Introduction to Simulation - Lecture 1 Example Problems and Basic Equations Jacob White

Thanks to Deepak Ramaswamy, Michal Rewienski, Luca Daniel, Shihhsien Kuo and Karen Veroy

Outline
Uses For Simulation
Engineering Design Virtual Environments Model Verification

Course Philosophy Example Problems

Power distribution on an Integrated Circuit Load bearing on a space frame Temperature distribution in a package

Circuit Analysis
Equations
Current-voltage relations for circuit elements (resistors, capacitors, transistors, inductors), current balance equations

Recent Developments
Matrix-Implicit Krylov Subspace methods

Electromagnetic Analysis of Packages

Equations
Maxwells Partial Differential Equations

Recent Developments
Fast Solvers for Integral Formulations

Structural Analysis of Automobiles

Equations
Force-displacement relationships for mechanical elements (plates, beams, shells) and sum of forces = 0. Partial Differential Equations of Continuum Mechanics

Recent Developments
Meshless Methods, Iterative methods, Automatic Error Control

Drag Force Analysis of Aircraft

Equations
Navier-Stokes Partial Differential Equations.

Recent Developments
Multigrid Methods for Unstructured Grids

Engine Thermal Analysis

Equations
The Poisson Partial Differential Equation.

Recent Developments
Fast Integral Equation Solvers, Monte-Carlo Methods

Micromachine Device Performance Analysis

Equations
Elastomechanics, Electrostatics, Stokes Flow.

Recent Developments
Fast Integral Equation Solvers, Matrix-Implicit Multi-level Newton Methods for coupled domain problems.

Stock Option Pricing for Hedge Funds

Option Price Stock Price

t
Equations
Black-Scholes Partial Differential Equation

Recent Developments
Financial Service Companies are hiring engineers, mathematicians and physicists.

Virtual Environments for Computer Games

Equations
Multibody Dynamics, elastic collision equations.

Recent Developments
Multirate integration methods, parallel simulation

Virtual Surgery
Equations
Partial Differential Equations of Elastomechanics

Recent Developments
Parallel Computing, Fast methods

Biomolecule Electrostatic Optimization

+ +

+- +-

- - + + +
+

Ligand (drug molecule)

Equations

-+
Receptor (protein molecule)
Ecm protein

The Poisson Partial Differential Equation.

Recent Developments
Matrix-Implicit Iterative Methods, Fast Integral Equation Solvers

The Computer Simulation Senario

Problem too complicated for hand analysis
Toss out some Terms Macromodel Solve a Simplified Problem Simulate using a canned routine, a friends advice, or a recipe book

Works!
D R O P C L A S S

Way too slow

Develop Understanding of Computational complexity

Only works sometimes

Develop Understanding of Convergence Issues

No

Make Sense? Yes Anxiety

Faster Method
Right Algorithms

Robust Method
New Algorithms

Happiness

Fame

Course Philosophy
Examine Several Modern Techniques
Understand, practically and theoretically, how the techniques perform on representative, but real, applications Why Prove Theorems? Guarantees, given assumptions, that the method will always work. Can help debug programs. The theorem proof can tell you what to do in practice.

Power Distribution for a VLSI Circuit

+ 3.3 v

Cache

ALU

Decoder

Power Supply Main power wires

Is there at least 3v across the ALU ?

