How can I successfully run this simulation?
Hello,
I am trying to run the FEKO Solver for my model, but I keep running into this error (using MoM solver):
ERROR 33238: Not enough memory available on the hard disk for an out-of-core file
I have tried deleting things to make more room and I have tried switching to double precision, but it still won't work.
I have also tried using MLFMM but I get this error:
ERROR 32945: It is highly recommended to use the conventional MoM, since the MLFMM near field matrix fill exceeds 20% of the full MoM matrix
When I ran the MLFMM solver I switched to single precision because it was recommended.
I am not sure what I can do solve this issue. I have provided a copy of my model down below. Thanks!
Yolanda Gonzalez.
Answers
-
Hi Yolanda,
Unfortunately I can't download your model from the specified source. Maybe you can just attach it directly here?
Your problem is that you want to calculate a model that is too large for your computer (RAM). The required memory is determined by the number of mesh elements. These in turn depend on the maximum frequency on the one hand, and on the geometry itself on the other. If your geometry is very complex, it is best to simplify it. Very often electrically small geometric details (electrically small = small related to the wavelength) do not play a role for the result, but inflate the computational requirements unnecessarily.
I hope this helps!
Best regards,
Torben0 -
Hello,
Thank you for responding! How would you recommend I simplify my geometry? Would I have to change it the model in AutoCAD?
Also, I am unable to add the model directly because the file is too large. Please let me know if this link works.
Thanks,
Yolanda Gonzalez
0 -
Hi Yolanda,
The download has now worked, and I can see that your model is a complex PCB. Simulating complex PCBs is anything but recommended in Feko, due to the extremely small geometric details. There are much better tools for this, such as Altair Pollex.
Nevertheless, I will try to improve your model a bit so that you can do your own investigations.
Best regards,
Torben1 -
Hi Yolanda,
I think that the dimensions of the PCB are wrong and it needs to be scaled. Presently the substrate has a size of 50 m x 40 m
What's the actual size?
Best regards,
Torben1 -
Hi Yolanda,
The frequency is defined as 2.45 Hz. Could it be that it is supposed to be 2.45 GHz?
1 -
Hello,
Sorry for the late response! I was on break
Yes, the frequency is supposed to be in GHz. Also, the units were supposed to be in mm instead of m.
I changed the units to mm and changed the frequency to 2.45GHz. I ran the simulation again but got the same error.
ERROR 33238: Not enough memory available on the hard disk for an out-of-core file
Yolanda Gonzalez
0 -
Hi Yolanda,
the finite thickness of the metallic components should not be modeled. This only costs large amounts of memory and runtime due to the many mesh elements. Instead, you should only use surfaces to which you then assign a "virtual" thickness. I did this with your model, and all metallic surfaces are now copper with a thickness of 35 µm (variable t_copper, you can change this as you wish).
Could you please check again whether 35 µm is assumed correctly? In your model, the finite thickness is 2 µm, which seems a bit unrealistic to me. The thickness of your substrate is also unusually thin at 14 µm.
If the thicknesses are correct, you should next consider which of the metallic surfaces have no influence on the antenna and can therefore be omitted. This can save you an enormous amount of memory and runtime. You should also check whether the back of the substrate is also metallic.
As the model has now been simplified by me, about 23 GB of memory is required for the simulation (see attachment).
Best regards,
Torben1 -
Hello,
How were you able to create surfaces for the copper? When I import my model, it sets all of the components to "primitive".
Also, why did you decide to use an edge port instead of a wire port?
Yolanda Gonzalez
0 -
Also, I can confirm that the finite thickness (thickness for the copper) is 2 µm and the substrate thickness is 14 µm.
Yolanda Gonzalez
0 -
Hello,
After reviewing your model, I was able to figure out how to create the surfaces for the copper. I added metallic properties to the copper and defined its virtual thickness to be 2 µm. I added dielectric properties to the substrate. I have attached my model below, please let me know if you are able to download it.
I tried to run the simulation, but I got this error:
ERROR 2701: A wire segment at a connection point is too thick as compared to the triangle size
Is this why you changed the port to an edge port?
Yolanda Gonzalez
0 -
Hi Yolanda,
Maybe it would be a good idea to have a look at the Getting Started.pdf and also at the Example Guide.pdf. This is what I usually recommend to new Feko users to get an overview and some basic knowledge. This will help you to get started quickly without stumbling over the usual beginner's questions.
