9
Figure 2.4:Simple dendritic analysis strategy for determining branch currents
which the current is known and then it is easy to work out the currents that °ow
through the branches.For example,in Figure 2.4 it is known that 5mA are enter-
ing the branch at X and that devices at Y and Z draw 3mA and 2mA respectively,
then the value of current which °ow through the branches are obvious.
As this method only works for the most basic of examples,Lyric is in need of a
proper circuit analysis tool,which is the purpose of this project.A key feature
for any such tool is that it needs to be interactive - that is,it should quickly show
how small changes to the circuit e®ect how the circuit behaves.This is important
so chip designers can create reliable designs.
Computer-based circuit analysis tools have been developed since the 1960's.In
order to understand how circuit analysis tools work and the methods and tech-
nologies available,a few circuit analysis tools are explored here.
2.4 Existing Circuit Analysis Systems
2.4.1 ECAP II
ECAP II,Electronic Circuit Analysis Program,is a circuit analysis program that
was ¯rst devised in 1971.It has no graphical user interface and instead features an
input language for users to enter the details of a circuit design,with FORTRAN
sub-routines [2].
The method by which the program carries out circuit analysis is not by us-
ing nodal analysis or mesh analysis but rather by a process called the mixed
method.This method utilises the idea that Ohm's law can be expressed in a
matrix form involving the admittance matrix for certain branches of the circuit
and the impedance matrix for the rest of the branches [3].Impedance is an ac
circuit term which is analogous to resistance in a dc circuit and admittance is
analogous to conductance in dc circuits which is the reciprocal of resistance [13].
10
The admittance branches are then analysed with a method called the cutset
method and the impedance branches by the mesh method.The two sets of
linear equations are then combined into one matrix and solved.The method
of solution,rather than using the straight-forward Gaussian elimination,uses
implicit integration based on the variable order approach of Gear [10] as modi¯ed
by Brayton et al [4].This method approximates a solution and then iterates -
converging towards the actual solution.This only takes one or two iterations for
a dc problem [2] and is faster than the Gaussian approach.
An advantage of ECAP is that it contains a modify feature [3] which is based
on Kron's tensorial method as explained later in this paper.The modify feature
allows the matrix of solutions for a circuit to be updated in relation to changes in
some branch resistances with only trivial computation compared to re-calculating
the entire solution from scratch.This is a feature that is needed for a circuit
analysis tool for Lyric since it needs to be an interactive design tool and so this
method could be adopted.
A possible disadvantage of this program is the complex method by which the
systems of equations are constructed and solved.It would be much simpler to
construct the matrices using nodal analysis and solve them using Gaussian elim-
ination.It's true that ECAP's method is faster but is this increase in complexity
worth the time gain?Also,with the speed of modern computers the increase,in
speed may be minimal.
2.4.2 SPICE
SPICE,Simulation Program with Integrated Circuit Emphasis,is a circuit simu-
lation program developed in the 1970's at the University of California,Berkeley.
It is a derivative of another circuit analysis program,CANCER,that was also
developed at Berkeley [23].
SPICE is the most widely used circuit analysis program in the market place and
has many derivatives of its own such as SPICE2,SPICE3 and PSPICE which is
the pc version of the software owned by Cadence Design Systems.
Like ECAP,SPICE was designed to have no user interface as standard and was
used via an input language.Programs were written in the input language,saved
and then compiled into an executable [8].SPICE took as input a netlist ¯le
which describes the circuit,which was then fed to the executable to perform the
analysis.More recent incarnations of SPICE have a full graphical user interface
and circuit designs can be entered into the system by placing circuit components
onto a canvas to form a schematic for the circuit.
The way SPICE works is shown in the block diagram in Figure 2.5.At the heart
11
Figure 2.5:The SPICE °ow diagram [8]
12
of SPICE is nodal analysis as seen in steps 3 and 4.For linear dc circuits,these
are the only stages that are necessary.The inner loop,steps 2 - 6,iterates to
¯nd the solution for non-linear dc circuits,while the outer loop,7 - 9,extends
the process for ac analysis.
As this project's scope is only linear dc circuits,steps 3 and 4 are the impor-
tant ones.As mentioned above,SPICE's analysis mechanism of choice is nodal
analysis.SPICE creates the matrices for the systems of equations in the form
with the G matrix being a matrix of the conductances (1/resistance),the I vector
being a vector of branch currents and the V vector being the vector of the node
voltages.The V vector is where the solutions will appear at the end of the
analysis.
SPICE uses a shortcut called matrix construction by inspection to create the ma-
trices rather than creating a system of nodal equations using Kirchho®'s current
law.This inspection method allows SPICE to create the system matrices as each
element is read in from the input ¯le at the start of execution.
"Matrix construction by inspection builds the system matrices and
identi¯es the location of each element in the matrices as soon as the
nodes connected to the element are de¯ned [15]"
For solving the matrices,SPICE uses a form of Gaussian elimination called LU
(or triangular) decomposition.This method of solution is faster than normal
Gaussian elimination.Standard Gaussian has a complexity of O(n
3
/3) whereas
LU decomposition has a complexity of O(n
3
/3 -n/3) where n is the number of
rows in the matrices [14].
Later incarnations of SPICE replaced nodal analysis with an improved system
called Modi¯ed Nodal Analysis (MNA).MNA was designed to rectify the short-
comings of standard nodal analysis.These drawbacks include the fact that nodal
analysis treats voltage sources ine±ciently and is incapable of including current-
dependent elements.Another issue is the fact that the nodal method does not
produce the currents through the independent voltage sources as part of the out-
put [12].These have to be calculated later using Ohm's law at each of the circuit
nodes and so adds to the computation required.
13
MNA resolved all of the problems of the nodal approach while preserving its ad-
vantages [12].MNA produces the currents through the voltage sources as well
as the node voltages as part of its output while also reducing the complexity of
computation.While the order of the matrix equation is increased (the algorithms
only overhead),the equations are generally highly sparse and can take advantage
of e±cient routines to solve sparse matrices.Also MNA indicates the most ef-
¯cient ordering for the LU decomposition process,speeding up the process of
¯nding the solutions [32].
