Resonant Wire Bond Vibrations Tom Barber, Dave Charlton, Bill Murray

1
Resonant Wire Bond Vibrations Tom Barber, Dave
Charlton, Bill Murray T.W.

• The CDF story.
• Fourier analysis.
• Quick tests at Birmingham.
• FEA calculations.
• Tests of bond wires from real barrel and EC
  opto-packages and K5.
• Possible Trigger veto algorithms.

2
CDF Story

• CDF had fully functional SVX II at start of run
  but started to lose modules during run.
• All deaths of modules occurred during fixed
  frequency running (trigger torture tests and
  problem with SVX II architecture at high rates).
• All caused by resonant wire bond vibrations. No
  further deaths since fixed frequency running has
  been avoided.
• Will implement a simple trigger veto on fixed
  frequencies.

3
CDF Tests

• Large changes in currents during readout.
• Test bonds in magnetic field showed resonances
  and bonds broke quickly once resonance was hit.
• Video clip on web
• http//www-cdf.fnal.gov/upgrades/silicon/TASK-Forc
  e/line3/wtext.mpg

4
Bond Wire Breaks

• Bond wires break after seconds to minutes on
  resonance (CDF and ATLAS Pixels. Also seen with
  bonds on barrel SCT test hybrid).
• Deliberately badly made bonds break immediately.

Fatigue Failure
Pull Failure
5
Fourier Analysis

• Frequency distribution of genuine random triggers
  very broad ? very little power within a narrow
  resonance. Explains why CDF have seen no deaths
  since they avoided fixed frequency operation.
• Consider a resonance with
• f15 kHz
• G500 Hz
• L1 trigger rate of 75 kHz
• fraction of the power within the resonance 8
  10-3.
• Explains why CDF dont see any wire bonds
  breaking with random triggers.

6
First Tests at Birmingham

• Used Birmingham superconducting magnetic field
  facility which can go up to 8T.
• Used video camera to look at wire bonds to search
  for resonances. We had to run at B0.5T because
  fringe field caused loss of camera resolution but
  we increased the current by a factor of 4 to
  compensate.
• Used wire bonds of the correct length for the
  barrel opto-package VDC ? VCSELs.
• Found strong resonances around 15 kHz.
• Wire bonds broke after few minutes on resonance.
• Not just a CDF problem !

7
Calculations(Tom Barber )

• Analytic calculations give scaling laws for
  resonant frequency f and amplitude a for length l
  
• Note the wire bond is not under tension ? not
  like a violin string !
• Confirmed by FEA calculations.

8
FEA Calculations

• FEA for barrel opto VDC ? VCSEL wire bonds.
• Resonances for B field in plane of loop much more
  dangerous than if B field is perpendicular to
  plane of wire bond loop.
• Amplitude smaller by factor of 240 for B field
  perpendicular to plane of bond wire loop cf in
  plane.

9
ATLAS Tests

• Bonds tested

10
Resonances

• With dogleg orientated so that B field is in
  plane of bond wire loop (not the ATLAS
  configuration) strong resonances seen with video
  camera around 17 kHz (see transparencies).
• B1.8 T, I(VCSEL) 10 mA (2nominal)

11
F16 kHZ
12
f16900
13
f16950
14
f17 kHz
15
F17050
16
F17100
17
f17150
18
f17200
19
F17250
20
F17300
21
F17350
22
F17400
23
F17450
24
F17500
25
F17550
26
F17600
27
Back emf

• If bond wires can be directly connected to the
  pulse generator (not the case for barrel dogleg)
  can see back emf from vibrations.
• Pulse the wire with a step at low frequency and
  observe the voltage at the bond wire on the
  scope.
• FFT gives peaks at resonance frequency.

Test bond wire 3 mm. B in plane of bond wire loop.
Resonance 18 kHz
28
Results with ATLAS tests

• Barrel dogleg VDC?VCSEL bonds with correct
  orientation of B field.
• No resonances seen with camera in scan of 10-100
  kHz.
• Amplitude ltlt 25 mm.
• K5 VDC? hybrid bonds.
• No resonances seen with camera in scan.
• Amplitude ltlt 25 mm.
• Forward opto PCB ?VCSEL.
• No resonances seen with camera in scan 50-110
  kHz.
• No resonances seen with back emf.
• Amplitude ltlt 25 mm.

29
What Do We Do ?

• We didnt see any resonances for the bonds we
  will actually use ? do nothing.
• Use precautionary principle, two options
• Pot foot of wire bonds to dampen vibrations. This
  works but what are the long term risks ? How
  would we apply this in production. What would we
  do with already assembled parts?
• Avoid fixed frequency triggers (demonstrated in
  CDF). No cost or extra risk involved.

30
Trigger Veto Algorithms

• Aim is to veto on fixed frequencies but to keep
  genuine random triggers.
• Two algorithms simulated
• CDF
• FFT

31
CDF Algorithm

• NNumber of triggers with time difference between
  successive triggers equal within DT.
• Plot veto probability for random triggers as a
  function of N for different DT.
• N10, DT1 ms, L175 kHz,
• veto rate 2.7 10-5 Hz

32
FFT

• Most powerful algorithm because would detect
  clocks superimposed on random triggers.
• Simulate worst case
• Random triggers L175 kHz
• Random triggers L175 kHz plus clock f11 kHz.
• Perform FFT (over 10 ms) and take maximum
  amplitude component in range 10-100 kHz.

33
Calibration Runs

• Triggers generated by software so we can do what
  we want
• Dither frequency by 10G decreases power by
  factor 11. G 500 Hz ? Df 5 kHz.
• Only effect is to slow down calibration runs by
  3 to 40.