A few comments on your model:
- Wire ports must not lie exactly on an interface. Your port should therefore either be in the substrate or above it. For example like this:
I prefer edge ports just because they feel more "elegant"
- All geometry parts that touch each other (in your case substrate and copper parts) must be united in a Union. Otherwise they will be meshed individually, resulting in intersecting triangles.
- The physical thickness is still present in the model F_antenna_3D_11.29.iges.cfx. Wrong model attached? To get the single faces I simply copy them (the ones I want to keep) and then delete the rest (you've probably already guessed that):
- Due to the complexity of the model, a lot of memory (~50 GByte) and runtime is required for the simulation. If you want to simulate PCBs in Feko, you should consider whether there are parts that are not relevant and can therefore be omitted.
I attached the model with the correct copper thickness of 2 µm. Note, I also changed the bottom layer of the substrate to copper (just guessing).
Best regards,
Torben1 - Wire ports must not lie exactly on an interface. Your port should therefore either be in the substrate or above it. For example like this:
-
Hello,
Thank you so much for the help! I really appreciate it!
I was wondering if you could help me understand this error (error 2757).
I made some changes to the port and got this error when I tried to run the simulation.
Thanks,
Yolanda Gonzalez
0 -
Yolanda Gonzalez said:
Hello,
Thank you so much for the help! I really appreciate it!
I was wondering if you could help me understand this error (error 2757).
I made some changes to the port and got this error when I tried to run the simulation.
Thanks,
Yolanda Gonzalez
Hi Yolanda,
You have to distinguish between errors and warnings. Feko LOVES warnings
In this case, Feko points out that at least 10 mesh triangles are small compared to the defined thickness of the copper. But that shouldn't be a problem. There are always very small triangles somewhere...Best regards,
Torben1 -
Hello,
I ran your version of the model (the one you sent me) through the simulation and it worked great! I tried to make changes to my version of the model according to what we have previously discussed and your version of the model, but I am still struggling with it. I tried making some adjustments and ran it through the simulation, but I got the storage error again.
I have attached my updated model below. If you are unable to open it, please let me know. If you could take a look at it and let me know what you think, I would greatly appreciate it.
Thanks,
Yolanda Gonzalez
0 -
Hi Yolanda,
You still have the physical thickness of the copper in your model. You need to selct all the faces you want to keep and delete the rest. Else you will have a LOT of mesh elements, extremely close to each other.
In the attached video you see what I mean. (Sorry that some steps in the video take a long time. There are just a lot of faces until it is reduced.)
Best regards,
Torben1 -
Torben Voigt_20420 said:
Hi Yolanda,
You still have the physical thickness of the copper in your model. You need to selct all the faces you want to keep and delete the rest. Else you will have a LOT of mesh elements, extremely close to each other.
In the attached video you see what I mean. (Sorry that some steps in the video take a long time. There are just a lot of faces until it is reduced.)
Best regards,
TorbenHello,
Thank you for the video! That helped clear up the steps you were talking about earlier.
My simulation runs now but it takes a very long time to complete. I tried to let the simulation run over night, but I got an error telling me that it had quit due to an error in RunFEKO. Do you know why or how to fix something like this?
Thanks,
Yolanda Gonzalez
0 -
Torben Voigt_20420 said:
Hi Yolanda,
You still have the physical thickness of the copper in your model. You need to selct all the faces you want to keep and delete the rest. Else you will have a LOT of mesh elements, extremely close to each other.
In the attached video you see what I mean. (Sorry that some steps in the video take a long time. There are just a lot of faces until it is reduced.)
Best regards,
TorbenAlso, here is an updated version of my model.
0 -
Yolanda Gonzalez said:
Also, here is an updated version of my model.
Hi @Yolanda ,
You have set "continuous (interpolated) frequency range", which basically ensures that Feko only calculates as many frequencies as are required to accurately map the entire course. Unfortunately, this can lead to problems if some frequency ranges are difficult to calculate. In your case, the main problem is probably that you have set 1 Hz as the lower frequency limit. Feko is an HF solver, and of course cannot be used for DC or extremely low frequencies. You may still get accurate results at a few kHz (with double precision set), but certainly not much lower.