Results from [12] shows MNA outperforms another common approach to circuit
analysis - the tableau method.The advantage of SPICE's mechanism is that it
uses algorithms which are reliable and easy to use.The algorithms have been
around for decades and are proven to be useful and there is also a lot of supporting
literature which would be useful in implementing a SPICE-like system in Lyric.
The disadvantages include the fact that an O(n3) algorithm on matrices with
thousands of lines may take a fair amount of time to compute.Also,unlike ECAP,
SPICE does not allow quick re-analysis when there have been small changes to
the circuitry,the analysis has to be restarted from the beginning which takes
time.Quick re-analysis is fundamental to any analysis tool to work well with
Lyric.
It may be useful to use some of SPICE's methodology to create a circuit analysis
tool for Lyric but it is not the whole answer.
2.5 Other algorithms
2.5.1 Kron's Method of Tearing
This method invented by [16] is supposed to make it possible to solve very large
physical problems in easy stages.The method involves tearing a large system of
equations up into a large number of small subdivisions,solving each subdivision
separately and then interconnecting the partial solutions by a set of transforma-
tions to obtain the solution of the original system [16].It is important that the
tearing is carried out such that each component part produced is independent of
every other part so that each can be solved independently before amalgamating
them to form the whole solution.
The main advantage of this tearing (or tensorial) technique is that the amount of
computation required to solve the system of equations is reduced by a signi¯cant
amount.For example,taking the system I = YE,where I,Y and E are matrices,
and dividing it into n parts,calculating the inverse E = ZI is reduced by a factor
14
of 2/n
2
[16].If n = 100,the calculations are reduced to a fraction of about
1/5000.
Other advantages include the fact that if several of the subdivisions are identical,
the reduction in computation increases dramatically.Also,if a system is already
solved and is then altered in any manner,the solution of the altered system
need not be started from scratch.Only the solutions of the altered portions are
changed,the rest of the solutions remaining unchanged [16].
In [3],a method is presented by which Kron's method may be used to study the
e®ects of varying a single resistance in a circuit.This method permits the nodal
solution matrix to be updated in accordance with changes in one or more branch
resistances.This feature would be very good in a circuit analysis system built
for Lyric as it would allow quick recalculation of the circuit properties as the
engineers make changes to their designs.
It is also stated in [3],however,that using this technique to compute a sequence
of solutions as a given resistance is varied through several di®erent values may
introduce round-o® errors that accumulate fromone step to the next.This would
invalidate the results acquired from the analysis.
This implementation of Kron's tensorial method may provide an ideal solution to
quickly calculating the a®ects of small changes made to a circuit by chip designers.
It remains to be seen however if the round-o® errors can be avoided.
2.5.2 Cholesky Factorisation
Cholesky factorisation is a technique for solving sparse matrices.The method
states that if matrix A is positive de¯nite,then there exists a unique lower tri-
angular matrix G with positive diagonal entries such that A = GG
T
[11].
This method does not require pivoting as in the LU decomposition method for
solving matrices and hence its performance is twice as fast [31].The technique is
also fairly easy to implement,however it can only be used on matrices that are
symmetric and positive de¯nite.
2.5.3 Hierarchical Analysis Algorithm
This algorithm is based on the well-known strategy of divide and conquer and
compromises the following:(1)The power grid is divided up into local and global
grids,(2) macromodels are generated for the local grids,(3) the admittance
(1/resistance) matrix is sparsi¯ed,(4) the global grid is modelled using the local
15
grids and ¯nally,(5) simulating the local grids where desired [44].
As with other divide and conquer algorithms,this method bene¯ts from solving
many simple problems rather than one large problem.Further improvements in
computation are gained by the sparsi¯cation of the matrices so that they may
be solved quickly using e±cient sparse matrix solving algorithms.[44] shows
that the hierarchical analysis method is signi¯cantly faster than conventional
non-hierarchical methods.
The disadvantage of this method is that implementation may be di±cult due to
the complex nature of the algorithm.Also,the algorithm relies on a hierarchical
structure in the design of the circuit which often may not be the case.
2.5.4 Random Walk Algorithm
This is a new method for analysing circuits,more speci¯cally power grids,based
on the relationship between random walks and electrical networks [27].The
method is inspired by the work of Doyle and Snell [28] which interprets the
relationship between resistive networks and probabilities.
[27] shows that using Kirchho®'s laws,the voltage at any node in a circuit is a
linear function of the voltages at its neighbours.This is analogous to a random
walk problem so that for any power grid problem,a mathematically equivalent
random walk problem can be constructed i.e characterised by the same set of
equations [27].
By ¯nding an approximate solution to the randomwalk problem,an approximate
solution to the circuit problem is also found.The solution is found by conducting
a certain number of experiments on the random walk and averaging out the
voltages returned to give the approximated solution.
"If this amount is averaged over a su±ciently large number of walks
by playing the'game'a su±ciently large number of times,by the law
of large numbers [42],an acceptably accurate solution can be obtained
[27]"
There is an accuracy-runtime trade o® with this algorithm that means the more
accurate the solution you want,the longer it will take to compute.The number
of walks carried out,and so the accuracy of the solution,is governed by a user
speci¯ed variable.This algorithm should be capable of an accuracy of above 99%
however.
The advantages of this algorithm include the speed at which a solution is found.
16
The algorithmhas linear complexity [27] and so when testing against other meth-
ods is found to be much faster.Another plus point is that the algorithm localises
computation;it can calculate a single node voltage without having to solve the
whole circuit which is a desirable feature in an interactive design and analysis
tool.
A disadvantage of this method is that the algorithm seems to be fairly complex
and so implementation might be di±cult.Also the algorithm only produces node
voltages so the branch currents would have to be worked out afterwards using
Ohm's law,which adds to the computation time.The other disadvantage is that
the results are close approximations rather than actual voltages and so there is
a small margin of error introduced.
2.5.5 Hierarchical Random Walk Algorithms
The hierarchical random walk algorithms combine the divide and conquer idea
of the hierarchical analysis algorithm with the random walk method [28].This
method proves to work roughly three to four times faster than the standard
random walk method while also being more robust.
While being very fast this algorithm does have the drawback of being more com-
plicated to implement than the two algorithms it is based on.It also has the
same problem that the branch currents are not calculated as part of the analysis
and so will have to be done separately at the expense of extra computation.
