Part a - Sensitivity Analysis in ADS
Uploaded by AgilentEEsof on 18.02.2010
Transcript:
Hello and welcome to another presentation and demo on design for manufacturing in ADS,
but this one is going to be on sensitivity analysis.
This slide represents the DFM design process for MMIC, and you can see once you start your
design, start your nominal design here, usually designers, the next step they optimize their
design. What I’m showing here, once you have the setup, the circuit setup to be optimized,
you can automatically access the sensitivity analysis tool. It takes a few seconds to get
a fast response or a fast analysis on the sensitivity of the components in your design.
So let’s see how sensitivity analysis works in ADS.
Basically in ADS sensitivity analysis what it does, it takes one component at a time
in your design. For example, here if we have a capacitor C1 what sensitivity analysis does
is it increases the capacitor C1 to C1 prime. And C1 prime is one times E-6 larger than
C1. So it increases the component by a very, very small factor, which is one times E-6,
one times 10 to the -6 of the nominal value, and by doing this it will look for the response,
what is the gain? What is the return loss? What is the noise figure, whatever you’re
looking for.
So it looks for the response at the nominal value and it also looks at the response of
the circuit at the new value, which is one times ten to the minus six of the nominal
value. And you can see the difference in the response, we represent it with delta R, and
the difference in the component value, which is delta C here. So basically the sensitivity
is calculated to be delta R divided by delta C, and this is the amount of change you get
in the response relative to that little amount, one times ten to the minus six factor, of
the component value. Okay?
So I just want to mention here that sensitivity analysis is local. It’s localized. That
means it is taken at one component at a time regardless of the interaction of the other
components. We just take resistor R, do the sensitivity analysis only on that resistor,
then we do capacitor, then we do the other components. But not all together varying,
so it’s localized, okay, to kind of give you a quick analysis on the sensitivity of
that part.
I want to mention, in the previous video, we have other videos and presentations on
yield sensitivity histograms, all the parts vary at the same time and it gives you much
more meaningful results on the sensitivity of that component relative to everything moving
at the same time. But sensitivity analysis again is localized, it’s done on one component
at a time. But it’s really quick, in a few seconds you can access that, you can simulate
it. In like a few seconds you get an idea of how sensitive these components are. But
I do recommend to use the yield sensitivity histograms and the design of experiments to
make a full understanding of the behavior of your design for manufacturing.
So here’s an example of a run in ADS. Notice here I have the input matching network capacitor
C1, I have the interstage matching network capacitor C8 and C4, and I have the output
matching network capacitor C6 and C7, and I’m looking here at the S22 spec. Notice
that C1 input matching network and C8 interstage matching network has no effect, no sensitivity
to the output return loss. But as we move closer to the output matching network we find
out C4, which is close to the output matching network, it’s in the interstage matching
network, has a little effect sensitivity to the S22. But the biggest sensitivity is coming
from the capacitor in the output matching network, which makes sense.
So C6 is the red X component here, or the sensitive part, and C7, which might be a big
huge bypass capacitor, had no effect because it’s a big capacitor just for bypassing
purposes. So this is how you can tell, you list the components on the X-axis and you
list the measure you’re looking at on the Y-axis, and just visually you can see clearly
what component is more sensitive than another.
One thing I want to mention in ADS when you do sensitivity analysis you can access two
types of analysis, two types of results. We have the absolute sensitivity and we have
the normalized sensitivity, but they both give you a different meaning to the sensitivity.
And I do have a document written specifically to explain the difference between absolute
sensitivity and normalized sensitivity, please feel free to access that by asking for it
from the Agilent representative. But let me illustrate to you in the next slide normalized
sensitivity calculation is more meaningful to designers.
I’m a designer and always, 99% of the time I always access the normalized sensitive analysis.
I recommend that you use it because basically what it does (skip in tape) the results have
more meanings to you. It gives you the percent change in the response versus a 1% change
in the component value. It’s all normalized, you have many components for every 1% change
in that component value you can see the percent change in the response; in the gain, in the
noise figure, in the power output. So this is normalized, it’s very easy to compare
which component is sensitive more than another component.
In the absolute sensitivity calculations what ADS calculates or gives you, it gives the
change in the response, like the change in the gain for example, versus one unit change
in the component value. If you have a 10-ohm resistor, for one unit change, for 11 ohm,
from 10 to 11 ohms what is the gain change? But be careful when you compare apples with
apples here, when you compare the units. This is why I recommend the normalized sensitivity
calculation, it’s all normalized to percent change in the component value so you can really
compare, you can compare the analysis, sensitivity analysis relative to all components equally
because they're all changed by the same factor, percent.
So let me now show you a demo using ADS. Okay, so let’s go to ADS (skip in tape) the same
LNA, the KU band LNA design I used in the other demos, the design of experiments and
the yield sensitivity histogram demo. By the way, this project is available in ADS examples
directory under microwave circuits, it is the KU band LNA DFM circuit. So you can access
it and simulate it yourself.
But again this is the input matching network and this is the interstage or the FET structure,
here is the resistor lines in the FET. And here’s the output matching network, the
output port, the DC port which takes the voltage supply, DC port. So as I said before, designers
usually go through the nominal design followed by optimization. So notice here –
but this one is going to be on sensitivity analysis.
