Agenda for SCT links meeting
The meeting was held on May 21, 1997 at 9:00 in room 40-SS-C01.
09:00 Introduction (R. Kowalewski)
09:10 Expected fluences in the SCT (A. Weidberg)
09:20 Overview of LED-PIN system , prototype results (A. Weidberg)
09:40 Summary of LED irradiation studies (J. Beringer)
10:10 Summary of PIN irradiation studies (S. Oglesby)
10:25 First test results on LDC and DORIC chips (D. White)
10:40 Coffee
11:10 Overview of shielded twisted pair system (R. Kowalewski)
11:10 Summary of performance (R. Kowalewski)
11:25 System optimization (R. Kowalewski)
11:35 Material budget (S. Marconi)
11:50 Cost of LED-PIN and STP links (R. Kowalewski)
12:00 Experience with VCSELs
o radiation hardness, reliability, coupling to fiber (Bern)
o experience with VCSELs for the LAr optical links (O. Martin)
o packing efforts (Birmingham)
12:30 Pixel data transmission requirements (M. Holder)
12:40 Lunch
14:00 Implications for other SCT subsystems
o FE electronics and modules (A. Grillo)
o mechanics and cooling (R. Kowalewski)
14:15 Discussion led by R. Kowalewski covering
o Performance
o Radiation hardness
o Material budget
o Cost
o Applicability to Pixels
o Further study of VCSELs
15:30 Coffee
15:50 More discussion, decisions
17:00 End
*********************************************************************
The aim of the meeting was to choose between the two proposed link
solutions (LED-PIN optical links or shielded twisted pair links).
Presentations of work relevant to each system were given during the
morning session. The afternoon was used to discuss the advantages and
disadvantages of the two systems and make a recommendation.
* * *
Tony Weidberg showed the expected fluence in the SCT for 10 years of
LHC operation. The highest calculated fluence in the SCT is
1.3*10^14 n/cm^2, but this fluence has a 50% uncertainty, giving
1.9*10^14 n/cm^2 as the expected fluence upper limit. Alternatively
one can require that the detectors and FE electronics die before the
links. The fluence at which these latter devices cease to work well
enough to be usable (S/N < 8) is not precisely known, but given
current information appears to be below 3*10^14 p/cm^2, which
corresponds to 1.8*10^14 n/cm^2, about the same as the expected fluence
upper limit.
Tony presented a table of NIEL equivalents in silicon (relevant to PIN
diodes) and GaAlAs (relevant for LEDs) to translate this fluence into
equivalent doses of 24 GeV protons and 1 MeV neutrons to facilitate
comparison with irradiation studies at the PS, ISIS, and the
Birmingham Dynamitron.
The fluence target given above (in 1 MeV neutron equivalents) was
1.8*10^14 n/cm^2. The radiation hardness of the PIN diodes was judged
against this number. For the LEDs the equivalent fluence of 24 GeV
protons was calculated to be 1.4*10^14 p/cm^2.
His transparencies are available at
http://wwwcn1.cern.ch/~kowalews/public/weidberg-fluence.ps.
* * *
Tony then gave an overview of the LED-PIN system. As this is well
documented in the TDR and in a draft NIM article, it will not be
described in these minutes. See links from web page
http://wwwcn1.cern.ch/~kowalews/public/linksreview.html.
* * *
Juerg Beringer then gave a summary of the LED irradiation and lifetime
studies carried out at Bern. Most of the information he presented is
very well documented in their draft paper, available at
http://www.lhep.unibe.ch/atlas/ledlife/welcome.html.
New preliminary results were presented on the re-irradiation of LEDs.
A batch of LEDs that had been previously irradiated to fluences of
about 2*10^14 p/cm^2 at the PS were annealed and reirradiated. The
reirradiation of ABB LEDs was done at the Dynamitron to 3*10^14 n/cm^2,
giving a total fluence of about 3*10^14 p/cm^2. The reirradiation of
the GEC devices was done at ISIS to 5*10^14 n/cm^2, giving a total
fluence of approximately 5*10^14 p/cm^2. The reirradiated devices were
once again annealed. After annealing they appear to have a better
relative light output (RLO) than devices given an equivalent dose in
one irradiation. This may suggest that the radiation damage at a
given fluence is less severe at lower dose rates.
The conclusion of Juerg's talk was that the NIEL scaling hypothesis
is confirmed by their damage constant fits, and that GEC LEDs annealed
with a current of 50 mA give sufficient light after irradiation to
maintain a S/N of about 65, making them adequate for use in the SCT.
He cautioned that 3/58 GEC LEDs gave less than 10% of the initial light
output and were considered as "failing"; he added, however, that only
one of these devices was a hard failure and that the other two might
be recoverable in ATLAS.
