Slide sem t

1
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Design of Integrated Inductors Through Selection
from a Database Created Using Electromagnetic
Simulation and Neural Networks
Romão Kowaltschuk1,2 Wilson Arnaldo Artuzi
Jr.1 Oscar da Costa Gouveia Filho1 1 - UFPR
Universidade Federal do Paraná 2 - Copel
Companhia Paranaense de Energia

September, 2003
2
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
OUTLINE

1. INTRODUCTION
2. INDUCTORS DESIGN
3. ELECTROMAGNETIC SIMULATION
4. RESULTS OF ELECTROMAGNETIC SIMULATION
5. NEURAL NETWORKS
6. CONCLUSIONS

3
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
INTRODUCTION
- Objective Transceptor complete integration. -
A problemPassive devices (inductors) integration.
Technology
Advantages
Negative Points

• - Speed
• - Highly resistive substrate
• Passive component construction
• is not difficult

- Low density of integration - High cost
Bipolar
GaAs
- Speed - High Fan-out avaiability
- Low density of integration
- Low power - High density of integration - Low
cost
- Conductive substrate
CMOS
Bipolar/CMOS
-Isnt avaiable in standard manufacturing
plants
- Joins advantages of bipolar/CMOS
4
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
INDUCTORS DESIGN
Design Variables
Inductors
- Devices with no standard design
- Too many variables to be chosen in design
Project Techniques
- Empirical formulations
- Analythic formulation derived from
electromagnetic theory
- Electromagnetic simulation (finite elements
and numerical methods)
5
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Basic Electrical Model of the Inductor
Lumped Parameters Considering the Spiral
CS capacitance of overlaying metal
layers Rs conductivity of spiral metal Ls
high frequency inductive effects of
that occur in the spiral metal
Lumped Parameters Considering the Substrate
Cox oxide capacitance RSi silicon
conductivity CSi high frequency capacitive
effects that occur in the semicondutor
6
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Electromagnetic Simulation - An Alternative
Solution
Belief the electromagnetic simulation
gives a good evaluation of
results, concerning the variation of reactance
with frequency, but it demands a lot of effort!
Solution to do electromagnetic simulation
automatically!
7
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Automatization of Electromagnetic Simulation
An inductor base case editor program for batch
simulation
An electromagnetic specific purpose simulator
(ASITIC) capable of providing continous outputs
for simulation cases of thousands of devices.
A program to classify results, due to the huge
amount of data! (a simple software written in VB6)
8
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Database Description
Geometric Specification of Inductors
Results of Electromagnetic Simulation
Typical Device Specified in Database
9
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Results of Electromagnetic Simulation
10
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
11
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
12
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Design Example
Database Variable
Search Criteria
- Normalized inductive reactance between input
and output terminals
5,0 nHltjX/jwlt 5,2nH
Radius lt 200 ?m
- Inductors spiral circumscript radius
- Resonant Frequency
f gt 3,5 GHz
Partial Vision of the Answer to the Requested
Question
Spiral Ident. Number
Radius (?m)
Operational freq. (MHz)
Normal. Induc. Reactance (nH)
Resonant Frequency (GHz)
175 .... 175 150 ..... 150
200 .... 1600 200 ..... 1400
5,094 .... 5,132 5,090 ..... 5,104
6,922 .... 8,796 7,088 ..... 6,580
931 .... 931 1048 ..... 1048
13
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Creating the Inductors Electrical Database Using
Neural Networks
Question how could it be possible to decrease
the time spent in eletromagnetic simulation?

A possible answer evaluating some values of the
electrical parameters database using neural
networks trained using a smaller set of data
obtained by eletromagnetic simulation.



14
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
Results Obtained Using Neural Networks
15
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
16
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
CONCLUSIONS

• The proposed design method enables evaluation of
  inductive reactances for a wide range of
  frequencies and can justify the development of
  the sofware tools and the avaiability of computer
  resources necessary to realize it.

• The evaluation of normalized inductive
  reactances, through electromagnetic simulation is
  the only theoretical model that shows the
  designer a trustable performance of inductors,
  as frequency varies in a wide range.

17
Design of Integrated Inductors through Selection
from a Database Obtained by Electromagnetic
Simulation and Neural Networks
CONCLUSIONS

• The alternative design method of creating some
  of the values necessary to complete a searchable
  database employing neural networks has achieved
  reasonable results just for evaluating reactances
  of big and medium size inductors (outer sides gt
  100 µm).
• For smaller devices, the performance of neural
  networks is not acceptable. The values obtained
  are worth just for indicating a range of values.