MRB on SLAC NCR 226 Excess of Disconnected MCM Channels

1
MRB on SLAC NCR 226Excess of Disconnected MCM
Channels

• R.P. Johnson
• November 9, 2004

2
1. Description of Non-Conformance

• MCM part number LAT-DS-00898, SN 11378 was found
  to have 151 channels disconnected between the
  amplifiers and the pitch adapter.
• This MCM had zero non-conductive cracked traces
  (from the visual inspection) and zero dead
  amplifiers.
• This problem was found Friday, November 5, at
  SLAC while testing 42 MCMs using the new
  electronic pitch-adapter test fixture.
• Earlier the same week 1 MCM already attached to a
  tray and SSDs at GA was found to have about 300
  disconnected channels.
• Also, at GA and Pisa 2 trays were found to have
  a disconnection of 1 or more bias lines, which
  may be from the same root cause.

3
Inspections and Tests

• A new fixture designed at SLAC solves our old
  problem of not being able to test the connections
  to the amplifier inputs until SSDs are attached.
  This became a critical problem when the MCM
  mounting changed from screws to gluing.
• Initially our assurance was based on
• Wire bond quality control at Teledyne
• Visual inspection of all wire bonds and bond feet
  at MIP-2 at Teledyne
• Sample testing of several preproduction MCMs by
  probing each pitch-adapter trace.
• After an MCM on Tower-0 was found with 25 cracked
  traces, we did the following
• Visual inspection for cracks in pitch-adapter
  (and manual probing with a DVM for continuity)
  traces before shipment to Italy. This appears to
  be quite effective, albeit with a rejection rate
  of 14, requiring no more than 8 such cracked,
  nonconductive traces.
• Tested with 200 thermal cycles whether the
  incidence of non-conductive cracks gets worse (it
  does not).
• Stepped up existing efforts to get an electrical
  test method developed.

4
The Electrical Test Method

• Use a conductive, flexible strip (zebra strip) to
  short circuit 100 of the pitch-adapter traces to
  ground.
• Bias HV is turned off, to protect it.
• AVDDA must be turned on, but the resistance of
  the zebra strip is high enough that it does not
  draw enough current to cause any problem.
• Run a charge-injection scan and measure the
  response of each channel.
• Channels with inputs shorted to ground will look
  dead.
• Channels not connected to the pitch adapter will
  look alive, unless the amplifier is already dead
  for some other reason.
• So, if a channel measures good in the normal
  electrical test, with the input floating, AND it
  measures dead with the zebra strip applied, then
  we are certain that it is a good, connected
  channel.
• If a channel measures good with the zebra strip
  applied, then either
• it is disconnected from the pitch adapter, or
• the zebra strip did not make contact with the
  pitch adapter (false negative).

5
Pitch-Adapter Electrical Test Fixture
Designed by Peter Amnuaypayoat
STORAGE CASE BASE
ZEBRA CONNECTOR
SOFTWARE/ ELECTRICAL FIXTURE
MCM WITH PITCH ADAPTER
TEST FIXTURE
CONNECTOR SAVER
GROUND LEAD
6
Disconnected Ch. vs. Cracked Traces
Visual inspection and test data from 41 MCMs.
Note that the number of disconnected channels is
probably a little bit overestimated, due to
occasional false negatives given by the test
fixture. (False positives are not possible,
unless the connection is intermittent.)
7
Disconnected Ch. vs. Cracked Traces
Visual inspection and test data from 41 MCMs.
The 42nd MCM had 151 disconnected and 0 cracked
traces.
8
Bias Disconnects

• The pitch-adapter electrical test fixture will
  not detect problems with the bias traces.
• Our visual inspection rejected MCMs with even 1
  cracked nonconductive bias trace, but this does
  not catch all problems.
• Up to now we relied on
• 100 electrical test of the PWB.
• Visual inspection of the bias resistor.
• Wire-bond QC and visual inspection at MIP-2 at
  Teledyne.
• Still, there are 2 cases of MCMs in Italy with
  bad (sometimes intermittent) bias connections.
• Therefore, we are now doing the following
  procedure on all MCMs before shipping to Italy
• Turn the bias voltage down to 2.5 V, and turn on
  the MCM power.
• Using a voltmeter, probe each of the 16 bias
  traces past the bend region, and verify that the
  appropriate voltage (2.5 V or 1.5 V) is seen.