One application problem which generates large systems of equations is the problem of distributing power to the various parts of a Very Large Scale Integrated (VLSI) circuit processor. The picture on the left of the slide shows a layout for a typical processor, with different functional blocks noted. The processor pictured has nearly a million transistors, and millions of wires which transport signals and power. All one can really see be eye are the larger wires that carry power and patterns of wires that carry signals, boolean operations such as and and or . A typical processor can be divided into a number of functional blocks, as diagrammed on the layout on the left. There are caches, which store copies of data and instructions from main memory for faster access. There are execution units which perform boolean and numerical operations on data, such as and, or , addition and multiplication. These execution units are often grouped together and referred to as an Arithmetic Logic Unit (ALU). Another main block of the processor is the instruction decoder, which translates instructions fetched from the cache into actions performed by the ALU. On the right is a vastly simplified diagram of the processor, showing a typical 3.3 volts power supply, the 3 main functional blocks, and the wires (in red) carrying power from the supply to the 3 main functional blocks. The wires, which are part of the integrated circuit, are typically a micron thick, ten microns wide and thousands of microns long ( a micron is a millionth of an inch). The resistance of these thin wires is significant, and therefore even though the supply is 3.3 volts, these may not be 3.3 volts across each of the functional blocks. The main problem is we address is whether or not each functional block has sufficient voltage to operate properly.

Load Bearing Space Frame

Joint Beam

Droop Cargo

Attachment to the ground

Vehicle

Does the Space Frame droop too much under load ! ?

In the diagram is a picture of a space frame used to hold cargo (in red) to be lowered into a vehicle. The space frame is made using steel beams(in yellow) that are bolted together at the purple joints. When cargo is hanging off the end of the space frame, the frame droops. The main problem we will address is how much does the space frame droop under load.

Thermal Analysis

Does the engine get too hot ! ?

Above is a picture of an engine block, which is typically solid steel or aluminum. The heat generated by the gas burning in the cylinders must be conducted through the engine block to a wide enough surface area that the heat can be dissipated. If not, the engine block temperature will rise too high and the block will melt.

Design Objectives for the VLSI problem

+ 3.3 v

Cache

ALU

Decoder

Select topology and metal widths & lengths so that a) Voltage across every function block > 3 volts b) Minimize the area used for the metal wires

Design Objectives for the Space Frame

Droop

Select topology and Strut widths and lengths so that a) Droop is small enough b) Minimize the metal used.

Thermal Analysis
Select the shape so that a) The temperature does not get too high b) Minimize the metal used.

First Step - Analysis Tools

Droop + 3.3 v Cache ALU Decoder

Given the topology and metal widths & lengths determine a) The voltage across the ALU, Cache and Decoder. b) The droop of the space frame under load.

Who uses VLSI Tools ?

Several big companies

IBM, Motorola, TI, Intel, Compaq, Sony, Hitachi Non functional prototype costs - Increases time-to-market - Design rework costs millions

Once a VLSI circuit is designed, it is fabricated using a sequence of sophisticated deposition and etching processes which convert a wafer of Silicon into millions of transistors and wires. This processing can take more than a month. If the circuit does not function, the design flaw must be found and the fabrication process restarted from the beginning. For this reason, just a few design errors can delay a product for months. In a competitive market, this delay can cost millions in lost revenue in addition to the cost of redesigning the circuit. In order to avoid fabricating designs with flaws, companies make extensive use of simulation tools to verify design functionality and performance.

Who uses VLSI Tools ?

1000s of small companies

Small companies make application circuits disk drives, graphics accelerators, CD players, cell phones. What is the cost of non-functional prototypes ? - Out of business.

Thousands of small companies design VLSI circuits for applications as diverse as peripherals for personal computers as well as signal processors for audio, video and automotive applications. These small companies cannot afford the cost of fabricating prototype designs that do not function. The very survival of these companies depends on using simulation tools to verify designs before fabrication.

Who makes VLSI Tools ?

Company employees sales Cadence 4,000 Synopsis/ 5,000 Avanti Mentor 2,600 Graphics Market cap.

1.3 billion 3.8 billion 1.5 billion 6.9 billion .6 billion 1.4 billion

Companies compete by improving analysis efficiency.

Modeling VLSI circuit Power Distribution

+ 3.3 v

Cache

ALU

Decoder

Power supply provide current at a certain voltage. Functional blocks draw current. The wire resistance generates losses.
SMA-HPC 2003 MIT
25

Each of the elements in the simplified layout, the supply, the wires and the functional blocks, can be modeled by relating the voltage across that element to the current that passes through the element. Using these element constitutive relations, we can construct a circuit from which we can determine the voltages across the functional blocks and decide if the VLSI circuit will function.