2.6 Conclusions
The method as employed by SPICE is the most easy to understand and implement
while being the best supported and probably the most reliable.However,as it is
not a suitable solution as an analysis tool in an interactive design environment,
it cannot be the whole solution for this project.
Using the SPICE method as ¯rst pass analysis would provide a good base for
further analysis.Then Kron's tensorial method or a random walk algorithm or
some other similar algorithm could be useful for providing quick re-analysis when
small changes are made to the circuit.
17
Chapter 3
Requirements Analysis and
Speci¯cation
3.1 Requirements Analysis
This chapter aims to determine all the goals of the system to be created and
set them out as a speci¯cation for the design stage.The best way to elicit the
requirements for a new software system is to enlist the help of the experts in
the area.Since this circuit analysis system is to be incorporated into the Lyric
Physical Design Framework,it makes sense to talk to the engineers at Pulsic
Limited who make the software and therefore know what they want from a new
tool.
Jeremy Birch is a co-founder of Pulsic and currently works as the Chief Tech-
nology O±cer.He is widely recognised within the company to be the foremost
expert on the software and has been the principal contact within the company for
this project.The documentation of the requirements process here comes largely
fromdiscussions with Jeremy Birch,with some input froma few other employees.
3.1.1 Type of Circuit
As mentioned in Section 1.4 on page 2,this project is focussing just on the DC
aspect of IC circuits and ignoring the AC aspect in order to help set a manageable
task.This still leaves a very large scope for the project,so it needs further re¯ning.
Electrical circuits can contain active devices and passive devices.Active devices
are either sources of energy,driving electrical current around the circuit and hence
delivering continuous energy to it [33],or else they have the ability to control the
18
°ow of electricity.The most common active device is the transistor.Passive
devices are those which consume energy rather than produce it,like resistors,
inductors and capacitors [33].
Since active devices control the °ow of electricity,they are much harder to model
than the passive devices which only consume it.To make this project of achievable
size,the circuit analysis tool that is to be produced will operate on circuits
consisting of passive elements only.
Further more,since modern integrated circuits can contain more than ten mil-
lion components with interconnects in between,this project will concentrate on
analysing speci¯c sections of the chip.These sections are known as'ring and
mesh'.They consist of a'ring',which is a relatively large power track,and lo-
cated within it is a'mesh',which is a grid of thin wires,the edges of which are
connected to the ring.The mesh carries power from the ring to the components
situated beneath.
Ring and mesh circuits are suitable for this project since they can be easily
modelled.The wires are modelled as resistors,the connections between the mesh
and the power ring can be modelled as independent voltage sources and the drain
of current from the components below the mesh can be modelled as independent
current sources.
3.1.2 Inputs
To perform circuit analysis,the system needs to know the architecture of the
circuit - that is,how the nodes of the circuit are connected together by the
branches (wires),and also where any independent voltage sources and current
sources are connected.
In a stand-alone library with no graphical user interface,this information can
be fed into the system manually.The user could simply enumerate the nodes,
branches,voltage sources and current sources and type the details into a command
line interface.This is similar to how OrCAD SPICE works,albeit without the
graphical user interface.In SPICE,users drop onto a schematic the components
to create the circuit they desire before running any analysis.Entering the circuit
information via a command line interface might be useful for implementing small
circuits for test purposes,but it would be very tedious.
If the circuit analysis systemis to be integrated into Lyric then it is likely that the
circuit information will have to be inputted in the same way that the information
is entered into Lyric which is via a netlist ¯le.This is a special ¯le format that
contains all the information for the circuitry in a chip design which is read and
stored appropriately.
19
3.1.3 Outputs
As stated in Section 2.1 on page 5,the system created by this project is intended
to help chip designers reduce the possibility of electromigration and IR drop
occurring.
Since electromigration is caused by an excess in the °ow of current through a
wire,the system needs to be able to calculate and output the value of the current
°owing through every branch between the nodes of the circuit.In order to make
it easy for a chip designer using the system to see where there is an over°ow of
current,it would be useful for the system to highlight where there is a current
that exceeds acceptable limits.It would be simple then for the user to see the
places where the wire should be fattened to reduce the stress.
Apossible addition to this functionality which would be a bonus to the user would
be for the system to highlight areas where the current °ow is signi¯cantly under
the maximum capacity of a wire.With this information,the user could make
the wire thinner and thus possibly make the overall design smaller,resulting in
a smaller physical chip size.
To recap on Section 2.1,IR drop is the phenomenon whereby voltage drops ex-
cessively across a circuit,possibly resulting in not enough power reaching certain
areas of the chip.Therefore,the system needs to output the voltage at every
node across the circuit.To make it easier for the user to see where problems
may occur,it would be of bene¯t if the system could highlight nodes where the
voltage is below some predetermined threshold level.
An enhancement to this functionality would be for the system to indicate where
a node is receiving a large excess in voltage.This might allow the chip designer
to reduce the power input into the chip.
3.1.4 Speed
The initial attempt at circuit analysis,when the system¯rst calculates the values
for voltage and current across the speci¯ed circuit,does not have to be partic-
ularly fast.When Lyric's router tool runs and joins up all the components in a
chip design with wire tracks,it can take an hour or more,depending on the size of
the design.So,relative to this,conducting circuit analysis will be comparatively
fast anyway so there is no real need to strive for speed for the ¯rst iteration.The
focus at this point is simply accuracy and reliability.
The results of the ¯rst circuit analysis will often show areas of the circuit where
currents and voltage fall outside the speci¯ed parameters.In a mesh with ten
20
thousand nodes,and thus more than ten thousand branches,there maybe hun-
dreds of such errors that need correcting.
Fixing the errors highlighted by the circuit analysis is done by either fattening
the wires or thinning them.For instance,if there is too much current °owing
through a wire then the wire needs fattening to safely accommodate the current
°owing through it.
However,a single fattening of the wire will not normally ¯x the problem.By
fattening the wire,the resistance of the wire is reduced causing more current to
°ow through it,which could mean the fattening has little e®ect.The solution
then is to fatten or thin the wire iteratively and re-analyse to see the e®ects
of the action and ¯nd the optimum wire width.The wire needs to be fattened
enough to safely cope with the current °owing through it but not so much that
the amount of current increases enough to negate the fattening.The thinning
process is similar to this.