9
2. Overstress Analysis Additional Testing

• We tried out the test fixture on the 4 nonflight
  MCMs that we recently passed through 200 thermal
  cycles from ?30C to 85C.
• 3 were not bad, with the worse case being 15
  disconnected channels.
• 4th had 60 to 80 disconnected traces, depending
  on the test trial. The results were variable,
  and we found clear cases of false positives. The
  pressure from the fixture was causing
  reconnections of some of the failed traces.
• This experiment would be more interesting if we
  had a measurement before the thermal cycles as
  well as after.
• A more relevant test will be to check the
  connections at MIP-3 (i.e. at Teledyne) and then
  again after thermal cycles and burn-in at SLAC.

10
3. Suspected Root Cause

• We suspect failure of the wire bonds where they
  connect to the pitch adapter. They are not
  visible (black encapsulant), but we have indirect
  evidence
• Cracked traces in the visible region of the pitch
  adapters are eliminated by our thorough visual
  scans.
• Our visual inspection at MIP-2 ensures that the
  wire bonds were there and were initially
  connected (we couldnt fail to see 151
  disconnected or missing bonds!).
• The MIP-2 visual inspection also rules out that
  any significant number of the disconnected
  channels could be due to pitch-adapter cracks
  close to the cut edge.
• The very low failure rate of wire bonds on the
  power and digital I/O GTFE connections nearly
  rules out wire-bond failure on the ASIC side of
  the bond.
• The pitch-adapter substrate is the more difficult
  material to wire bond to.
• So, we believe that either the bonds failed
  during encapsulation or later during thermal
  cycling. The intermittent ones must have failed
  after encapsulation in order to have the foot
  still so close to the pitch-adapter trace.

11
Impacts to Inventory

• The 1-in-42 (2.4) occurrence of large numbers of
  disconnected bonds is not a big problem for MCM
  availability.
• The 14 loss due to gt8 cracked traces is a big
  problem. At that rate we have to review again
  whether we will have enough parts to complete the
  build, given that this is not the only source of
  loss of MCMs (e.g. Tower-0). I suggest that we
  consider raising the maximum number of bad
  channels from 8 to 15. This is the limit used at
  the tray level and corresponds to a worst-case
  inefficiency of 1.

12
5. Corrective Action

• Test all MCMs with the PA test fixture both at
  MIP-3 and before shipping to Italy. We need to
  manufacture one more fixture.
• Review and possibly improve the Teledyne
  wire-bonding controls. The present procedure
  calls for the following
• Plasma clean in Tepla (Recipe 1) prior to wire
  bonding.
• At the start of each shift, wire bond the test
  pattern on a single die on the wire bond set up
  PCB. Perform 100 destructive pull test. During
  actual wire bonding, perform Non destructive Pull
  Testing (NDPT) on every 3rd unit. NDPT test on 2
  die, 30 wires on a single PCB only.
• Update and re-release the burn-in and test
  procedure LAT-TD-02367.
• Verification plan
• Test our remaining MCMs with the PA fixture asap.
• Consider doing extra thermal cycles on some
  non-flight MCMs, with tests before and after.

13
6. Effectiveness of Corrective Action

• The PA test fixture is almost guaranteed to find
  the missing connections, the only exception being
  cases where the pressure from the fixture results
  in a temporary reconnection. This will not
  happen on all or even most of any large number of
  missing connections on a given MCM.
• There is some concern that the tray-level thermal
  cycles could produce more disconnections.
• The burn-in procedure thermal cycles are more
  severe in terms of number of cycles (20 vs. 4)
  and range (?30C to 85C vs. ?30C to 55C).
• This would be the reason to try some more
  extensive thermal cycle testing on non-flight
  MCMs.

14
7. Recommended Disposition

• Scrap (use for EGSE) the MCM that is the subject
  of NCR 226.
• Return all MCMs to SLAC that have not been tested
  with the PA fixture and bias-circuit hand
  probing.
• Begin mounting to trays those MCMs that passed
  the PA and bias probing tests.