Modeling the Circuit

Supply becomes
A Voltage Source

Power supply
current

V
+ Vs

V = Vs
Constitutive Equation

+ Voltage
Physical Symbol Current element

The power supply provides whatever current is necessary to ensure that the voltage across the supply is maintained at a set value. Note that the constitutive equation (in the figure) , which is supposed to relate element voltage (V) to element current (I) does not include current as a variable. This should not be surprising since voltage is always maintained regardless of how much current is supplied, and therefore knowing the voltage tells one nothing about the supplied current.

Modeling the Circuit

Functional blocks become

Current Sources

+
ALU

V
Is

I = Is
Constitutive Equation

Physical Symbol

Circuit Element

The functional blocks, the ALU, the cache and the decoder are complicated circuits containing thousands of transistors. In order to determine whether the functional block will always have a sufficient voltage to operate, a simple model must be developed that abstracts many of the operating details. A simple worst-case model is to assume that each functional block is always drawing its maximum current. Each block is therefore modeled as a current source, although one must assume that the associated currents have been determined by analyzing each functional block in more detail. Note that once again the constitutive equation is missing a variable, this time it is voltage. Since a current source passes the same current independent of the voltage across the source, that V is missing should be expected.

Modeling the Circuit

Metal lines become

Resistors

I
Physical Symbol

IR V = 0
Constitutive Equation (Ohms Law)

Circuit model

Length resistivity R= Area

Design Parameters Material Property

The model for the wires connecting the supply to the functional blocks is a resistor, where the resistance is proportional to the length of the wire ( the current has further to travel) and inversely proportional to the wire cross-sectional area ( the current has more paths to choose).

low Resistance

high Resistance

That the current through a resistor is proportional to the voltage across the resistor is Ohms law.

Modeling VLSI Power Distribution

Putting it all together

IC
Cache
+ -

IALU
ALU

ID
Decoder

Power Supply voltage source Functional Blocks current sources Wires become resistors Result is a schematic

To generate representation which can be used to determine the voltages across each of the functional units, consider each of the models previously described. First, replace the supply with a voltage source. Second, replace each functional block with an associated current source. Third, replace each section of wire with a resistor. Note that the resistors representing the wires replace a single section with no branches, though the section can have turns. The resulting connection of resistors, current sources and voltage sources is called a circuit schematic. Formulating equations from schematics will be discussed later.

Modeling the Space Frame

Bolts Struts

Ground

Load

Example is simplified for illustration

In order to examine the space frame, we will consider a simplified example with only four steel beams and a load. Recall that the purple dots represent the points where steel beams are bolted together. Each of the elements in the simplified layout, the beams and the load, can be modeled by relating the relative positions of the elements terminals to the force produced by the element. Using these element constitutive relations, we can construct a schematic from which we can determine the frames droop.

Modeling the Frame

Load becomes
Force Source

Fload
Mass Schematic Symbol x

Fx = 0 Fy = Fload
Constitutive Equation

Physical Symbol

Fload = Mass Gravity

The load is modeled as a force pulling in the negative Y direction ( Y being vertical, X being horizontal). Note that the constitutive equation does not include the variable for the loads position, following from the fact that the loads force is independent of position.

Modeling the Frame

Strut Beam Physical Symbol

Beam becomes
Strut

x1 , y1
v f

L = (x1 x2 )2 + ( y1 y2 )2 v L L f = EAc 0 L0
Constitutive Equation (Hookes Law)

x2 , y2

L0 = Unstretched Length Ac = Cross-Sectional Area

Design Parameters

E = Young's Modulus

Material Property

In order to model the steel beams in a space frame, it is necessary to develop a relation between the beam deformation and the restoring force generated by the beam. To derive a formula we will make several assumptions. 1) The beam is perfectly elastic. This means that if one deforms the beam by applying a force, the beam always returns to its original shape after the force is removed.