I have now set 1 GHz - 3 GHz and discrete frequency range (100 MHz steps), just for testing. Please let me know if it works now.
As written earlier, you should also think about simplifications. Certainly not all copper parts are relevant for the functioning of the antenna. This can greatly reduce the simulation time.
Best regards,
Torben0 -
Torben Voigt_20420 said:
Hi @Yolanda ,
You have set "continuous (interpolated) frequency range", which basically ensures that Feko only calculates as many frequencies as are required to accurately map the entire course. Unfortunately, this can lead to problems if some frequency ranges are difficult to calculate. In your case, the main problem is probably that you have set 1 Hz as the lower frequency limit. Feko is an HF solver, and of course cannot be used for DC or extremely low frequencies. You may still get accurate results at a few kHz (with double precision set), but certainly not much lower.
I have now set 1 GHz - 3 GHz and discrete frequency range (100 MHz steps), just for testing. Please let me know if it works now.
As written earlier, you should also think about simplifications. Certainly not all copper parts are relevant for the functioning of the antenna. This can greatly reduce the simulation time.
Best regards,
TorbenYou could also try to see if an infinite substrate might give you good results (see F_antenna_3D_12.7_alt_simplified_pGF.cfx). In the model I have also reduced the copper parts. The model can thus be simulated in a few seconds. The use of infinite dielectric layers is very efficient, but only appropriate if the effect of the edges of the subtrate does not play a major role.
0 -
Torben Voigt_20420 said:
You could also try to see if an infinite substrate might give you good results (see F_antenna_3D_12.7_alt_simplified_pGF.cfx). In the model I have also reduced the copper parts. The model can thus be simulated in a few seconds. The use of infinite dielectric layers is very efficient, but only appropriate if the effect of the edges of the subtrate does not play a major role.
Hello @Torben Voigt ,
I am not able to open the model's you are sending me. When I download them and try to open them, I get this error.
Yolanda Gonzalez
0 -
Yolanda Gonzalez said:
Hello @Torben Voigt ,
I am not able to open the model's you are sending me. When I download them and try to open them, I get this error.
Yolanda Gonzalez
Hi Yolanda,
You need to update Feko to version 2023.0.1:
Best regards,
Torben1 -
Hello,
Is it possible to run multiple simulations on one model without overwriting the previous results? I was able to complete a simulation, but I wanted to try different frequency increments. However, it won't let me run another simulation without overwriting the previous results.
Is there a way to do this?
Best,
Yolanda Gonzalez
0 -
Hello,
After reviewing my results and comparing them to the experimental results, I noticed there were some differences, and I was wondering if you might know why or what changes I should make to the model to get similar results to the experimental ones.
Here are the experimental results:
And here are my results:
For this simulation I used a start frequency of 1GHz and an end frequency of 3GHz. The number of frequencies was 200 and the frequency increments was 10.05MHz.
Best,
Yolanda Gonzalez
0 -
Yolanda Gonzalez said:
Hello,
After reviewing my results and comparing them to the experimental results, I noticed there were some differences, and I was wondering if you might know why or what changes I should make to the model to get similar results to the experimental ones.
Here are the experimental results:
And here are my results:For this simulation I used a start frequency of 1GHz and an end frequency of 3GHz. The number of frequencies was 200 and the frequency increments was 10.05MHz.
Best,
Yolanda Gonzalez
0 -
Yolanda Gonzalez said:
Hello,
After reviewing my results and comparing them to the experimental results, I noticed there were some differences, and I was wondering if you might know why or what changes I should make to the model to get similar results to the experimental ones.
Here are the experimental results:
And here are my results:For this simulation I used a start frequency of 1GHz and an end frequency of 3GHz. The number of frequencies was 200 and the frequency increments was 10.05MHz.
Best,
Yolanda Gonzalez
0 -
Yolanda Gonzalez said:
Hello,
Is it possible to run multiple simulations on one model without overwriting the previous results? I was able to complete a simulation, but I wanted to try different frequency increments. However, it won't let me run another simulation without overwriting the previous results.
Is there a way to do this?
Best,
Yolanda Gonzalez
Hi Yolanda,
You could create different StandardConfiguirations for this. Just make sure that you set Frequency to "Specify frequency per configuratrion" (instead of "Global"))
It is quite common to have a StandardConfiguration for the frequency band (e.g. for S-parameters) and a second one for just a single frequency (e.g. gain at resonance frequency).
