LArg FE-Links Meeting

Radiation levels and neutron energy spectra in ATLAS : consequences for transient errors
observed during irradiation of Gb/s optical links
    In preparation for the LArg electronics meeting in the afternoon, Marie-Laure first reported on the neutron irradiation tests performed on Gb/s demonstrator links based around the G-Link chipset from Hewlett-Packard. The results of the tests are fully described in papers submitted to conferences in summer 98 and can be seen here. In summary, tests at SARA with an mean neutron energy of 6.6 MeV gave around 1.5 errors per minute per link where the G-Link receiver cannot identify a frame to lock onto and the link is blind for several 100 mS. The tests at SARA were at a neutron flux approximately 500-1000 times larger than expected in ATLAS. The error rate drops rapidly as the neutron flux is decreased. The SARA high dose rate tests have been complemented by more ATLAS-like dose rate tests at a neutron facility in Göteborg, Sweden. In this test, three types of neutron source were used :

• D-D reaction (2.5 MeV neutrons) - no errors seen (even at SARA-like dose rates)
• D-T reaction (14 MeV neutrons) - errors seen
• AmBe source (spectrum from 1-12 MeV) - no errors seen for 14 hours, implying a threshold around 8 MeV at 90 % C.L.

Marie-Laure then went onto explain that it was crucial to fully understand the neutron spectra inside ATLAS at the position of the LArg electronics crates in order to assess the impact of such threshold behaviour. As already reported at a FE-Links working group meeting Paola Sala (Fluka) and Mike Shupe (Gcalor) have been looking into this problem. Their work is described
on the WWW. The two simulations agree to within a factor of five for the total fluence (4.6e13 [Fluka] -vs- 10e13 [Gcalor]) at the electronics boxes. Gcalor is thought to give a more realistic description of the ATLAS detector.

Finally, Marie-Laure discussed preliminary work aimed at comparing error rate results from SARA and Chalmers with the results from Fluka and Gcalor. 

Plans for the CERI Neutron Source in 1999
     Marie-Laure concluded by confirming that the links group have access to the CERI facility during 17th and 18th Feb. The idea of the test would be to define the energy thresholds for the links more clearly. She also mentioned other facilities for possible high energy neutron irradiations (60 MeV in Belgium), protons (CERI) and gammas (Stockholm).

Report from the Oxford Workshop
    Mark reported briefly on Second Workshop on Optical Read-out Technologies held in Oxford during 7th/8th January. He mentioned that a full set of transparencies would soon be available in the ATLAS secretariat and the organisers would try to get as many of the talks as possible onto the workshop WWW page. In this presentation, only items likely to be of interest to the LArg links group are described :