Apply force

Remove force

Lo
2) The beam does not buckle

L1 > Lo
buckling

Lo

Lo

Apply force No buckling

Buckling is an important phenomenon and ignoring it limits the domain of applicability of this model.

L1 > Lo

3) The beam is materially linear. For a beam to be materially linear, the force which acts along the beam is directly proportional to the change in length.

Lo
f= KL

f=0

f = K L

L1
To determine K consider that the force required to stretch a beam an amount (I) Inversely proportional to its unstretched length (It is easier to stretch a 10 inch rubber band 1 inch than to stretch a 1 inch rubber band 1 inch) (II) Directly proportional to its cross-sectional area (Imagine 10 rubber bands in parallel) (III) Dependent on the material (Rubber stretches more easily than steel). Combining (I), (II) and (III) leads to the formula at the bottom of the slide.

is

Modeling the Frame

Putting it all together

Load

How much does the load droop?

To generate a representation which can be used to determine the displacements of the beam joints, consider the models previously described. First, replace the loads with forces. Second, replace each beam with strut.

Formulating Equations from Schematics

Two Types of Unknowns Circuit - Node voltages, element currents Struts - Joint positions, strut forces Two Types of Equations Conservation Law Equation Circuit - Sum of Currents at each node = 0 Struts - Sum of Forces at each joint = 0 Constitutive Equation Circuit - element current is related to voltage across the element Struts - element force is related to the change in element length
SMA-HPC 2003 MIT
35

Conservation Laws and Constitutive Equations

Heat Flow
1-D Example

Incoming Heat

T (0)
Near End Temperature Unit Length Rod

T (1)
Far End Temperature

Question: What is the temperature distribution along the bar

T

T (0)
SMA-HPC 2003 MIT

T (1)

x 36

Conservation Laws and Constitutive Equations

Heat Flow
Discrete Representation

1) Cut the bar into short sections 2) Assign each cut a temperature

T (0)

T (1)

T1

T2

TN 1 TN
37

SMA-HPC 2003 MIT

Conservation Laws and Constitutive Equations

Heat Flow
Constitutive Relation

Heat Flow through one section

Ti

Ti +1 hi +1,i = heat flow =

Ti +1 Ti x

hi +1,i
Limit as the sections become vanishingly small T ( x ) lim x 0 h ( x ) = 38 x SMA-HPC 2003 MIT

Conservation Laws and Constitutive Equations

Heat Flow
Conservation Law

Two Adjacent Sections

control volume

Incoming Heat (hs )

Ti 1 hi ,i 1

Ti

hi +1,i Ti +1

x Net Heat Flow into Control Volume = 0 SMA-HPC 2003 MIT hi +1,i hi ,i 1 = h s x

39

Conservation Laws and Constitutive Equations

Heat Flow
Conservation Law

Net Heat Flow into Control Volume = 0

In com ing H eat ( h s )

hi +1,i hi ,i 1 = hs x
T i 1 hi , i 1 Ti
x

hi + 1, i Ti + 1

Heat in from left

Heat out from right

Incoming heat per unit length

Limit as the sections become vanishingly small

SMA-HPC 2003 MIT

lim x 0 hs ( x ) =

h ( x ) T ( x ) = x x x

40

Conservation Laws and Constitutive Equations

Heat Flow
Circuit Analogy

Temperature analogous to Voltage Heat Flow analogous to Current

1 = R x T1
+ -

TN
+ -

vs = T (0)

is = hs x

vs = T (1)
41

SMA-HPC 2003 MIT

Formulating Equations
Two Types of Unknowns Circuit - Node voltages, element currents Struts - Joint positions, strut forces Conducting Bar Temperature, section heat flows Two Types of Equations Conservation Law Equation Circuit - Sum of Currents at each node = 0 Struts - Sum of Forces at each joint = 0 Bar Sum of heat flows into control volume = 0 Constitutive Equations Circuit element current related to voltage Struts - strut force related to length change Bar section temperature drop related to heat flow
SMA-HPC 2003 MIT
42