This slide represents the DFM design process for MMIC, and you can see once you start your
design, start your nominal design here, usually designers, the next step they optimize their
design. What I’m showing here, once you have the setup, the circuit setup to be optimized,
you can automatically access the sensitivity analysis tool. It takes a few seconds to get
a fast response or a fast analysis on the sensitivity of the components in your design.
So let’s see how sensitivity analysis works in ADS.
Basically in ADS sensitivity analysis what it does, it takes one component at a time
in your design. For example, here if we have a capacitor C1 what sensitivity analysis does
is it increases the capacitor C1 to C1 prime. And C1 prime is one times E-6 larger than
C1. So it increases the component by a very, very small factor, which is one times E-6,
one times 10 to the -6 of the nominal value, and by doing this it will look for the response,
what is the gain? What is the return loss? What is the noise figure, whatever you’re
looking for.
So it looks for the response at the nominal value and it also looks at the response of
the circuit at the new value, which is one times ten to the minus six of the nominal
value. And you can see the difference in the response, we represent it with delta R, and
the difference in the component value, which is delta C here. So basically the sensitivity
is calculated to be delta R divided by delta C, and this is the amount of change you get
in the response relative to that little amount, one times ten to the minus six factor, of
the component value. Okay?
So I just want to mention here that sensitivity analysis is local. It’s localized. That
means it is taken at one component at a time regardless of the interaction of the other
components. We just take resistor R, do the sensitivity analysis only on that resistor,
then we do capacitor, then we do the other components. But not all together varying,
so it’s localized, okay, to kind of give you a quick analysis on the sensitivity of
that part.
I want to mention, in the previous video, we have other videos and presentations on
yield sensitivity histograms, all the parts vary at the same time and it gives you much
more meaningful results on the sensitivity of that component relative to everything moving
at the same time. But sensitivity analysis again is localized, it’s done on one component
at a time. But it’s really quick, in a few seconds you can access that, you can simulate
it. In like a few seconds you get an idea of how sensitive these components are. But
I do recommend to use the yield sensitivity histograms and the design of experiments to
make a full understanding of the behavior of your design for manufacturing.
So here’s an example of a run in ADS. Notice here I have the input matching network capacitor
C1, I have the interstage matching network capacitor C8 and C4, and I have the output
matching network capacitor C6 and C7, and I’m looking here at the S22 spec. Notice
that C1 input matching network and C8 interstage matching network has no effect, no sensitivity
to the output return loss. But as we move closer to the output matching network we find
out C4, which is close to the output matching network, it’s in the interstage matching
network, has a little effect sensitivity to the S22. But the biggest sensitivity is coming
from the capacitor in the output matching network, which makes sense.
So C6 is the red X component here, or the sensitive part, and C7, which might be a big
huge bypass capacitor, had no effect because it’s a big capacitor just for bypassing
purposes. So this is how you can tell, you list the components on the X-axis and you
list the measure you’re looking at on the Y-axis, and just visually you can see clearly
what component is more sensitive than another.
One thing I want to mention in ADS when you do sensitivity analysis you can access two
types of analysis, two types of results. We have the absolute sensitivity and we have
the normalized sensitivity, but they both give you a different meaning to the sensitivity.
And I do have a document written specifically to explain the difference between absolute
sensitivity and normalized sensitivity, please feel free to access that by asking for it
from the Agilent representative. But let me illustrate to you in the next slide normalized
sensitivity calculation is more meaningful to designers.
I’m a designer and always, 99% of the time I always access the normalized sensitive analysis.
I recommend that you use it because basically what it does (skip in tape) the results have
more meanings to you. It gives you the percent change in the response versus a 1% change
in the component value. It’s all normalized, you have many components for every 1% change
in that component value you can see the percent change in the response; in the gain, in the
noise figure, in the power output. So this is normalized, it’s very easy to compare
which component is sensitive more than another component.
In the absolute sensitivity calculations what ADS calculates or gives you, it gives the
change in the response, like the change in the gain for example, versus one unit change
in the component value. If you have a 10-ohm resistor, for one unit change, for 11 ohm,
from 10 to 11 ohms what is the gain change? But be careful when you compare apples with
apples here, when you compare the units. This is why I recommend the normalized sensitivity
calculation, it’s all normalized to percent change in the component value so you can really
compare, you can compare the analysis, sensitivity analysis relative to all components equally
because they're all changed by the same factor, percent.
So let me now show you a demo using ADS. Okay, so let’s go to ADS (skip in tape) the same
LNA, the KU band LNA design I used in the other demos, the design of experiments and
the yield sensitivity histogram demo. By the way, this project is available in ADS examples
directory under microwave circuits, it is the KU band LNA DFM circuit. So you can access
it and simulate it yourself.
But again this is the input matching network and this is the interstage or the FET structure,
here is the resistor lines in the FET. And here’s the output matching network, the
output port, the DC port which takes the voltage supply, DC port. So as I said before, designers
usually go through the nominal design followed by optimization. So notice here –