The chip designer may chose to do the fattening/thinning of the wires in the
design by hand.More likely though,since there may be hundreds of wires in
a design that require the iterative fattening/thinning process,there will be an
automated ¯xing tool that does the job.
Since there will be a lot of iterations of making changes to the design and then
re-analysing of the circuit,it is imperative at this point that the circuit analysis
is done very quickly.The automated ¯xing tool has to wait on the results of the
analysis before making its next change and so it cannot have to wait long or the
whole process would take an impractical amount of time.
The focus therefore during this area of the systems usage is speed,but still with
as high accuracy as possible,although some slight trade-o® may be acceptable.
3.1.5 Accuracy
Since the circuit analysis is being performed on integrated circuits,the width
of the wires is tiny and so the values of the currents and voltages are also very
small.The current °owing through any wire in the circuit may range from one
amp down to a few nano amps while it also must be possible to detect a drop
in voltage of one ten-thousandth of a volt between adjacent nodes.The system
will have to be able to manipulate these small numbers accurately and ensure
the possibility of round-o® errors is avoided
21
3.1.6 Reliability
The system to be produced by this project is just simply for academic purposes;
Pulsic wish to integrate it into their Lyric software.This means that the system
will have to live up to the reliability expectations of the business world.Pulsic's
customers will expect a high level of reliability since they will have spend a lot
of money on the Lyric tool.
This implies that the testing performed in this project should be as rigorous as
possible to catch any bugs and errors in the system.
3.1.7 Platform Versatility
Pulsic provides versions of its software to run on a wide variety of platforms to
suit all customer needs.The range of platforms includes Linux,Solaris,HP-UX
and Windows.Further more,Pulsic support di®erent varieties of each of those
platforms.For Linux for instance,Lyric is available to work with SuSE 9.0,SuSE
9.1,SuSE 10,RedHat 7.2 and RedHat 8.2.
If this system is to be integrated into Lyric,then it too must be able to run on
all of these di®erent platforms.This means taking into account things such as
variations in the compiler available etc.Thorough testing should be employed to
ensure the system runs correctly on all platforms.
3.1.8 Security
It is important that the software Pulsic release is protected so that other com-
panies cannot dissect the code and learn all of the company's design secrets.
Therefore,if this system is to be integrated into the Lyric,it is important that it
does not violate any security protocols put in place by Pulsic so that the software
remains secure.
3.1.9 User Interface
If the systemis to be integrated into the Lyric software,it will have to use Lyric's
graphical user interface rather than having one of its own.This will probably
involve adding a menu and some toolbar buttons to the GUI for controlling the
circuit analysis tool's functionality.The circuitry for circuit analysis to be per-
formed on will probably be chosen by selecting sections of a chip design displayed
22
in the Lyric GUI.The output of the system will also have to be displayed in the
Lyric graphical user interface,in some appropriate fashion.
The changes made to the Lyric GUI will have to be made in such a way to keep
it consistent,so as not to confuse users.The additions for the new system should
be clear and compact to ensure ease of use.
3.1.10 Heat Map
As the number of components on a chip continues to increase along with the den-
sity in which they are packed,overheating is becoming a serious factor a®ecting
the performance of integrated circuits.
As part of its functionality,the system could calculate the heat generated by
a chip based on the results of circuit analysis.This would require calculations
involving thermodynamics and heat conductivity of the various materials from
which an integrated circuit is constructed.The results would then be displayed
in a'heat map'superimposed over the image of the circuit.
However,the complexity of the calculations required for this task to be performed
is huge and therefore this requirement is considered to be out of the scope of this
project.
3.2 Requirements Speci¯cation
3.2.1 Functional Requirements
²
Perform circuit analysis on DC ring and mesh circuits containing passive
components
²
Take as input the circuit architecture,including the nodes,branches,volt-
age sources and current sources
²
Perform an accurate ¯rst iteration of circuit analysis
²
Execute fast further iterations of circuit analysis
²
Output the voltages at each of the nodes
²
Output the currents through each of the branches
²
Display the results graphically on the Lyric GUI
23
3.2.2 Non-functional Requirements
²
Output accurate results
²
Have good reliability
²
Uphold Pulsic's security protocols
²
Be easy to use
²
Be able to run on all the platforms that Pulsic support
3.3 Constraints
3.3.1 Hardware Constraints
The author has been given use of an AMD Athlon PC with 512Mb of RAM by
Pulsic.The author also has their own Pentium 4 3.1GHz notebook with 512Mb
of RAM and the author also has use of the PCs available on campus at the
University of Bath.
3.3.2 Software Constraints
The author has use of the Windows XP operating system on their notebook,
the SuSE 9.0 Linux operating system on their Pulsic machine and the UNIX
operating system on the University of Bath computers.Servers at both Pulsic
and the university also provide compilers for most mainstream languages that
the author may wish to use.
3.3.3 Time Constraints
The ¯nal deadline for the project is Monday 8th May.
3.4 Summary
This ends the Requirements Analysis and Speci¯cation chapter.The next chapter
investigates the various tools available for implementing the system and chooses
the best ones for the task.
24
Chapter 4
Tools and Technologies
4.1 Programming Languages
There are various languages available that could be used to implement the current
analysis system.Here is a discussion of some of those languages,leading to a
decision on which should be used.
4.1.1 C
Advantages
C has a simple,small core language with extra functionality being provided by
a large range of libraries which include maths and ¯le handling functions [36].
This allows programmers new to the language to quickly become pro¯cient in the
basics,while the more complicated routines are readily available when needed.
The language the supports procedural programming paradigm [36] which lends
itself to modular design.While C does not force modular design upon the pro-
grammer,good programmers can use the language to create a cohesive modular
structure.The modules are usually separate ¯les which can be compiled sepa-
rately.This would be useful for this system since it has been designed as a set of
modules with speci¯c functionality.
C is extremely °exible and allows the programmer to have low level control over
the computer.As such,the programmer has full control of memory allocation
and pointers to create program solutions which are tailored to the problem.The
low level nature of the language also results in the language being fast in its
25
execution.This is a feature which would be welcomed for this system as there
are parts of it which need to run as fast as possible.