Best regards,
Torben0 -
Yolanda Gonzalez said:
Hello,
After reviewing my results and comparing them to the experimental results, I noticed there were some differences, and I was wondering if you might know why or what changes I should make to the model to get similar results to the experimental ones.
Here are the experimental results:
And here are my results:For this simulation I used a start frequency of 1GHz and an end frequency of 3GHz. The number of frequencies was 200 and the frequency increments was 10.05MHz.
Best,
Yolanda Gonzalez
Hi Yolanda,
Your results (S11 > 1) suggest that something in the model is not correct. You might want to check if your port(s) are not defined correctly or maybe the mesh is extremely coarse? Were there any warnings from the solver?
Best regards,
Torben0 -
Torben Voigt_20420 said:
Hi Yolanda,
Your results (S11 > 1) suggest that something in the model is not correct. You might want to check if your port(s) are not defined correctly or maybe the mesh is extremely coarse? Were there any warnings from the solver?
Best regards,
TorbenHello,
There were only two warnings. One was that the triangles were too thick as compared to the lateral dimensions (2757). And the other was that further warnings of ID 2747 would be suppressed (32383). But other than that, there were no warnings.
I have added a PPT of my results from the simulations if it is any help. I have also added the newest version of the model so that you can see the changes we made to the port. I moved the port to the opposite side from where it originally was, and it yielded different results but still not the ones we need.
Thank you so much for your help with this! I really appreciate it!
Best,
Yolanda Gonzalez
0 -
Yolanda Gonzalez said:
Hello,
There were only two warnings. One was that the triangles were too thick as compared to the lateral dimensions (2757). And the other was that further warnings of ID 2747 would be suppressed (32383). But other than that, there were no warnings.
I have added a PPT of my results from the simulations if it is any help. I have also added the newest version of the model so that you can see the changes we made to the port. I moved the port to the opposite side from where it originally was, and it yielded different results but still not the ones we need.
Thank you so much for your help with this! I really appreciate it!
Best,
Yolanda Gonzalez
Hi Yolanda,
- When activating "Connectivity" in the Display Settings you can see that all of the copper faces are not connected to the substrate. In fact, they are 2 µm above the substrate.
- The port also looks a little strange. Maybe a wire port would be more feasible.
- I would suggest replacing the finite substrate with an infinite substrate layer. The thickness of 14 µm is so thin that it would have to be meshed very finely (this is probably the reason for your results). An infinite substrate (planar greens functions) saves an enormous amount of memory and runtime. The size of the original substrate also suggests that the outer edges have little influence, which is why an infinite substrate is legitimate.
Please have a look at the attached model.
Best regards,
Torben0 - When activating "Connectivity" in the Display Settings you can see that all of the copper faces are not connected to the substrate. In fact, they are 2 µm above the substrate.
-
Torben Voigt_20420 said:
Hi Yolanda,
- When activating "Connectivity" in the Display Settings you can see that all of the copper faces are not connected to the substrate. In fact, they are 2 µm above the substrate.
- The port also looks a little strange. Maybe a wire port would be more feasible.
- I would suggest replacing the finite substrate with an infinite substrate layer. The thickness of 14 µm is so thin that it would have to be meshed very finely (this is probably the reason for your results). An infinite substrate (planar greens functions) saves an enormous amount of memory and runtime. The size of the original substrate also suggests that the outer edges have little influence, which is why an infinite substrate is legitimate.
Please have a look at the attached model.
Best regards,
TorbenHello @Torben Voigt ,
1) The reason for the copper being 2 µm above the substrate is because when I was deleting the unnecessary faces (to create a virtual thickness) I only kept the top faces instead of the bottom faces. I will change that on my model on my end.
2) As for the port, I had moved it to the opposite side because, after discussing with the professor, that is where the power should originate from. Because the opposite side is not "even", the port looks a little weird. I will try a wire port and see if this changes my results.
3) Also, after reviewing your model, I will see how the infinite substrate effects the results of the simulation. I am running your model to compare our results and to try and find any major differences. I will get back to you once I have the results.
Thank you for your help!
Best,
Yolanda Gonzalez
0 - When activating "Connectivity" in the Display Settings you can see that all of the copper faces are not connected to the substrate. In fact, they are 2 µm above the substrate.
Altair Employee