Bond wire application at 145 GHz.

[shmily0447]

Hi all,

I was trying to use co-planar bond wires to conncet an antenna and and IC that both work at 145 GHz. The bond wires bridge a separation of 300 um. It was found that:

(1) the maximum |S21| is -3 dB, which is okay for me
(2) the 3-dB transition bandwith is quite narrow as no more than 5 GHz
(3) the impedance bandwidth at each port is also no more than 5 GHz

I have tried:

(1) capacitive compensation
(2) impeance matching

But the bandwidth does not seem to increase.

Does anyone have ideas about how to increase the bandwidth? Any other unrelated comments or suggestions are also appreciated.

 

[WimRFP]

I am not an expert at 2 mm wavelength, but some things I can mention:

your gap is 300 um, so the length of the bond wires is about 0.25 lambda I think? You very likely have impedance steps in the path due to the different characteristic impedance of the bonding wire path. A quarter wave length of different impedance gives worst case mismatch. Can you post an image of what you are doing. If possible also mention the souce and load impedance.

The freedom you may have to change the characteristic impedance of the wires is the spacing between signal wires and return wires, and the number of wires.

 

[shmily0447]

Thanks for the information. I changed the separation to 100 um, however the situation does not change. The performance is still bad. Attached pls find the picture. All comments are welcome.

 

[WimRFP]

I am trying to understand what you did.

I assume the orange sturcture (right side) is a short piece of co-planar wave guide with ground under it and certain impedance. However the center strip conductor seems to continue as a microstrip (with higher impedance) that uses the ground plane. Are you feeding the right side as a microstrip?

The 6 bonding wire also form a kind of co-planar wave guide.

On the left side I see a very small orange trace is this a microstrip that uses the groundplane below the blue dielectric?

The pads for the return path bond wires on the blue dielectric, are they connected to the ground layer?

 

[kh4n]

what happens when you change wires to ribbons?

 

[shmily0447]

Your understanding for the whole structure is right. The pads on the blue substrate are properly grounded.

PS. the substrate thickness for the white color is 100 um, dielectric constant 5.75.
the substrate thickness for the blue is 5 um, dielectric contsant 2.7.

- - - Updated - - -

In our lab, there are many a few susccessful demonstration of bond wire connection at freqs higher than 145 GHz. So we think that we should stick to the usage of bond wires. The ribbons feature much lower insertion loss in my opinion, however, we do not have facilities in lab to impelement them.

 

[WimRFP]

@kh4n: The ribbons will reduce the Zc of the bond wire CPW, but it will not reduce it with factor 3 I think. What I don't understand is that the current structure has - 3 dB S21 bandwidth of just 5/145 = 3.4%. I can hardly image that this all happens because of bad impedance match of the bond wire CPW. So there must be something else going wrong. I hope Shmily will give us more info.

I was thinking of some weird common mode resonance of the bond wire CPW together with the rectangular capacitive ground planes at the right side of the circuit. But that should definitely change when changing the seperation.

 

[shmily0447]

I was trying to detect some possible resonances since the bandwidth is toooooo narrow, but nothing achieved at this moment. I am afarid that those are all the information that I can provide here. It is a project that we should not disclose too much. For my understanding, it is the first time that the bond wires are used at 145 GHz, so barely any refereces could be found. Do you guys have anything for me to refer?

 

[WimRFP]

Some things: What is the Zc of the thin microstrip on the blue dielectric?
What is the Zc of the bond wire CPW (you can simulate this with a seperate simulation, or use something as ATLC (sourceforge) ).
Similar questions for the transmission lines of the right part of the circuit. In my opinion there are severel impedance disontinuities

Do you believe that the floating ground planes at the right side of the circuit (orange rectangles) provide sufficient ground? Or are they grounded as well.

Can you run THIS simulation setup with ribbons, of so does it give better results? Why THIS? Just to rule out some other issue, so I mean to just change the wires into ribbon and keep all others the same. If you know it should be OK with ribbon, but you can't reproduce it in this simulation, then something else is wrong.

I believe the ohmic loss of ribbon will be less, but I can't explain the very narrow relative bandwidth of your circuit, it is just a few percent.