Through the use of special keywords,C allows the programmer to create new data
types to suit the needs of the programsolution.This is particularly advantageous
for this project as some unique data structures will be needed to store the data
for the objects in the system such as nodes,branches etc.
The Lyric Physical Design Framework is programmed largely in C.If this system
is also programmed in C,it would be easy to integrate it into Lyric which is an
obvious advantage.Furthermore,the author has extensive experience of using C
which would mean implementing the system in this language would be easier.
Disadvantages
One disadvantage of C is that memory allocation has to be done explicitly by
the programmer.This can lead to errors when memory is not allocated properly.
Allocated memory is also not initialised automatically [36] and so this too has to
be done by the programmer to avoid getting bizarre values when trying to access
data structures.If C is used,these disadvantages mean that the author will have
to be particularly vigilant when allocating the memory for the data structures
which will doubtlessly be needed.
C is not a strongly typed language.This means that the compiler will often not
complain when a variable is declared as one type and then used to store a value
of a di®erent type.Therefore,if using this language,the author will have to
be careful when writing the program to ensure variables are used as declared to
avoid getting any unusual behaviour.
C has some idiosyncracies in its syntax which can catch the programmer out and
cause unexpected behaviour in the program.A well known example of this is
accidentally using if (x = 0) instead of if (x == 0).The former will set x to be
zero rather than testing if x is equal to zero.This will usually cause the program
to run di®erently than expected,but it will not be picked up by the compiler.
Other such syntax issues include the"dangling else"and loops that only iterate
over the ¯rst line if a block if the curly braces are forgotten.These issues mean
the programmer has to be very careful when writing a program to ensure the
code will perform exactly as they are expecting.
The disadvantages above can mean debugging C code can be problematic which
means the author will need good debugging skills.
C also has no automatic garbage collection which means the programmer has to
explicitly free all allocated memory.
26
4.1.2 C++
Advantages
Since C++ is derived from C,it inherits many of the advantages of C such as
support for procedural programming,the ability to create bespoke data structures
and its general °exibility.
C++ supports object-oriented programming [35].This allows programs to be
organised as a collection of classes and objects,which,enthusiasts claim,makes
programming large systems easier.If the circuit analysis system had been de-
signed with object oriented programming in mind,this would be an advantageous
feature.However,the author is more comfortable with the procedural paradigm
which is why the system has been designed in a modular fashion instead.
Despite the extra functionality included into the language to support object ori-
ented programming,C++ remains a fast language which is an advantage for the
iterative circuit analysis tool which relies on speed.
The majority of Pulsic's source code is written in C,while a good portion is
written in C++.This means that if the system was written in C++,it would be
easily integrated into the Pulsic software.
The author has little experience with C++,but the fact that it is a derivation of
C and that the author has a lot of experience using C,learning C++ should not
present too much of a challenge.
Disadvantages
Since C++ is derived from C,it inherits many of the disadvantages of that
language too,such as the idiosyncracies of the syntax and the lack of garbage
collection.
While the core of the C++ language is small like C,there are many libraries
that deal with the additions for objected oriented programming which can make
it seem like a large and complicated language [35].If programmers try to use
all of the functionality at their disposal,then it can be a di±cult language to
master.The additional functionality can also make C++ a slower language than
C.Furthermore,while the standard library has been developed by many users
over the years to become useful and reliable,many other libraries have not.This
can cause problems when trying to develop programs that utilise these libraries.
The C++ language contains constructors and destructors that are supposed to
27
help with memory allocation.In reality however,they can often hide a lack of
deallocation of memory which then leads to memory leakage.
The error messages produced by a C++ compiler are notorious for being ex-
tremely complicated and di±cult to understand.This will mean choosing C++
to implement this system will require a lot of e®ort when it comes to debugging
the program.
4.1.3 Java
Advantages
Java's syntax is heavily derived from that of C/C++ [40] and so it should look
familiar and be easy to learn for those programmers who have experience with
either of those languages.The author has had some experience with Java,which
is an advantage should Java be chosen as the language to be used to implement
the circuit analysis system.
Java hides the details of the machine from the programmer.For instance,the
programmer does not have to worry about pointers or explicit memory alloca-
tion.Java has automatic garbage collection [40],so implementing the system in
this language would mean the author would not have to worry about explicitly
freeing the memory used by any data structures.These features should make
implementing the system easier.
Since it is compiled to bytecode which can run on any machine with a Java
Virtual Machine (JVM),Java is an extremely portable language [6].This is an
advantage for this project since the circuit analysis system has to be able to run
on a variety of di®erent machines and operating systems.Simply by installing a
JVM on each of the platforms,the program would be able to run without any
changes to it.
Disadvantages
Java is a pure object oriented language [40].This means the design of the system
would have to be changed to be organised in terms of classes and objects rather
than the modular design it currently has,if Java is to be used.The author is also
more comfortable with the procedural programming style rather than the object
oriented approach.
The inclusion of features such as an automatic garbage collector and the fact it is
an interpreted language means that Java is generally recognised as being a fairly
28
slow language [6].This is a disadvantage because,as discussed previously,speed
is important in certain parts of the system.
As Java hides away the complexities of the underlying machine away from the
programmer,the language may be considered to be less °exible than C/C++.
There may be occasions when implementing the systemwhere having tight control
over the machine may be bene¯cial,but this is something which Java will not be
able to allow.
As a company Pulsic does most of its programming in C or C++.This means
that it will be more di±cult to integrate a system programmed in Java into
Pulsic's software than if the system were programmed in either C or C++.
4.1.4 Python
Python is an object oriented language that is good for graphical user interface
programming.This is because of its Tkinter package which is a thin object
oriented layer on top of Tcl/Tk - a graphical user interface toolkit [41].As such,
Python is the language that Pulsic use to implement all their GUIs.
Pulsic also have a customised set of GUI widgets known as pwidgets.pwidgets
are created simply by putting a'wrapper'around the normal Tcl/Tk widget set
to give the widgets a look and feel that make them unique to Pulsic.
