Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

Problem with me or my vendor?

Status
Not open for further replies.

Hawaslsh

Full Member level 3
Joined
Mar 13, 2015
Messages
164
Helped
5
Reputation
10
Reaction score
7
Trophy points
1,298
Location
Washington DC, USA
Activity points
3,422
Hello all,

I attached a much more detailed analysis with all the measured and simulated data, but I will give a summary here as well.
For over a year we have been working with a specific vendor and specific flexible substrate; Dupont Pyralux AP915R, 5 mil of polyimide with copper cladding on both sides. We have had three fabrication runs for various projects. Each run contains a few test coupons we can measure to ensure our circuit models, substrate, and fabrication are all within tolerance.
24 GHz patch.jpg

The first fabrication run centered around 24 GHz and we had a 24 GHz patch antenna test coupon to test. To model and predict the antenna performance, we use Cadence's Microwave Office. For this first fabrication run we relied on the material's data sheet when modeling the substrate within Microwave Office. This first run worked as predicted. There were slight variations in the predicted loss and resonant frequency, but nothing out of the ordinary when comparing simulations to reality.
5800 Patch.jpg

The second fabrication run centered around 5.8 GHz and we had a 5.8 GHz patch antenna test coupon to test. Again we used Microwave office to design and predict the performance of the test coupon. The simulated antenna modeled fairly well as compared to the fabricated antenna, see attachment for the actual data.
slot.jpg

The third run fabrication run had two projects, centered around 5.8 and 11.3 GHz. Instead of patch antennas, the test coupons were slot antennas. The slots were designed using the same method as before, however, the measured values were drastically different than our models predicted. There is always some variation between simulated and measured results, but this discrepancy was too large to be attributed to fabrication tolerance and slight variations in the substrate.
I tried a lot of different things in my model to try and pinpoint where I might have been wrong. Out of frustration I tried a crazy change to the dielectric constant of the material, from the nominal 3.4 to 4.4. That change alone made both the 5.8 and 11.3 GHz models match their measured counterparts.
The attachment has a lot more detail and data to back up my claim, but I think the substrate my vendor used the last run is different than my prior runs. And not slightly different, VERY different. Am I ok to call out my vendor (as nicely as possible) and ask if something changed or if they actually used the right substrate? Is there some other reason someone can think of for my last fabrication run to be so drastically different than my prior runs?

Sorry for the long post, happy to provide more context.
Thanks Sami
 

Attachments

  • Anomalous Results with Recent Flex Antennas.pdf
    583.8 KB · Views: 89
  • PyraluxAPclad_DataSheet.pdf
    80.8 KB · Views: 84
  • EI-10142-Kapton-Summary-of-Properties.pdf
    2.4 MB · Views: 104

I'd start with requesting a detailed build sheet
(or whatever you call it in your world) which
has the materials used. declared plainly. Then
you can read it before making accusations.

The ~ 30% difference I think could have many
explantions (e.g. thickness, skin effects,
proximity effects) that want checked before
you pin the tail on a random donkey.
 

    Hawaslsh

    Points: 2
    Helpful Answer Positive Rating
The ~ 30% difference I think could have many
explantions (e.g. thickness, skin effects,
proximity effects)

1617170837647.png

A varied thickness was one of the first thing I tried, in hindsight I should have included it in the presentation. This flex material only comes in certain thicknesses; 5 mil being the thickest. I plotted the simulated results for 1 mil, that doesn't seem to the culprit.

Not sure why skin effects would have any affect here. 35 um of copper is will beyond any skin depth issues and my EM simulator is taking into account metal thickness and real resistivities.

As to proximity, all the antennas were measured the same way. Any proximity affects would have been noticed during other measurements.

I pinged the vendor for a build sheet and made sure they included all the info they had about the substrate. Will be interesting to see if the latest run was done with a new lot of their DuPont material.

Thanks,
Sami
 

Can you show the mesh details (mesh preview) in the radial stub region?

Can you upload the geometry in DXF format, so that we can double check using another EM solver?
 

    Hawaslsh

    Points: 2
    Helpful Answer Positive Rating
Hello,
Thanks for taking the time!
Can you show the mesh details (mesh preview) in the radial stub region?
Mesh_Preview.png

I started with a grid size of 50um, and the adaptive meshing algorithm did the rest.

Can you upload the geometry in DXF format, so that we can double check using another EM solver?
Yes, that would be much appreciated. Usually I have access to a few other tools, but working from home I can't simply walk over and restart the license servers. I attached a zip file with 6 files. Three DXF files: the 24 GHz Patch, the 5.8GHz Patch, and the 5.8 GHz slot. Also included are 3 s1p files which are the measured results for each of the structures.
The datasheet for the material was posted before, but the details of the substrate I used; er = 3.4, Tand = 0.003, thickness = 125 um, and metal thickness on both sides of 35um.
Please let me know if you need any additional information, happy to provide more context.
Thanks again for any and all help,
Sami
 

Attachments

  • DXF_and_Measured_Data.zip
    41.7 KB · Views: 87

I simulated your 5.8 GHz slot antenna using Empire XPU (FDTD), with very fine meshing. Results with the stackup that you documented are closer to measurement than yours.

