1. 1 ECAL/HCAL FEB and CROC. FEB serialisers Serialiser tests GANIL Irradiation test. CROC bit flip error modifications and tests Calorimeter Detector/Electronics Meeting on behalf of the LAL Group Frédéric Machefert Wednesday June 25th, 08

2. 2 Analog Device serializer problem The present serialiser in M24 (trigger) do not accept overshoot on its input DS90CR215 (National Semiconductors) The trigger BGA outputs present such features A “protection” starts working in the DS90 The component pulls plenty of current (~1/2A) Starts to increase its temperature The component still works We may suppose it can not bear such conditions for a long time Think for a replacement Two components may suit the board requirements Texas SN65LVDS95 Fairchild FIN1215

3. 3 Texas and Fairchild components in tests Boards have been modified to receive the Texas and Fairchild components Still large oscillations But no harmful protection The oscillation should be painless for the Fairchild part Fairchild has the preference Sampling region Signal should be stable Sampling region Signal should be stable Threshold 2V

4. 4 At LHCb SEL are triggered indirectly by neutrons The probability to produce a sufficiently ionising fragment is low The neutron flux is large in the experiment Only some ions from the neutron-Si collisions have a deposited energy (LET) sufficient to produce a SEL The energy of the ion should be greater than 1 or 2 MeV. Maximum LET reachable is 15MeV.mg -1.cm 2 GANIL irradiation : motivations and conditions (I) Ionising fragment : SEL small probability! neutron Si LET>6MeV.mg -1.cm 2  1.5MeV.  m -1 LET max =13/15MeV.mg -1.cm 2

5. 5 Probability to produce a fragment of energy greater than E MeV from neutron collision (1GeV) But we also know that E>2MeV Ion is N, O,..., Si GANIL irradiation : motivations and vonditions (II) (1GeV incident neutron) Supposing that all these fragments produce a SEL, the probability to trigger a SEL from a 1GeV neutron is of the order of 2x10 -6 To determine the probability to produce a sufficiently ionising fragment it is necessary to evaluate the probability per μm times the range of the ion considered and to sum up over the various ion species

6. 6 LHCb : flux of particles in the electronics vicinity Origin : Calorimeter Origin : Interaction Pt MeV -1.cm -2 Energy (MeV) L=5x10 32 cm -2.s -1 In the region of the electronics We expect 4x10 9 incoming neutrons of more than 30MeV in a year (10 7 s) per cm 2 Less than 8000 ionising fragments (cm -2 ) capable of producing a SEL But most of them have a LET around 6 MeV.mg -1.cm 2 Just a few can reach a LET of 16 MeV.mg -1.cm 2

7. 7 GANIL irradiation We used Krypton (70MeV/A) Deposited energy is small (a few MeV.mg -1.cm 2 ) Use an absorber (various thickness) to reach LET up to 16 MeV.mg -1.cm 2 Dose is of the order of 5krad/2x10 7 particles 16 MeV.mg -1.cm 2 Absorber : 500  m - Al

8. 8 GANIL results Texas No SEL was observed Supposing that the LET of the fragments from neutron collision is always max. Texas was tested for more than 7500 years without SEL (a single part) ! Illustration of the power of a Krypton test at GANIL Safe radiation tolerance Fairchild Several SEL have been detected SEL rate clearly depends on the LET (threshold effect) Taking same “a priori” as for Texas A SEL every 180 years (single part) This is VERY pessimistic We may try to evaluate rate more carefully (more “realistic assumptions”)

9. 9 Fairchild : specific study Taking into account The SEL rate dependence with respect to the LET The probability to produce ions of Various types Leading to specific LET along their path Rate goes down to 4x10 -7 SEL per year ! We may assume that Fairchild is also radiation safe ! Modelisation of the SEL cross-section SEL contributions from the various ion species

10. 10 Calorimeter Detector/Electronics Meeting CROC Bit flip : problem The BGA cannot cope a two frequent bit flip on two many outputs BGA levels fluctuate (GND and VCC) This is not observed on the PCB (Monitored - Tried to mount huge capacitors) Application notes indicate : Inductance inside the BGA Large output buffer bit activity Vcc/GND derive ~ L. di/dt For us : Large input flip means large output flip The effect is larger if internal register activity Input is a comparator with respect V cc /2 if V CC fluctuates, may imagine duty cycle change (signal fronts are not vertical)

11. 11 Calorimeter Detector/Electronics Meeting CROC Bit flip : solutions tried Internal activity (Spying) does not help Do we have a problem when normal LHCb functioning ? A test was planned at 1MHz in the pit : cannot be done in the present situation We tried to correct the problem by changing the firmware of the FE PGA 1) Problem is linked to the current flow in the BGA Set output in slow slew rate : slower current variations standard is fast slew rate but slow slew rate is good enough for us 2) Imposing constraints on the routing Keep internal registers away from the output buffers 3) Do NOT flip all the bits EXACTLY at the same moment Out of the 23 bits (out of the FePGA), ~8 LSB often flip Time shift for the 4 first ADC bits 1212 21 bits Clock 19 MSB 4 LSB  t~ 4/5ns

12. 12 Calorimeter Detector/Electronics Meeting CROC Bit flip : tests performed At LAL Procedure 1MHz L0 triggers 8 boards in the crate Random capture of the events with CROC Spy Could not observe any single error of of several 100k L0s passed At CERN Re programming of the CROC in crate 19 (full crate) Test procedure is the same as the one at LAL Could not observe any error on ~100k L0s Still, we wish to be able to perform a 1MHz test with the “real” DAQ Crate 19 and 10 already use a “modified firmware” Have been operated for one week No ECS modification (exactly identical CROC configuration as previous firmware) Latest version will be loaded in the CROC this week

13. 13 Calorimeter Detector/Electronics Meeting CROC GOL tuning Modifications of the CROC firmware have consequences on the timing of the bits sent to the GOL Before : After : Margin is still OK Reference or point of functioning in standard condition