Since the circuit analysis system will be integrated into Lyric and will use the
same GUI,this project will use Python for any additions that need to be made
to the interface.This is the obvious choice so that there wont be any need to
integrate a foreign programming language into the GUI software.Furthermore,
its imperative that any additions to the GUI retain the same look and feel as the
rest of it.Therefore it is necessary that the added widgets are created from the
pwidgets widget set,which means using Python.
4.1.5 Conclusion
Having considered the pros and cons of the major languages available for imple-
menting the circuit analysis system,the decision has been made to use C.This is
because the author has extensive knowledge of the language which should reduce
implementation time and also the majority of the Lyric source code is written in
C which should mean integrating the two systems should be an easier task.
Furthermore,C is a fast language which will help enable the system to meet its
performance requirements,and the language supports the procedural program-
29
ming paradigm which is the author's preferred programming style.
As for implementing the graphical user interface of the system,the only option
really is to use Python since it will enable an easy integration into Lyric and also
maintain Lyric's look and feel.
4.2 Integrated Development Environments (IDEs)
An integrated development environment (IDE) is a piece of software that assists
programmers to develop programs and usually consists of a source code editor,
a compiler,build automation tools and a debugger [39].They are generally
supposed to make developing software easier.
There are several IDEs available for the C programming language such as Eclipse,
C-Free and the CodeForge IDE.However,the author has never used an IDE and
feels comfortable programming without the aid of one.Therefore an IDE will not
be used when implementing the circuit analysis system.
4.3 Compilers
A compiler is a piece of software that converts the source code of a program
written in a programming language into machine code that can be executed by a
computer.There are many compilers available for the C programming language,
but the author has always used the GNU Compiler Collection (GCC) compiler
and feels most comfortable with it.
GCC is produced by the GNUProject and is free software,distributed by the Free
Software Foundation (FSF).Since it is open source software,GCC has been de-
veloped using the open source model in which users are treated as co-developers.
Users are given access to the source code and are encouraged to develop new fea-
tures,¯x bugs,update documentation and so on.Advocates of the open source
development methodology claim that in many ways it is better than the closed
source method (commercial development).The open source enthusiast Eric S.
Raymond goes so far as to suggest that the open source methodology is able to
produce higher quality software than any other methodology or technique [29].
These claims can be partly backed up by studies that have shown that the time
taken from bug discovery to bug ¯x is shorter in open source development.
The result for GCC is a highly developed and tested piece of software that should
be reliable and e±cient.For these reasons and since the GCCcompiler is available
both on the University of Bath machines and on the computers at Pulsic,GCC
30
will be the compiler of choice for this project.
4.4 Debuggers
Debuggers are software tools which help programmers identify the errors in their
source code.There are several debuggers available for C,such as the DDD and
GDB debuggers available on the Pulsic computers.
Much of the debugging required can be performed manually with the careful
insertion of print statements.However,for the di±cult to ¯nd problems such as
errors involved with pointers and memory allocation,the GDB debugger will be
utilised.
4.5 Revision Control Systems
Revision control is concerned with keeping track of the development and multi-
ple revisions of the same units of information.Revision control systems keep a
repository of ¯les which developers'check out'in order to work on them before
checking them back in and merging the changes made into the previous version.
A history of the revisions is kept to ensure that the ¯les can be reverted back to
a previous version if necessary.These systems allow a team of people to work on
the same ¯les without having to worry about the con°icts between their work or
fear of causing damage.
Apopular revision control systemis the Concurrent Versions System(CVS) which
is available on the Pulsic computers.However,since there will be just one de-
veloper working on the circuit analysis system,there is no need for a central
repository of ¯les for concurrent development and so CVS will not be utilised in
the development of the system or its documentation.
4.6 Backing Up Files
Unforeseen technical problems can occur within a computer,such as the hard
drive failing or becoming corrupt,that can lead to the loss of ¯les.For this
reason important ¯les should not be kept in just one location,they should be
'backed up'.This means copying the ¯les to some other storage device or devices
so that if any ¯les are somehow lost,there is always a reserve copy.
31
As a precaution in this project,the system's ¯les will be regularly backed up to
the university's servers,and also to a USB memory stick to allow for ¯le recovery
on the move.
4.7 Testing Technologies
Since Modi¯ed Nodal Analysis revolves around solving equations involving ma-
trices,Maple has been chosen as a platformfor testing the circuit analysis system
as it is being developed.
Maple is a computer algebra system and as such is excellent at manipulating
matrices quickly.It is intended that,during development the system will be
tested by giving it some simple inputs and then printing out the matrices that
are constructed and the solutions that are calculated.Meanwhile,for the same
inputs the matrices are constructed by hand and then the solutions found by
using Maple.Comparing the results will show whether the system is working
correctly or if some changes are required.
4.8 Summary
In summary,the language to be used for implementing the core systemis C,while
Python will be used for some of the GUI work.The compiler will be GCC,the
debugger will be GDB and Maple will be used for the testing of prototypes.The
next chapter will outline the high level design of the system.
32
Chapter 5
High Level Design
This chapter outlines the high level design decisions made in order to create a
system that satis¯es the requirements laid out in the Requirements Analysis and
Speci¯cation chapter.
5.1 System Structure
The system has been designed in a modular fashion so that the system consists
of a collection of units with a distinct functional purpose.It has been designed
this way as the problem lends itself more to a procedural approach rather than
the object oriented approach which would have led to the system being designed
as a collection of classes and objects.
Figure 5.1 displays the overall high level design of the system.It shows that the
system is broken down into ¯ve modules:the input module,the initial circuit
analysis module,the iterative circuit analysis module,the current calculation
module and the output module.
Following is a brief description of each of the modules:
²
Input - this module is responsible for receiving the input of the system
(circuit layout information) and storing it.This will involve entering the
data into the system's data structures in an organised and e±cient manner.
²
Initial Circuit Analysis - this module will carry out the ¯rst iteration of
circuit analysis on the circuit information held in the system's data struc-
tures.The focus of the circuit analysis at this point is to determine accurate
values for the voltage at each of the node in the circuit design.The nodal
voltages will then be saved back to the data structures ready for output.
33
Figure 5.1:The high level design of the circuit analysis system
²
Iterative Circuit Analysis - this module will take the results of the initial
analysis and perform further iterations of circuit analysis.The here is to
recalculate the nodal voltages very quickly in response to minor changes
made to the circuit so that errors in the design can be ¯xed in an e±cient
manner.
