FAST AND ACCURATE MODELING OF

1
FAST AND ACCURATE MODELING OF EMBEDDED PASSIVES
IN MULTI-LAYER PRINTED CIRCUITS USING NEURAL
NETWORK APPROACH Q.J. Zhang, M.C.E. Yagoub,
X. Ding Dept of Electronics, Carleton
University, Ottawa, Canada R. Goulette, R.
Sheffield, and H. Feyzbakhsh Nortel Networks,
Ottawa, Ontario, Canada
2

• Dr. Qi-jun Zhang
• Professor
• Department of Electronics,
•  
• Carleton University,
• 1125 Colonel By Drive
• Ottawa, ON., K1S 5B6
• Phone  [phone]
• Email qjz_at_doe.carleton.ca

3
Acknowledgement This NCMS / AEPT Consortium
work is performed under support of the U.S.
Department of Commerce, National Institute of
Standards and Technology, Advanced Technology
Program, Cooperative Agreement Number
70NANB8H4025.
4
Outline

• Introduction
• Neural Network Modeling
• Training of Neural Models
• Neural Models of Passive Devices
• Examples
• Conclusion

5
Features

• New approach to modeling of high frequency
  effects of embedded passive components in
  multilayer printed circuits based on artificial
  neural networks.
• Training data generated by EM simulators
• Models are trained to learn the S-parameters of
  the passives versus physical and geometrical
  parameters.
• Models are fast and represent the EM based
  information of the components.
• Models can be used for efficient design of
  high-frequency circuits and systems.

6
Multilayer Perceptrons
(MLP) Structure
(Output) layer L
(Hidden) layer l
(Hidden) layer 2
(Input) layer 1
xn
7
Input-output Computation of a Neural Network
Model
2
2
1
1
8
Neural Net Training
Objective to adjust W such that minimize
(y - d)2 W x
S
9
Training Error dpk is the kth element of
dp where dp represents the measured/simulated
output y for the input xp, TR is the index set
of training data. is the kth
output of the neural network xp is the input
sample w is the weight vector
10
Neural Network Computation process xi is
the ith input to the neural network, Nl is the
number of neurons in layer l, is the
output of ith neuron of lth layer,
represents weight of the link between jth neuron
of l-1th layer and ith neuron of lth layer,
is the bias parameter of ith neuron of lth
layer, L is the total number of layers.
is a sigmoid function for the layers and
linear function for the output layer.
11
Training Validation Process
12
Embedded Resistor Neural Model
Embedded Resistor
Neural Model Representing EM Behaviors
13
Embedded Capacitor Neural Model
Embedded Capacitor
Neural Model Representing EM Behaviors
14
General Steps in 3-D EM Data Generation
.OPEN EM SIMULATOR .OPEN PROJECT . EXECUTE THE
FOLLOWING STEPS

• . DRAW the Structure
• . ASSIGN Materials for 3-D Objects
• . ASSIGN Boundaries for 2-D Objects
• . SELECT Frequency Range, Accuracy...
• .SOLVE
• .POST PROCESSOR Edit and Save Solution

.CLOSE EM SIMULATOR
15
EM-Based Neural Network Training Process
16
EXAMPLES OF NEURAL MODELS OF EMBEDDED
RESISTORS
17
EM-Based Neural Network Training Process
EM Simulators (Sonnet Lite Ansoft-HFSS)
S-Parameters (Training Data)
EM Based Neural Models
NeuroModeler
18
Embedded Resistor ( L W )
Length 8 mils Width 8 mils Resistivity 100
W /square
R
W
er
L
19
Comparison of S-parameters of the Embedded
Resistor for L W
Original EM data () EM based neural model
(__)
20
Embedded Resistor ( L 4W )
R
W
er
L
Length 32 mils Width 8 mils Resistivity 100
W /square
21
Comparison of S-parameters of the Embedded
Resistor for L 4 W
Original EM data () EM based neural model
(__)
22
Embedded Resistor ( W 4L )
Length 8 mils Width 32 mils Resistivity 100
W/square
R
er
W
L
23
Comparison of S-parameters of the Embedded
Resistor for W 4L
Original EM data () EM based neural model
(__)
24
Conclusion

• A neural network approach for modeling embedded
  passives in multilayer printed circuits have been
  presented.
• The models can be obtained by learning the data
  from electromagnetic simulators.
• The technique takes into account the 3-D EM
  effects in embedded passives during modeling and
  simulation of high-frequency circuits and
  systems.
• It helps making the design of multilayer printed
  circuits more accurate and efficient,
  contributing to overall reductions in design
  cycles.