If we add solder resist, that shifts down the antenna resonance a lot! If we just use one thick flat layer for solder resists, that over-estimates the actual effect. In the plots below, I tried to model solder resist accurately using 3D extrusion with the actua conductor outline.

In summary, I think your frequency shift is from simulation accuracy: frequency should shift down if you include solder resist, and even without solder resist your resonance looks a bit high compared to my results. You can try finer mesh density, especially the junction at the radial stub seems to be sensitive to meshing.

What do you think?
BR, Volker
 

Attachments

  • meshlines.png
    meshlines.png
    132 KB · Views: 101
  • crosssection.png
    crosssection.png
    50.9 KB · Views: 112
  • resist.png
    resist.png
    70.9 KB · Views: 109
  • slot_5g8_s11.png
    slot_5g8_s11.png
    111.2 KB · Views: 115

    Hawaslsh

    Points: 2
    Helpful Answer Positive Rating
Axiem (which probably you used also) get identical results as you, but HFSS with the same settings shows results identical as your measurements.
So in this particular case, I would trust more your measurements, than some simulations...
 

Attachments

  • Axiem.jpg
    Axiem.jpg
    116.8 KB · Views: 99
  • HFSS.jpg
    HFSS.jpg
    176 KB · Views: 108

    Hawaslsh

    Points: 2
    Helpful Answer Positive Rating
But S11=-11dB in HFSS is not "identical" to measured -23dB?
 

Response frequency is identical...deep of S11 may have minor importance if is already below -10dB
 

Response frequency is identical...deep of S11 may have minor importance if is already below -10dB

I am sceptical ... from what I have seen with this slot antenna model, we get (too) low resonance frequency if the radial stub area has (too) low mesh density. Can you double check with more mesh iterations?

Also, I see a strong effect from solder resist for this slot antenna (but not for the patch antennas). It is a surprise if you get the "correct" results with an incomplete stackup.
 

Thanks a lot! This was exactly what I needed. Sometimes you get so laser focused its hard to see the obvious.

If we add solder resist, that shifts down the antenna resonance a lot!
I tried to add the solder mask in my prior work, but the 2.5D simulator I am using does a pretty poor job when the metal thickness is greater than a dietetic layer attempting to cover it. I followed up with our vendor for a more detailed datasheet on their LPI and an approximate thickness they would predict. Once I switch to a FEM solver I will start to include the layer.
Can I ask what material properties you assigned to the layer? The datasheet for the PSR-9000 FXT Series resist they use was pretty darn limited.

Axiem (which probably you used also) get identical results as you, but HFSS with the same settings shows results identical as your measurements.
So in this particular case, I would trust more your measurements, than some simulations...
Yes, i did use Axiem. We had good success with Axiem in regards to the patch antennas, so we made the drastic assumption it would be OK with the slot. I suppose that might have to do with the fact Axiem usually assumes an infinite ground plane, and our slot most definitely is not infinite. I tried the same problem is Analyst, Microwave office's FEM solver. It got closer to the measured value, but still quite a bit off from the measured results.
Our HFSS license server got booted off our network, once I have wrestled IT into submission, I will try it with HFSS.
Follow up question, how did you set up your port in HFSS? A wave port at the end of the transmissions line, or something different?

Thanks to all! This forum is one the of the most friendly, helpful, and knowledgeable places on the internet.
 

I used er=4 for solder resist, thickness 20µm. That is what PCB manufacturers have specified in other projects.

What you can do with planar solvers like Axiem, to get an idea about solder resist influence: embed the entire metal into a layer of solder resist that covers metal height + 10µm on top. That is too thick then in the no-metal areas, but it will give an idea if the circuit is sensitive to solder resist. Frequency shift with that simple "planar thick" solder resist was about twice of the more accurate "conformal coating" model.

Regarding your Axiem model(s), I would repeat that with finer mesh. You cannot expect accurate results with such a coarse mesh around the slot. We expect hight edge currents around the slot, and such big mesh cells can't model that accurately.

Good luck!
Volker
 

Attachments

  • axmesh.png
    axmesh.png
    232.4 KB · Views: 96

When I increased on both simulators the mesh density (doubled), the S11 frequency response went up in frequency about 70MHz from initial resonance (1.3%).
Changing frequency response (up or down in frequency) when increasing the mesh density is a complex mechanism, and is mainly related to the complexity of the structure, because convergence is not always monotonic. From this reason, not always when increase the mesh density will give better reults.
 

Status
Not open for further replies.

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top