²
Current Calculator - since the circuit analysis modules only calculate the
nodal voltages,this module calculates the currents through the circuit
branches.This will be done by applying Ohm's law to a pair of nodes
and the branch between them.The results will be saved back to the data
structures.
²
Output - this module is concerned with returning the results of each iter-
ation of the circuit analysis to the user.The most important method of
output will be via a graphical user interface,although the results of the
circuit analysis will also be printed to a ¯le.This is so users will have a
record of the results,but also because it will be useful for system testing.
5.2 Initial Circuit Analysis Algorithm
As stated in Section 3.1.4 on page 20,the requirement of the initial circuit analysis
phase is accuracy and reliability with the speed of the algorithm of less impor-
tance.With this in mind,the decision of which method to use for this phase of
34
circuit analysis is to be made by examining the methods used by existing circuit
analysis systems.The rationale for this line of thought is that existing circuit
analysis systems which are or have been used in the commercial world will use
methods which are accurate and reliable,since it is demanded by the users.
It was shown in Section 2.4 on page 10 that two of the most well known circuit
analysis systems are ECAP II and SPICE.SPICE performs circuit analysis by
using the well known nodal analysis method,while ECAP II uses a more obscure
method known as the mixed method.These two methods could be implemented
in this project so a decision has to be made on which to use.
Nodal analysis is a straight-forward concept and there is extensive literature on
the subject explaining the process and its implementation.The mixed method
however is a much more di±cult algorithm and the availability of pertinent lit-
erature is much less.
The success of the two circuit analysis systems may also be an indication of the
relative merits of their two analysis methods.SPICE is the most popular circuit
analysis tool available today,while ECAP II is rarely used if at all.This may
indicate that nodal analysis is the better basis for a circuit analysis tool.
For these reasons,this project's circuit analysis system will follow the SPICE
method.However this still leaves the question of which version of the nodal
analysis method to use.Early versions of SPICE were based on standard nodal
analysis while later versions utilised the Modi¯ed Nodal Analysis (MNA) ap-
proach.
The problem with standard nodal analysis is that it can become complicated
when the e®ects of voltage sources have to be taken into account;either a sep-
arate equation has to be written for each source,which increases the number of
equations to be solved,or the supernode method must be used which makes the
implementation more di±cult [5].Since the ring and mesh circuits that the sys-
tem has to analyse has many virtual voltage sources that have to be considered,
using the standard nodal analysis could be problematic.
MNA has no such problems trying to analyse the e®ects of voltage sources.Fur-
thermore,it is easier to implement algorithmically on a computer [5],which is a
signi¯cant advantage.For these reasons,the circuit analysis system will imple-
ment MNA as the basis for the initial circuit analysis module.
35
5.3 Matrix Inversion Algorithms
When Modi¯ed Nodal Analysis is applied to a circuit with only passive elements
(resistors) and independent current and voltage sources,as will be the case in
ring and mesh circuitry,the result is a matrix equation of the form shown in
Equation 5.1.
Ax = z (5.1)
For a circuit with n nodes and m independent voltage sources:A is an (n + m)
* (n + m) matrix and consists only of known quantities,x is an (n + m) * 1
matrix (or vector) that holds the unknown quantities and z is an (n + m) * 1
matrix that holds only known quantities.
Equation 5.1 is solved through matrix manipulation,forming Equation 5.2 from
which the solutions can be easily read-o®.Creating Equation 5.2 requires ¯nd-
ing the inverse of matrix A.There are several methods for ¯nding the inverse
of a matrix and so following will be a discussion of some of those techniques,
culminating in a decision of which will be used in this project.
x = A
¡1
z (5.2)
Gauss Jordan Elimination is a well known method for manipulating matrices.It
is the most basic technique available and is straightforward,understandable and
"solid as a rock"[26].Its ease of implementation,however,is o®set by a number
of drawbacks,one of which is that it is not a very fast technique.This,along
with the other drawbacks which will not be detailed here but can be found in
[26],led to it being stated that:
"Gauss-Jordan elimination should usually not be your method of ¯rst
choice,either for solving linear equations or for matrix inversion."[26]
LU decomposition is another technique for manipulating matrices and is the
method of choice for SPICE.The algorithm is fairly easy to implement and has
an operations count which is roughly a factor of three better than Gauss Jordan
Elimination,while having no obvious °aws.These features have resulted in LU
decomposition being described as the preferred way to solve the linear set of
equations of the form shown in Equation 5.1 [26].
As mentioned in Section 2.5.2 on page 15,Cholesky decomposition is a technique
that is relatively easy to implement while being a factor of two faster than LU
36
decomposition.However,since the algorithm only works on square matrices that
are both symmetric and positive de¯nite and the matrices produced by MNA can
not be guaranteed to be of this form,Cholesky decomposition has to be ruled
out as a candidate for use in the circuit analysis system.
The ¯nal possibility to be considered here is Kron's method of tearing.While this
method is faster than those examined here,it is disadvantaged by being by far
the most di±cult to implement and the fact that it is prone to round o® errors.
After considering the options,it has been concluded that LU decomposition will
be implemented to solve the matrix equations generated by the Modi¯ed Nodal
Analysis.While it is not the fastest choice,as stated in Section 3.1.4 on page
20,speed is not the most important requirement of the ¯rst iteration of circuit
analysis.LU decomposition is however fairly fast,is easy to implement and is
used by SPICE which indicates that it is a reliable mechanism.
5.4 Iterative Circuit Analysis Algorithm
As stated in Section 3.1.4,the iterative circuit analysis'most important require-
ment is its speed.It must produce results fast enough for an automated tool to
perform corrections to the circuit design e±ciently.Just using Modi¯ed Nodal
Analysis with LUdecomposition will not be su±cient since calculating the inverse
of a large matrix will take too long.Therefore a di®erent algorithm is required
for this section of the system.
The key for any algorithm to be suitable for this task is the ability to localise
computation so that when a small change is made to the resistance of a branch
in a circuit,the whole circuit does not have to be recalculated.The literature
review in Section 2 provides two candidate algorithms for this.
