Liverpool BeO Forward Hybrid with ABCD2-P
Introduction
A Liverpool BeO Forward Hybrid has been populated with 6 x ABCD2-P (thinned)
to one side only. The chips are mounted onto individual Aluminium foil
substrates using conductive glue. Wire bond connections, of imporatance,
are:
-
Aluminium foil substrate bonded to AGND only.
-
Ring-A pads on ABCD2 bonded to Ring-A pads (this is not AGND) on Forward
Hybrid.
The ABCD2 has Edge Detect enabled and is using Edge Data Compression ,
unless stated.
40MHz clock mark-space ratio set to 1:1.
Nominal values of VCC = 3.5V and VDD = 4.0V (measured on
hybrid).
VCC and VDD are floating, Bias Card BC96 is not used.
Single trigger data collection.
Chip Location and ID
Chip Location |
1
|
2
|
3
|
4
|
5
|
6
|
Wafer ID |
P3-14
|
P3-15
|
P3-00
|
P3-07
|
P2-07
|
P4-15
|
* ABCD2 with no detector
* ABCD2 with detector
*ABCD2 oscillation fix
ABCD2 performance with no detector
Digital tests indicate that all 6 (1 to 6) chips work:
-
All chips interpret commands correctly.
-
Clock mark-space ratio is not critical.
-
Fix still required for 'last chip' trailer problem.
Analogue testing reveals:
-
Only 5 chips are actually working. Chip 4 produces no analogue output (Internal
calibration failure?).
-
F.E.Bias has broad operating range, 230uA - 285.2uA with no degradation
in overall performance.
Optimum settings are tabulated below:
Chip
|
F.E.Bias (uA)
|
Shaper Bias (uA)
|
Strobe Delay
|
1
|
239.2
|
18
|
25
|
2
|
239.2
|
18
|
25
|
3
|
239.2
|
19.2
|
25
|
4
|
n/a
|
n/a
|
n/a
|
5
|
239.2
|
19.2
|
21
|
6
|
239.2
|
19.2
|
23
|
Results
Using the settings, shown above, for F.E.Bias and Shaper Bias, the following
results have been obtained for Input noise (fC), Threshold (mV) and Gain
(mV/fC) vs Channel for nominal Qinj of 2.0fC and 3.0fC.
Observations
-
Gain is typically 150mV/fC at 2.0fC.
-
Input noise is typically 560e at 2.0fC.
-
Chips 1, 2 and 3 exhibit similar characteristics whereby oscillation can
clearly be seen for thresholds of less than 125mV (typical).
-
Chips 5 and 6 show no oscillation, even with threshold set to 0V. See
Scurve plots above.
ABCD2 performance with Wedge
detector
Conditions
-
Wedge detector W12 (72mm), Strip pitch 70um (outside) to 54um (inside).
-
Single chip (Chip 6) bonded to detector.
-
F.E.Bias still has broad operating range, 230uA - 285.2uA with no degradation
in overall performance.
-
Strobe delay has changed to 31. There appears to be increased sensitivity
to this setting, moving +/- 2 counts from this setting reduces chip performance.
-
No modifications to Melbourne support Card or Forward Hybrid detector bias
network.
Results
Using nominal settings for the F.E.Bias and Shaper Bias, 239.2uA.and 20.4uA
respectively, plots are shown for Input noise (fC), Threshold (mV)
and Gain (mV/fC) vs Channel for nominal Qinj of 2.0fC and 3.0fC.
Observations
-
Gain is 150mV/fC at 2.0fC.
-
Input noise is typically 1375e at 2.0fC.
-
Still no observed oscillation!
ABCD2 Oscillation at Low
Threshold
Of the 5 'working' chips only 2 work correctly in that no oscillation is
observed at low threshold. The cause of this oscillation has never really
been properly understood and thus the ability of being able to resolve
has proved problematic.
Why 2 chips work, which both come from different wafers, compared to
the remaining 3 chips (from the same wafer) which do oscillate initially
suggested that the problem could be processing related. Recent correspondence
has indicated that this is not the case, it appears that the main contributary
factor is the electrical interface between the backplane of the chip and
the foil substrate that the chip is glued to, this should be a low Ohmic
connection. The integrity of this connection is determinded by the conductive
epoxy and its application.
-
Reworking of the chip backplane and foil substrate connection i.e. additional
conductive epoxy has been added (where possible). This has resulted in
a dramatic improvement in chip performance!
-
This rework has been done to the 3 oscillating chips. Chips 2 & 3 no
longer oscillate, chip 1 has a different Scurve characteristic.......but
this is probably due to the fact that not all the VCC and AGND bond connections
are made on the chip.
Results
Using the nominal settings for F.E.Bias and Shaper Bias, the following
Scurves have been obtained for nominal Qinj of 2.0fC:
Chips 1, 2 & 3 are not bonded to the detector. Chips 5 & 6 are
bonded to the detector.
Again using the nominal settings for biasing, not yet optimised, the following
results have been obtained for Input noise (fC), Threshold (mV) and Gain
(mV/fC) vs Channel for nominal Qinj of 2.0fC.
Chips 1, 2 & 3 are not bonded to the detector. Chips 5 & 6 are
bonded to the detector.
Return to homepage
Last modified: 03/02/99