G. Bolla Purdue University For the CDF Silicon Task Force Many thanks to: CDF collaborators SiDet (FNAL Silicon Detector facility) TD (FNAL Technical Division)

1
G. BollaPurdue UniversityFor the CDF Silicon
Task ForceMany thanks toCDF
collaboratorsSiDet (FNAL Silicon Detector
facility) TD (FNAL Technical Division)
Wire-bonds failures induced by resonant vibration
in the CDF Silicon Detector
2
Outline (really a time-line)New Failure Mode
well after commissioning
3
First observations and immediate reaction

• The failures were tracked down to anomalous
  trigger conditions during data taking.
• Trigger TORTURE tests (to explore dead-time at
  high trigger rates)
• Fake triggers generated according to the fixed
  beam structure. Synchronous!.
• High Dead time runs (due to problems in the
  experiment)
• SVX3 chip can go in a HIGH occupancy (100) mode
  if 5th L1A is issued (buffer is only 4 deep)
• Long readout time drives CDF into issuing a
  trigger as soon as there is a free buffer. This
  happens with a fixed time interval between each
  other. Synchronous!.
• Failures concentrated on an integrated time of a
  few hours to be compared with 2 Years of
  operations.
• This must be something that we can avoid!
• Introduction of a rate limit at L1 for the
  experiment
• Only reaction possible with the limited
  knowledge.
• Started the investigation

4
Symptoms of the failures

• DOIM failures
• No more data from the module.
• Only feedback is the current consumption on the
  power lines
• Module still responds to all the detectable
  inputs.
• Sudden drop on the Port-Card current consumption
• Loss of 1/5 of the power consumption
• Occurred on both ISL and SVXII
• DVDD Jumper connection
• No more useful data from the Z side of the module
• F side still operable and symptoms free
• Increase by 10-15 of the Analog power
  consumption (AVDD) and consequent trips.
• Extra AVDD current is strongly dependent on what
  the Digital part of the chip is doing
• Failures are confined to SVXII only
• Symptoms have been successfully reproduced on the
  bench (with spare modules) and are consistent
  with the loss of continuity on specific power
  line at well identified locations in the hardware

5
What are the power lines that fails?

• DOIM TX Power
• (Dense Optical Interface Module Transmitter)
• Vertex 2002 presentation by S. Hou
• 8 bits clock optical link that transmit the
  data from the tracker to the VME boards.
• Laser diodes are powered with a differential
  power supply
• Granularity of the failure indicates wire-bonds
  breaking/fusing.
• Other single point failures are not consistent
  with the symptoms (very specific)

SXV3 chip Digital power (DVDD) from F to Z side.

• Same connection as for everything else.
• All control signals
• All data lines
• Analog power as well
• 5 times higher current in steady state condition
• Never Less than 2 times higher

6
What do they look like
7
Multiple lines of investigations

• Power surges in time scales short enough not to
  trip the Power Supply and still capable to fuse
  the wirebonds.
• Extensive measurements of the energy needed to
  blow the wirebonds
• Extensive analysis of the energy stored locally
  in the detector
• Extensive analysis/measurements of the time
  response of the tripping circuitry of the power
  supplies
• This mechanism was ruled out. More than a factor
  of 5 safety margin
• Aging of the various components in the power
  lines.
• Focused on the vias on the small jumper
• Accelerated aging tests (high Temperature high
  current)
• No failures up to 20 years equivalent of data
  taking
• (we stopped after 4 weeks at 120 C with 1 A per
  VIA)
• Power surges on the complex ground of the system.
• Hard to replicate on the bench
• A deep review of the GND connections. Nothing
  really came out of it.
• Fatigue on the wire-bond heel induced by Lorentz
  Forces.
• Similar work done in the past for completely
  different reasons (IEEE Transaction on
  Components, Hybrid and Manufacturing Technology,
  Vol 14, NO 4, December 1991)