* * *
Stephen Oglesby presented recent results on PIN irradiation studies - see
http://wwwcn.cern.ch/~oglesby/pin_results/pin_results.html
The PINs were irradiated at the Dynamitron in 3 periods up to a total
fluences of 1.2-1.8*10^14 n/cm^2 (multiply by 1.97/1.80 to get 1 MeV
neutron equivalent). The new irradiation studies confirm the results
obtained earlier with bulk Si PIN diodes, namely that their responsivity
plummets to near zero after doses below 10^14 n/cm^2. The "new style"
epitaxial PIN diodes, however, performed very well. The responsivity
dropped from 0.35A/W to 0.25A/W after the first irradiation period but
remained more or less constant during and after the next two irradiation
periods. The dark current of the epitaxial PINs went up with each dose
but remained below 25 nA while the bulk PIN dark current increased to
over 600 nA. The damage seen on the bulk PINs is unexpected and not
understood. Further irradiation studies of epitaxial PINs under proton
irradiation are planned.
* * *
David White was clearly pleased to report that the first tests performed on
the DORIC and LDC chips look very good. LDC produces the desired tunable
drive current. DORIC has not been fully tested, but tests performed with
20 and 40 MHz square wave signals gave the proper clock doubling and
extracted data. David pointed out that due to the lack of a phase lock
loop the mark-space ratio of the extracted clock was dependent on the
mark-space ratio of the input signal (sensitive to both rising and falling
edges). The input signal thus needs to be adjusted to keep the extracted
clock mark-space ratio better than 1 ns.
The chips were fabricated in the AMS 0.8 micron bi-cmos process (not
qualified rad-hard). Irradiation studies with neutrons will be performed
at ISIS in June.
* * *
Bob Kowalewski reviewed the use of shielded twisted pair (stp) cables for
SCT data transmission. The work done by Peter Lichard and Arno Beer of the
CERN TRT group made it possible to evaluate the use of stp cables in the
SCT in a rather short time and is gratefully acknowledged.
The contents of his talk are drawn from the draft INDET note available at
http://wwwcn1.cern.ch/~kowalews/public/stpsystem.ps.
He concluded that there is ample bandwidth available for 40 Mb/s data
transmission on AWG36 cables, and that further miniaturisation of the
cables was called for. The noise studies indicated that noise radiated
from the pairs is not a problem, but that noise picked up along the shield
must be considered. The use of AWG38 cables would reduce the material
budget to approximately 60% of the AWG36 thickness.
The 80m link from the patch panel to the ROD was not satisfactory with the
AWG28 cables used so far. The TRT community is working to improve on this.
* * *
Stefan Marconi then presented the material budget for the stp cables and
the optical links. The plots can be found in the draft INDET note referred
to above. The material budget for 4 AWG36 cables per module has a peak at
eta=2.2 of X/X0=2.5% in front of the last TRT hit and 1.0% in front of the
last SCT hit. The corresponding values averaged over eta<2.5 are about 1.4%
and 0.6%, respectively. Use of AWG38 cables and multiplexing of clock and
control signals could reduce these values by 0.6*3/4=0.45.
The material budget for optical fiber is a strong function of the yet to be
defined fiber protection. With no protection the material up to the last
TRT hit averaged over eta is about 0.2%, while with individual fiber sheaths
(o.d. 1mm) this increases to about 0.5%.
* * *
Bob Kowalewski then presented a costing of the two options. The prices for
the LED-PIN system have large uncertainties, and might well be underestimated.
The optical system also should have the cost of the off-detector driver and
receiver added to it. The cost breakdown can be seen at
http://wwwcn1.cern.ch/~kowalews/public/cost.html
The stp links would cost about 143 CHF/module. The LED-PIN optical links
would cost at least an additional 100 CHF/module.
* * *
The next section of the meeting was devoted to giving the status of the use
of VCSELs in radiation hard optical links. Juerg Beringer showed the results
from irradiation and lifetime studies. You are once again referred to
http://www.lhep.unibe.ch/atlas/ledlife/welcome.html
for details. The VCSELs obtained from Sandia National Laboratories in the
U.S. are very radiation hard and perform well in lifetime tests. The main
problem cited by Juerg was VCSEL-fiber coupling, since different modes are
excited as a function of drive current and failure to couple all the light
into the fiber results in non-linear and unstable transmission. In addition
he noted that back reflection of light into the VCSEL caused very noisy
operation. Reasonable results were obtained on a test bench by butt-coupling
the fibers.
Olivier Martin of the LAr community gave a summary of their experience with
Sandia VCSELs. They are working on an analogue link from the calorimeter
back end to the RODs. He also cited VCSEL-fiber coupling as the main issue,
and showed drawings of a proposed package from Amphenol. That package is
probably too heavy for use in the SCT, but it will be interesting to see if
the coupling problem is solved. Olivier cited cases where back reflected
light was thought to be responsible for killing the laser diodes that were
butt-coupled to fibers. Angle cut fibers would be one way to avoid this.