• MITEL SEMICONDUCTOR.  Mitel presented data detailing 7500 hour accelerated aging tests of VCSEL's. The data predicts < 1.5 failures in 10e9 hours at room temperature. At 100 C, the output power, threshold current and slope efficient of the VCSEL's remained constant through-out the 7500 hour test. Mitel also presented details of VCSEL and PIN arrays (4-way) built on lead-frames. The devices interface to standard MT ferrules and are being evaluated by the KTH and Oxford groups.
• HONEYWELL. Honeywell presented details of developments on Oxide Confined Aperture VCSEL's (as opposed to the more normal proton implanted devices). These VCSEL's have sub-mA threshold currents and could be driven directly from a G-Link transmitter chip, without the need for a laser driver. Preliminary data indicated that the reliability of the oxide aperture VCSEL's was compatible with current proton implanted Honeywell VCSEL's (MTTF > 10e7 hours at 40 C and 10 mA bias). Unfortunately these devices are currently in a developmental stage and Honeywell have no plans to commercialise them in the near future.
• HEWLETT PACKARD. HP confirmed that Gb Ethernet was the dominant standard for Gb/s links. They went on to say that the IEEE 802.3z standard had recently been approved for Gb Ethernet (V5.0 - June 98). At 850 nm, the Gb Ethernet standard requires a 62.5mm core graded index fibre to be able support traffic for a distance of up to 250 m (500 m for a 50mm core fibre). The low power G-Link chipset was also described in some detail (available Q199). Two Tx chips would be needed per FEB as the double frame functionality is not implemented. The chips run at 3.3 V and dissipate 1.4 W (as opposed to 4.4W@5.0V). The chipset is believed to use the same bipolar process as the standard G-Link.
• MOLEX. Molex presented many products, including :
• SC adapters and transceivers with in-built shutters to exclude dust and improve eye safety. These could be interested for use at patch panels to simplify eye safety conformity (usefully available in beige, blue, black and green (!)).
• 1.5 Gb/s VCSEL-based transceiver units (normal units are only specified for Gb Ethernet speeds [1.25 Gb/s])
• Le Clip. A non-magnetic replacement for the fiddly springs used to connect MT connectors together. The design concept came from CERN.
• FIBERCORE. Fibercore confirmed that fibre fabricated with no phosphorous in or near the core were likely to be radiation tolerant. This confirms LArg results on a fibre manufactured by Plasma Optics and resold by Alcoa-Fujikura.
• EUROPTICS. Different constructions of multi-optical ribbon cables were described. Some favourites are 8x8 fibres, 4 groups of 4x4 fibres and 5 groups of 5x4 fibres. Generally such cables are stiff and may not be useable within ATLAS.
• ERICSSON. Ericsson explained that they use Plasma Optics in Holland as their preferred source of multimode fibre nowadays. This is good news as a fibre from Plasmas has already been tested by the LArg links group and been found to be sufficiently radiation tolerant. Ericsson also confirmed that a standard 62.5mm core fibre could support Gb/s traffic at 850 nm. They also commented that in general such fibre have a more attractive bend radius than a 50 mm core fibre . Toplogies of different ribbons were discussed - maximum of 12 fibres can be readily ribbonised today. They also mentioned the importance of fibre management.

Report from Visit to Middlesex University (GaAs Gb/s serialiser)
Click here for all the slides
 
CHFET Serialiser Developments for CMS E-Cal
    Finally, Mark described progress with the CMS E-cal CHFET serialiser chip. This device is particularly appealing for LArg links as it can serialise 16 bits at 40 MHz (i.e.: two per FEB), encodes clock / maintains DC balance etc. a la G-Link (CIMT protocol), dissipates around 100 mW per chip and should be inherently radiation hard due to the properties of the CHFET process. The chip is designed to be transparently compatible with standard G-Link transmitters and includes an in-built VCSEL driver for standard COTS VCSEL's. Mark reviewed the status of the chip (mainly taken from Peter Denes' talk at the Oxford workshop), briefly :

• A first batch of chips has been produced (July 98) without the protocol implemented (added with external Xilinx chips). Chips were characterised and irradiated (protons at PSI). Chip functionality OK, but no BERT tests performed.
• Poor yield from wafer due to ESD problems. Honeywell have since helped to develop suitable protection for inputs.
• Pilot run to be submitted Feb. 99. All chips will use new ESD pads. Some chips will have CIMT protocol inbuilt.
• Next batch to be packaged (serialiser and VCSEL driver) in TQFP44 package.

Discussions
    There were several issues discussed :

• Bit error rate tester (BERT). Bernard Dinkespiler briefly reported on the status of the 'home made' 16 channel BERT. A two link prototype is nearing completion and full versions are expected in the Spring.
• Testbeam links. The need for a complete implementation of copper links for both of H6 and H8 was agreed. A timing problem at the FEB MUX (32 at 40 MHz converted to 16 bits at 80 MHz) encountered during a first attempt to integrate G-Link based links into the testbeam set-up in November 1998 would be revisited at the end of February once Laurent Serin had been consulted about the availabilty of the DAQ system. A final implementation filling a mini-ROD crate could wait until July (on H8). It was generally agreed that there was little point in fully equipping the testbeam with optical links at this stage due to the cost and manpower needed. Jacques confirmed that the noise levels due to copper read-out cables could be accepted.
• Strategy. There was a general discussion on how to ensure a radiation tolerant baseline FE-Link solution would be available given the current uncertainty surrounding the G-Link solution. This would be discussed by the LArg links group and a solution proposed in the future (Annecy LArg electronics workshop, 10-12 March perhaps).