Welcome to our site! EDAboard.com is an international Electronic 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.
Some people use the value of 1 nH per mm as a rule of thumb for typical wire bonds. In fact, it should be related to the radius as well as the length. It should also be related to the ground plane and the surronding such as coupled wire bonds. Basically, the current loop is critical. Regards.
Yes 1nH/mm is a good aproximation.
Of course as wire is shorter and thicker, as inductance is smaller.
2 parallel wire reduce inductance even if they are mutually coupled.
Two wire give less inductance than one wire (or ribbon) of the sum section.
For your case,
D=10 um is very thin, may be too mutch thin.
L=500um is very long, may be too long
Usually, gold, wedge, bond wire have a maximum L/D = 15
since i am using cascode configuration for my LNA and after adding bondwires of L=500 um and D=10um at the terminals of transistors the input impedance changes significantly. I have also added a degenration inductor which has very small value i.e less then 0.5 nH so can some tell me :
do i use same bondwire i.e with same lenght and D or can I use different values of bondwires?
what should be the typical value of bondwire inductance?
If possible, You may try to increase the wire diameter. 10 um (4 mils) is extremally thin. Less than suggested in most application. For example, the smaller HEMT made by Eudyna (Fujitsu) or Cel (NEC) have pads that support 17um thick wire. and is the 17 um Au wedge, the suggested wire!
The inductances of bonwires are not sufficient to design circuits. Also, mutual inductances and as well as package parasitic capacitances must be well modeled and should be taken into account when doing simulations. Especially over 2GHz, parasitic capacitances regarding to substrate,diepad and neighbourgh pins and bondwires must be very well modeled by using either Ansoft 3D Extractor or better HFSS 3D EM simulations.
I have had many bad troubles with these unwanted effects of a package at 5.8GHz and many of them are not solved yet..
The D=10um bonding wire is too thin. According to my experience, a typical diameter of 0.98mil/1mil/1.2mil(1mil=25.4um) gold wire is widely used in industry applications. As such, a very good rule of thumb to calculate the Parital Inductance is 1nH/mm, which is also the value you can get through most of the 3D qusia-static EM solvers such as Ansoft Q3D, PaksiE, Cadence APE3D etc. If you want to account for the bond wire effects at the system level, the Loop Inductance rather than the Partial Inductance should be used, which takes into account the effects of ground plane effect, coupling effect etc. And most of the time, a 3D EM full-wave solvers is needed(such as Ansoft HFSS, CST MicrowaveStudio).
Again, take note that a specified wire has fixed Parital inductance which, in most of the cases, is what the package parasitics L refers to. However, the same wire has different loop inductances when used in different systems, which is what a system design engineer need to take care of.
I am no expert, but I would suggest to isolate the problem to the bond wires,
try some of these configurations suggested and measure your results.
if there is little or no change to S11, then it is probably something else as BigBoss stated it could be, but if you see significant changes then it would logically be the BWs
This topic is quite interesting. Many of us just gave some comments based upon previous experiences and some common senses. I am curious how much of what we discussed is true or very solid, and I tried to simulate some wire bond cases with the dimensions discussed here using a full-wave EM simulator (IE3D). Following are what I got and the 3 projects are included in the attachment.
As you can see, L is certainly decreasing with diameter. However, the rate is not so high. I believe it should be in the LOG scale. For the coupled wire bonds, the self-L just decreases slightly while the mutual-L can be quite significant.
As you can, the 1 nH/mm rule is still approximately true.
-------------------cases and results----------------------------------
100 um GaAs substrate
75 um by 75 um Bondpads
Bondpads Center-to-Center Distance: 400 um
Wire bond height (from base): 98.5 um
Approximate Wire bond length: 500 um
Approxiate total length (including bondpads): 600 um
Wire bond to wire bond center to center distance: 100 um
One 17 um thick One 10 um thick Coupled 17 um thick
Hi, Kella: You might pick the wrong item. The evaluation license works for the Windows. Only the IE3D engine is working for Linux and it is a separated item. Please download ALL PRODUCTS FROM ZELAND for the Windows eval. Regards.
I got this info from the manufacturer of GaAs pHEMT Transistor :
You just have to use it with :
- Wu = 20 (unit finger width)
- N=6 (number of fingers).
Note that the model includes only intrinsic Ls, Lg and Ld inductors.
You should add in your design the bonding wire inductance (0.8nH/mm
for a single 25.4µm wire).
The model is validated to 60GHz and can be used for linear analysis
as well. You can thus extract S-parameter to 60GHz.
Hi, Keller: Yes. 0.8 nH / mm is a good value. I can see it matches the IE3D results always perfectly. Certainly, the L should be decreasing with frequency. Also, the mutual L may be critical if you use multiple wire bonds in the fingers. Regards.
The data is not measured one. I simulated in the IE3D Full-Wave EM Simulator and I attached all the results earlier. Here are the results with more (selected)frequency points. It is interesting to see the calculated L(1,1) value matches the specification from the manufacturer (provided by Keller in the previous e-mail). Certainly, the manufacturer of the transistor claimed the model of L = 0.8 nH/mm will work up to 60 GHz might be over estimate. For 500 um length, it is almost half a waveguide wavelength in a typical 100-um thick GaAs circuit. Lumped element model should not work. From what I see, the lumped element model should still work up to 15 GHz. Certainly, it will be much better to use the primary s-parameter results from full-wave EM simulators. Regards.
WB length = 500 um with 50 um extra length in each bond pad.
Coupled WBs are separated by 100 um from center to center.
The circuit should be modeled as distributed circuit over
about 15 GHz. Using lumped element may cause loss of accuracy.