John Wilson then discussed the status of the GEC packaging of Sandia VCSELs.
A meeting with Kevin Lear has taken place at GEC and both parties appear to
be interested in pursuing the packaging effort. GEC will package VCSEL arrays
and cut off 2 VCSELs for use in the existing LED-PIN package. Devices should
be available this summer.
Robert McLaren pointed out that VCSEL packaging has been successfully solved
for commercial links. Juerg Beringer responded that the packaging techniques
used there might prove too expensive for our needs (i.e. a large number of low
bandwidth links).
* * *
After lunch Martin Holder discussed briefly the special needs of the pixel
detector for their data transmission. He noted that the pixel community
hoped that the solution chosen for the SCT could be used for pixels as well.
They will follow with interest the development of the SCT links. A very
quick calculation was done of the number of links needed for pixel readout.
Using AWG36 cables at 40Mb/s would lead to an unacceptably large contribution
to the material budget (6-8%). Use of AWG38 cables at 80Mb/s would still
give a contribution of 2-2.5%. The use of optical fibers, on the other hand,
is made more difficult due to the increased fluence.
* * *
Alex Grillo summarized the impact of the two link choices on the FE
electronics and hybrids (input from Carl Haber, Tony Smith, and Alex).
There was a concern about impedance matching when driving a bus on the
hybrid with the LVDS signals from STP cables. Some on-hybrid repeater or
a DORIC-like chip might be needed in the stp system. It was noted that
redundancy of the clock and control links would not be needed if stp
was used. This would allow the elimination of the module interconnect
if one also dispensed with the no-beam clock synchronisation scheme.
Noise concerns regarding possible pickup from stp cables and the possible
injection of noise onto power supply lines by the optoelectronic elements
were expressed.
The major concern expressed was potential problems due to ground loops via
cable shields.
* * *
Bob Kowalewski reported a conversation he had with Eric Perrin concerning
issues for the mechanics and cooling. Eric didn't see major differential
advantages or disadvantages between the two link systems. His main concern
was regarding installation and access, since no connector was forseen for
the links between the FE and the muon crack patch panel in the TDR. He
feels strongly that a connector closer to the barrel is needed.
He cited the cooling contact for the optoboard as a minor complication, and
similarly thought the additional space required for stp cables could be
accommodated.
* * *
The meeting then turned to a discussion of the relative merits of the two
systems. A table was constructed with cryptic comments on different aspects
of the comparison. The resulting table is reproduced here:
LED-PIN link STP link
Data transmission OK OK
performance (should multiplex be used?) (can multiplex be used?)
Radiation LEDs ok to 2*10^14 p/cm^2 OK
tolerance PINs ok to 1.8*10^14 n/cm^2,
further tests needed
Material budget 1mm sheath: 0.5% (0.2%) 4 AWG36: 1.4% (0.6%)
Avg. X/X0 before bare fiber: 0.2% (0.1%) 3 AWG38: 0.6% (0.3%)
TRT (SCT)
Cost/module in CHF >244 143
Noise induced Noise injection on power Noise pickup on shields
into SCT FE lines via LDC or DORIC? Ground loops via shields
Reliability device failure; redundancy Noise problems may continue
needed to increase with system size
Size 1mm sheath: 2.36 4 AWG36: 2.54
(area in mm^2) bare fiber: 0.15 3 AWG38: 1.51
Power LED driver 10(20)mA => No additional power
96(176)mW (4-5% of (just LVDS outputs in
module power) common with fiber solution)
Use in pixels Radiation hardness of Material budget with existing
LEDs and PINs needs to cable routing as large as
fluences. VCSELs would X/X0=8%; optimized
be interesting if feasible. mass/bandwidth solution 2.5%
* * *
In the ensuing discussion substantial weight was given to potential system
noise problems if stp cables were used although no one claimed such a system
couldn't be made to work. The increase in material budget was also stressed
as a major concern (notably by Andy Parker). It was clear that the optical
solution would cost considerably more, but this was seen as the only major
drawback, since the assembled agreed that the GEC LEDs were sufficiently
radiation tolerant. There was a consensus that the optical link be chosen
for the SCT and that work on VCSELs (notably packaging) continue as an upgrade
path especially in light of their potential use for pixel readout.
* * *
A links meeting will be organized within the next few weeks to discuss the
work plan for 1997. A new coordinator will also have to be found as Bob
Kowalewski will be leaving the SCT in July.
*********************************************************************
These minutes can be found from web page
http://www.cern.ch/Atlas/GROUPS/INNER_DETECTOR/sctnew/Electronics/links/