The ¯rst of these is Kron's method of tearing,which although is based on ma-
nipulating matrices,manages to do so in a manner which localises computation
so that the whole matrix does not need recomputing.However,as mentioned
in Sections 2.5.1 and 5.3,this method is very di±cult to implement and is also
prone to round-o® errors.Therefore,Kron's method of tearing can be ruled out.
The second option is the randomwalk algorithm.This algorithmis based on per-
forming circuit analysis using the well known mathematical random walk princi-
ple.As such,the algorithm manages to compute a nodal voltage by performing a
series of random walks that only encompasses part of the circuit,thus localising
computation.In theory,this is an ideal solution for the iterative circuit analysis
module.
37
Figure 5.2:Convergence toward actual voltage over a number of walks
However,this algorithmis relatively new,with the paper introducing it only being
published three years ago.As a result,there is very little supporting literature
detailing the ¯ner points of the algorithm and its implementation.Therefore,the
random walk algorithm would be very di±cult to implement and so will not be
used in the circuit analysis system.
Since the two main candidates for the iterative analysis algorithmhave been ruled
out,the alternative is to design and implement a whole new algorithm.Although
the random walk algorithm has been ruled out,some lessons can be learnt from
it that can help the design of the new algorithm.
In the random walk algorithm,a number of random walks are performed across
the circuit to determine the value of a nodal voltage.As the number of walks
performed increases,the result found for the voltage oscillates and converges on
the true value as seen in Figure 5.2.The algorithm stops once the oscillation of
the voltage change drops below a certain speci¯ed value,thus limiting the amount
of computation.
The new algorithm which is to be developed will use this basic convergence tech-
nique and will follow the following basic steps in response to a change in resistance
of a branch:
²
Calculate the the new voltages at the nodes at either end of the modi¯ed
38
branch using Ohm's law.This is the ¯rst iteration of the algorithm.
²
Store the new voltages of these two nodes
²
Successive iterations will calculate and store the new voltages of nodes
connected to the nodes analysed in the previous iteration.In this way,the
circuit analysis will spread outwards from the initial branch.
²
When the new voltage for a node is calculated and it is di®erent to the
original voltage by less than 0.01 percent,the node is °agged and the circuit
analysis will not spread outwards past this node.In this way,the analysis
area will reach a maximum radius from the initial branch,thus localising
computation while also ensuring the calculations have an acceptable level
of accuracy.
²
When the maximum radius of nodes is reached,repeated iterations will
contract the area of circuit analysis back down to zero as the nodal voltages
converge towards their true new values.
²
The process stops when all new nodal voltages di®er from the values in the
previous iteration by less than 0.01 percent.
Through a series of iterations,this algorithm will converge on the new values of
nodal voltages caused by the change of resistance in one branch.
Figure 5.3 shows a section of a mesh circuit and gives a basic illustration of the
algorithm.The yellow branch in the centre has been fattened or thinned which
changes its resistance.The ¯rst iteration of the algorithm calculates the new
nodal voltages for the red nodes,the second iteration calculates the new voltages
for the blue nodes,and then the third iteration does likewise for the green nodes.
At this point,the change in the voltage is less than one percent so the changes are
su±ciently small to stop spreading the circuit analysis further.This process is
then repeated until all the nodal voltages have converged to within a one percent
oscillation in value,upon which the voltages are accurate enough to stop the
analysis.
5.5 Units To Be Used
The values that the system will have to deal with can range from one amp to a
few nano amps in current,with similarly sized values for voltages and resistances.
These wide ranging values bring up the question of what units to use for these
quantities.For instance,since the current is always likely to be less than an amp,
all the measurements for current could use milliamps or microamps as the base
unit.
39
Figure 5.3:Possible iterations of circuit analysis using the iterative algorithm
While this would avoid having to manipulate very small numbers,it could lead
to confusion when having to use equations such as Ohm's law where calculations
have to be made with voltages,currents and resistances.If they were in di®er-
ent base units such as milliamps,microvolts and ohms,incorrect answers may
accidentally be produced if it were forgotten that the numbers were in di®erent
scales.For this reason,the units that will be used in this system will be the
standard SI units of amps,volts and ohms.
5.6 Data Structures
The circuit analysis system will have to be able represent matrices and also store
data on the various components contained within a circuit design.Here follows
a brief outline of the required data structures:
²
matrices - the data structure to represent matrices will have to be some sort
of two dimensional array of °oating point numbers.The array will have to
be dynamically allocated however,as it will not be possible to know the
size of the circuit until after the program has started to run.
²
nodes - the information that will need to be stored about nodes will include
a unique identi¯cation number,the branches,voltage sources and current
sources connected to it and the node's voltage.
²
resistors - these are actually branches or wires in a circuit,but they are
modelled as resistors.The pertinent data that needs storing about branches
includes an id number,the two nodes between which the branch is connected
and the resistance of the branch.
40
²
current sources - an identi¯cation number,the current it draws from the
circuit and the node to which it is connected needs to be stored about each
current source in the data structures.
²
voltage sources - the information that needs to be stored about voltage
sources includes and id number,the node to which it is connected and the
value of the voltage it supplies to the circuit.
5.7 GUI
Since the circuit analysis system is to be integrated into Lyric,it will use Lyric's
user interface.A menu will be added to the interfaces toolbar from which the
circuit analysis can be initiated.The section of circuitry on which to perform
circuit analysis will be chosen by selecting an area of the circuit design with the
mouse.
It is often said that it is a mistake when designing a user interface to associate
meanings with particular colours since ten percent of men are colour blind [30].
However,the decision has been made that the best way to display the results
of the analysis so that they are clear is to use colour.The rationale behind this
decision is that it should be far easier to see if a wire is carrying too much current
by simply looking at its colour rather than interpreting a symbol or reading a
number which is annotated on the wire.To display the colour,a mask will be
laid over the display of the circuit that will render each node and branch of the
circuit a certain colour depending on their associated values.
The most obvious colours to use are the colours of tra±c lights.For instance
green is to be used for wires whose current is well within the limits,amber for
wires whose currents are approaching the limit and red for wires whose current
exceeds the speci¯ed limit.This way the colours used in the GUI are associated
with something that will be known by the vast majority of users.