Models are Always Wrong, Right Whats the Difference Between an $85 Model and a $2,000,000 Model - PowerPoint PPT Presentation

Loading...

PPT – Models are Always Wrong, Right Whats the Difference Between an $85 Model and a $2,000,000 Model PowerPoint presentation | free to view - id: 15df32-NmQzM



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Models are Always Wrong, Right Whats the Difference Between an $85 Model and a $2,000,000 Model

Description:

Tassie, not Taz. Colin McAndrew. NACDM, April 2002. Laws of Simulation and Modeling. A simulation is only as accurate as the models it is based on ... – PowerPoint PPT presentation

Number of Views:103
Avg rating:3.0/5.0
Slides: 48
Provided by: AjA85
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Models are Always Wrong, Right Whats the Difference Between an $85 Model and a $2,000,000 Model


1
Models are Always Wrong, Right?Whats the
Difference Between an85 Model and a 2,000,000
Model
  • Colin McAndrew
  • April 4, 2002

2
Australia, not Austria
3
Australia, not Austria
4
Taz
5
Tassie, not Taz
6
Laws of Simulation and Modeling
  • A simulation is only as accurate as the models it
    is based on
  • A model is (mostly) useless unless it is embedded
    in a simulator
  • Models are, by definition, inaccurate its just
    a matter of degree
  • Models generally trade off complexity (simulation
    time) for accuracy

7
Zeroeth Law of Modeling
Physics (reality) trumps Mathematics
8
Zeroeth Law of Modeling
Measurements trump Models
9
Zeroeth Law of Modeling
Measurements trump Models
Especially when Engineers Use Polynomials
10
Numerical Tricks May not Work
cubic spline
11
Zeroeth Law of Modeling
Measurements trump Models
Just be careful you know what you are measuring
12
Negative Capacitance Blooper
13
Negative Capacitance Blooper
14
Simple way to Get Capacitance
  • Terminal currents from AC analysis
  • Sequentially excite terminals k with 1j0 volts
    at w1, i.e. f1/(2p)
  • Change sign if

15
Oz is Acceptable
16
Aussie, not Ozzie
17
Victoria
18
Charge or Capacitance?
  • Theyre really the same, right?

19
Charge or Capacitance?
C
v
20
Charge or Capacitance?
C
v
21
Will the IC Fail?
Real circuitry
Protection circuitry
22
Metal Routing Missing
Real circuitry
Protection circuitry
23
Simulation Results
  • Capacitance based model did not conserve charge,
    isolated protection circuitry voltage ramped to
    infinity!
  • Charge based model showed the missing metal would
    NOT cause the protection circuitry to trigger
  • 2,000,000 of already manufactured parts were
    shipped and not scrapped

24
Victoria
25
Victoria
26
Melbn, not Mell-born
27
Milling Coal for the SECV
stress
speed
28
Milling Coal for the SECV
stress
speed
29
Milling Coal for the SECV
stress
speed
30
Melbn, not Mell-born
31
Melbn, not Mell-born
32
Eastern Suburbs
33
Asymptotic Correctness
  • Any model should give physically correct and
    reasonable behavior as any controlling parameter
    approaches an extreme
  • Temperature going very cold or hot
  • Geometry getting very large
  • It is amazing how infrequently this simple
    principle is applied
  • 1/W and 1/L preferred to L and W

34
Universal SPICE 1/f Noise Model
35
Universal SPICE 1/f Noise Model
(at a fixed total current)
36
Universal SPICE 1/f Noise Model
37
Universal SPICE 1/f Noise Model
38
Symmetry
  • Most MOSFET and resistor models are cast in the
    form IdsGVds
  • G(Vsb,Vdb,Vgb) is the effective conductance of
    the MOSFET or resistor
  • Consider symmetric biasingVdbVdb0Vx,
    VsbVsb0-Vx, Vds2Vx
  • Ids MUST be an odd function of Vx
  • G MUST be an even function of Vx
  • Limiting of Vds to Vsat MUST be odd

39
Eastern Suburbs
40
Eastern Suburbs
41
Eastern Suburbs
42
Eastern Suburbs
43
Kew
44
Kew
45
Kew
46
Kew
47
Kew
85
About PowerShow.com