8
How is the current on those lines?

• Both have a behavior that is trigger dependent
• DOIM
• The output of the current amplifier feeds either
  the laser Diodes (bit HIGH) or a dummy load (bit
  LOW)
• The 2 loads are supposed to be perfectly matched.
• Dynamic measurements have been performed on a
  sample of DOIM TX packages left over.
• The DI distribution is centered at 0 current with
  tails up to 20 mA.
• DVDD Jumper connection
• The digital part of the SVX3d (BE) chips have a
  current consumption that is strongly time
  dependent.
• lt 10 mA/chip while waiting for a trigger
• Doubles up during digitization (we have an on
  chip Wilkinson ADC)
• Grows up to 120 mA while reading out.
• Capacitors on the lines smooth up the behavior
• The time spent in the high current mode is
  occupancy dependent (on chip sparsification logic)

9
Bonds and Lorentz forces

• With a 1.4 T Magnetic Field, 200 mA of current
  and a 2 millimeter bond in a plane that is
  orthogonal to the field the forces are in the
  range of 5E-4 N (50 mg).
• The induced movement should not exceed 3-4 mm
  (hard to measure on a 25 mm Al wire)
• Resonant behavior is expected and the natural
  resonant frequencies are in the KHZ range
  depending on the length and shape of the wires
  (CDF is interested only up to 50 KHz)

10
Detection of the resonances

• With regular video equipment the resonances were
  detected.
• The amplitude of the oscillations is consistent
  with a few wire diameters
• Test wire-bonds while oscillating break in time
  scales of minutes.
• At 10-20 KHz this implies 10E6 - 10E7 cycles.

11
What happens at the heel
Wire-bonds break due to fatigue stress on their
heel induced by resonant vibration. These
resonant vibrations are a direct consequence of
the oscillating Lorentz forces induced by the
magnetic field on wire-bonds with non-DC current.
12
A closer look with 40K fps camera

• Single frames have been digitized and a
  quantitative analysis has been performed.
• Bonds were excited with limited number of pulses
  to measure the amplitude of the motion versus the
  number of pulses and the dumping ratio.
• 3 pulses (at the right frequency) are enough to
  induce motion

13
Very narrow resonances

• 2 natural resonant frequencies up to 50 kHz
  (range of interest for CDF) for realistically
  shaped bonds.
• With current pulses each resonant frequency f can
  be excited with pulses at f, f/2, f/4 etc
• The resonant system has a very high Q.
• The width of the resonance is about 1-200 Hz.
• Differently shaped bonds imply different resonant
  frequencies.

14
F.E.A.

• The system is very simple and can easily be
  simulated
• The agreement with the experimental results is
  very good

15
Operational response

• AVOID trigger conditions that could resonate the
  bonds.
• Administrative and Run-Control software.
• Understood and removed all possible spurious
  sources of the 5th L1A
• TS (Trigger Supervisor) firmware and Command
  Signal strength changes
• Current swing minimization
• Reduce the power output
• of the bus drivers
• (programmable)
• Minimization of noise
• occupancy
• Trigger Inhibit on potential resonances
• Count time between readout commands (Dti)
• If (Dti -Dti1)lt1msec C
• If C gt 5 pull the brakes

16
Possible solution (obvious)
Small drops of encapsulant (Sylgard 186 Silicone
Elastomer from Dowcorning) limit the oscillation
amplitude by more than a factor of 30 by covering
just the first 50-100 mm of the wire. We were
not able to break these wire-bonds!

• The small amount of encapsulant was placed by
  hand
• By placing the encapsulant only at the foot the
  problematic associated with not perfectly matched
  CTEs should be minimized
• No effort on our side toward any large scale
  technique (too late for CDF)

17
Conclusions

• Last fall the CDF experiment faced a crisis due
  to internal unrecoverable failures on the silicon
  detector.
• The source of the failures has been understood to
  be simple physics mechanisms that could have been
  taken into account during the design and
  construction of the hardware.
• Counter measures have been studied, developed and
  applied to the CDF experiment. Since these
  implementations are in place, no other failures
  have occurred.
• Similar problems could be faced by other
  applications.