Formulating Equations Circuit Example from Schematics Identifying Unknowns

1

vs
3 4

+ -

Assign each node a voltage, with one node as 0

Given a circuit schematic, the problem is to determine the node voltages and element currents. In order to begin, one needs labels for the node voltages, and therefore the nodes are numbered zero, one, two, N, where N+1 is the total number of nodes. The node numbered zero has a special meaning, it is the reference node. Voltages are not absolute quantities, but must be measured against a reference. To understand this point better, consider the simple example of a current source and a resistor.

0
1

1
In order for one Amp to flow through the resistor, V1 - V0 must equal one volt. But does V1 = 11 volts and V0 =10 volts Or is V1 = 101 volts and V0 = 100 volts ? It really does not matter, what is important is that V1 is one volt higher than V0. So, let V0 define a reference and set its value to a convenient number, V0 = 0.

Formulating Equations Circuit Example from Schematics Identifying Unknowns

i5
0 1

i1

i2

i4

i3

Assign each element except current sources a current

The second set of unknowns are the element currents. Obviously, the currents passing through current sources are already known, so one need only label the currents through resistors and voltage sources. The currents are denoted

i1, i2,

ib , where b is the total number of unknown element currents. Since elements

connect nodes, in an analogy with graphs, element currents are often referred to as branch currents.

Formulating Equations Circuit Example from Schematics Conservation Law

i5
i1 + i 5 i 4 = 0
0 1

i2
is 1 i 1 + i 2 = 0

i1

is 2 + is 3 i 2 i 5 = 0

is1

is 2
4

is 3
i3
3

i4

i 4 is1 is 2 i 3 = 0

i 3 is 3 = 0

Sum of currents = 0 (Kirchoffs current law)

The conservation law for a circuit is that the sum of currents at each node equals zero. This is often referred to as Kirchoffs current law. Another way to state this law, which more clearly indicates its conservation nature is to say
Any current entering a node must leave the node.

The conservation is that no current is lost, what comes in goes out. The green statement also makes it clear that the direction of the current determines its sign when summing the currents. Currents leaving the node are positive terms in the sum and currents entering the node are negative terms ( one can reverse this convention but one must be consistent).

Formulating Equations Circuit Example from Schematics Constitutive Equations

R5 i5 = 0 V2

R5
R2
R2 i2 =V1 V2
2

R1
R1 i1 = 0 V1

R4
R4 i4 =V4 0
4

R3
3

R3 i3 =V3 V4

Use Constitutive Equations to relate branch flow from plus node to minus currents to node voltages (Currents node)
Each element with an unknown branch current has an associated constitutive equation which relates the voltage across the element to the current through the element. For example, consider

R2 in the figure,

V2

i2 R2

V3

The constitutive relation for a resistor is Ohms law.

And in this case V = V2 -

V3 and I = i2.

1 V = I R
R I.

Onse should again take note of the direction of the current. If current travels from left node through the resistor to the right node, then the left node voltage will be higher than the right node voltage by an amount

Formulating Equations Circuit Example from Schematics Summary

Unknowns for the Circuit example Node voltages ( except for a reference) Element currents ( except for current sources) Equations for the Circuit example One conservation equation (KCL) for each node (except for the reference node) One constitutive equation for each element (except for current sources) Note that # of equations = # of unknowns

Many Applications of simulation

Picked Three Representative Examples Circuits, Struts and Joints, Heat Flow in Bar Two Types of Equations Conservation Laws Circuit - Sum of Currents at each node = 0 Struts - Sum of Forces at each joint = 0 Bar - Sum of heat flows into control volume = 0 Constitutive Equation Circuit current-voltage relationship Struts - force-displacement relationship Bar - temperature drop-heat flow relationship
SMA-HPC 2003 MIT
48

Summary of key points