- - - Updated - - -

CPW structures with narrow co-planar ground traces have reasonable differential to common mode coupling (and vice versa). I once had this in an interconnect (but at lower frequency). A common mode resonance between the two PCB's appeared introducing lots of ringing in the signal. We coudn't use a coaxial interconnect.

We solved is by increasing the number of "ground" wires to emulate wide ground strips on both sides of the signal conductors. From my head about 8 wires left and right of the two center conductor wires.

 

[shmily0447]

Tks for the feedback.

I am now using HFSS. Can you advise on how to simulate the capacitance of a single device in HFSS, like the pad.

They are well grounded I think. I tried to change the size of the grounded pads, without obvious effect on the transition performance.

- - - Updated - - -

PS. I've tried to increase the grounded pad as well as increase the number of bonding wires that connect the grounded pads on both side, but nothing happens.

 

[WimRFP]

I don't know HFSS, but you can draw the patch and use a lumped or embedded port scheme to drive the "patch capacitor". From the impedance or S11, you can calculate the capacitance (or inductance). You can even put a lumped/localized port under the rectangle at he position of the bond wires and see what happens. I would recommend you to check the HFSS manual for info on lumped/localized ports.

Hand calculation with the parallel plate formula will not work here, as when I look to the structure, your structure may even be larger than an electrically quarter wave (so it may appear as an inductor). If they appear as a low impedance, you don't need to use vias to provide ground for the bond wire CPW.

 

[shmily0447]

Yes, it appears very low impedance. I am wondering why the impedance is so low. Can you help me to explain this?

 

[WimRFP]

If is has very low impedance, it behaves more or less as a quarter wave open stub. You probably know the radial stubs as used for creating bias points and short circuits in microwave circuit. If it would resonate in a halve wave mode (like a regular patch antenna), the impedance would be relatively high.

I am still not convinced that your arrangement can't work, as the requirements are not stringent.

You say: "blue substrate is 5 um thick". If it is, the landing pad for the two signal bond wires would add lots of capacitance. This will cause reflection. Your image suggests a thicker dielectric (for the blue one).

 

[shmily0447]

On the blue device, the landing pads are grounded on a 5um substrate. Beneath the 5um substrate, there is a 100um substrate that shares the same ground with the white part. It suddenly comes to me that maybe there are substrate modes in the 100um substrate there.

 

[WimRFP]

So for the blue device, there is a local ground plane (with size equal to the blue device) 5um below the trace and landing pads. The landing pads for the CPW ground wires are connected to this local ground plane. I assume that this local ground plane is not grounded to the large one that is under the complete circuit. Are you sure your port that drives the blue circuit uses the small local ground 5um below the landing pads as port ground?

The large capacitance from the landing pad to the local ground of the blue device remains (unless you removed some of the ground below it).

 

[shmily0447]

Your understanding is right. The local GND is not grouned to the large GND of the white device. The port drives the blue device uses the local GND, which is 5um beneath.

I can not do anything with the local GND of the blue device, since it is already fabricated. I will try to reduce the capacitance by other ways.

 

[WimRFP]

I can't guess the capacitance of the landing pad and Zc of the narrow trace (all on the blue device). If it turns out that the landing pad (for the signal wires) introduce large reflection, can you add an inductive stub that goes to the local ground to counteract the capacitance? You may use one of the landing pads for the CPW ground wires as ground for the stub.

 

[shmily0447]

Good suggetion. I will try to see what happen.

 

[biff44]

at those frequencies, the bond wires are way too long. You have to do all u can to reduce the length of the bond wires. I would then use very soft gold ribbon, and laying it flat on the board (no loop height) bond the 3 ribbons. by using 3 ribbons, and positioning them carefully, you will better approximate a transmission line impedance as opposed to just a wire loop with parasitic inductance.

Finally, on the right of your drawing, you show 2 monster sized square pads, but no details on what they are. Are the connected thru via holes to a back side? If not, those pads LxW are probably resonant at 145 GHz, and need to be made much smaller.

 

[shmily0447]

The separation that the bond wires are going to bridge over is fixed by the process. So I have to go with it. Yes, make as close as possible that the bonding wires to the device instead of hopping to hight is a good way to improve the performance, which I've taken into consideration. The 'two monster pads' are also grounded, which are calculated not to introduce resonance at 145 GHz. I've been puzzling with this problem for more than one month......frustrated.