1Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB and RAT Status Report
Jay HauserJay Hauser
University of California Los AngelesUniversity of California Los Angeles
2Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB Hardware Since April Emu@OSU• Procured all FPGAs (Xilinx X2V4000)• Got fixed Virtex-2 mezzanine layout from Kan
• I/O pads are grouped in pairs, each pair needs to use the same clock.
• Made 40 X2V4000 mezzanine cards• 2 remaining mezzanine card problems (there were 3 simple fixes)
• Made 29 working sets (TMB-base, TMB-mezz, RAT)• 1 base board has a bad SCSI connector• 9 sets used for full-crate tests• No problems seen at FAST site
• Procured all parts for full production (520 sets)• Still waiting for some hex switches and front-panel parts (due soon)
• Full-crate tests done at OSU, no problems• May-June test beam problem found:
• ALCT inputs through RAT did not work at all.• RPC inputs through RAT seemed okay (minor problems).
3Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
RAT-to-TMB TTL Data Xmission• RAT gets everything (power, gnd, control signals) from TMB• Currently no connections to the common backplane
TMB
RAT
Backplane
NC
4Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
RAT2003A DetailTo TMBTo TMBGND only GND only
RPC connectorsRPC connectors
To TMB and 3.3v, To TMB and 3.3v, 1.8v power 1.8v power
ALCT ALCT LVDS-TTL LVDS-TTL convertersconverters
Spartan 2E Spartan 2E FPGA for FPGA for
RPCRPC
ALCT SCSI connectorsALCT SCSI connectors
Mechanical Only Mechanical Only
5Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
RAT-to-TMB Ground Problem• May-June test beam:
• Only the bottom connector on backplane
• 110 single-ended 80 MHz TTL signals from RAT to TMB.
• Reference ground and current return on 6 pins.
• “Typical” design used for all other modules in peripheral crate:• 1 GND per row of 5 pins, i.e. 25 GNDs per 125-pin connector
• Ground reference close to every signal
• Grounded to the backplane
It is not a DC current problem, it is an RF problem • 40 MHz noise
• 80 MHz signals, edges with 1-2 ns rise times.
6Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Ground Issues, cont.• RPC signals seemed to work at May-June t.b.
• They are buffered through FPGA close to connector
• May have slightly better RF characteristics (N.B. no CRC available).
• Also a problem: layout house for TMB used very thin ground and power planes• ½-oz thickness of copper
• Usual thickness is 1-oz or 2-oz
• Ground plane impedances therefore higher
• Bench tests show very significant ground plane noise
7Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB Ground minus RAT Ground
100 kHz ALCT
600 mV P-P
1 MHz ALCT
700 mV P-P
No ALCT triggers
300 mV P-P• 40 MHz noise depends
on ALCT trigger rate:
8Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB-RAT Ground Details
9Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
RAT-to-TMB Ground Problem• Bench investigations using CRC error checking of
ALCT data:• Checks for any bit errors in ~800 bytes of data.
• Rate of errors also seen to increase with ALCT trigger rate– Errors start when ALCT readout and triggering overlap in time
– Reaches ~100% before 1 MHz.
• Checks signal integrity through 8-bit DAQ path (out of 50 ALCT signals), i.e. many signals not checked.
• UC FAST site investigations using CRC:• Typically 0.5-2.0% errors seen.
• “New and improved” short backplane– 11 GNDs added, but on a different connector than TTL signals
– No observed change in the rate of CRC errors.
10Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Test Beam Study of TMB-RAT Interface
• CRC checks that every bit of ~800 bytes is transmitted correctly.
• May-June test beam:• TMB backplane fed GND to RAT with only 6 pins.
• ALCT data did not successfully go through at all.
• Sept-Oct test beam:• TMB backplane had additional connector with 11 more GND pins.
• ALCT errors were found only at high rates at the Sept-Oct test beam.
• General observation that CRC errors increase with rate agrees with bench tests.
11Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Sept. 2004 Test Beam Study
0.00E+00
5.00E-08
1.00E-07
1.50E-07
2.00E-07
2.50E-07
3.00E-07
0 100 200 300 400 500 600 700 800
Rate (KHz)
ALCT
CRC
Erro
r Frac
tion
• High-rate CRC error test:• Went up to 3.4 million muons
per spill before radiation alarms went off (ahem)
• Otherwise, ALCT-RAT-TMB worked fine
74.5 Million ALCTs
15 CRC errors
12Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Conclusions on RAT-TMB Data Transmission Tests
• Ground noise is quite large• Level of success of tests varies dramatically:
• Absolutely no success at May-June beam test
• Errors reproduced and understood with bench tests
• Small rate of errors observed at UC FAST site
• Very very small rate of errors observed at Sept beam test
• Variable level of success not really surprising for a noise problem• Sitting on the edge of a serious problem
• This situation is unacceptable for long-term CMS operation, we would like to eliminate this problem entirely
13Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Remedy
• Make the grounding for RAT like any other board in the peripheral crate:• Add one ground pin on every row of RAT pins
• Add backplane ground connections
• Requires more total pins:• Replace 55-pin X3 and 125-pin X18 connectors by two 110-
pin Type-A connectors
• 40 more pins, all “C” column pins used for grounding
• Also more robust alignment pins (noticed RAT alignment pins already broken before Sept beam test)
14Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Implications• Backplane layout needs modification
• ~3 months, according to Alex
• TMB and RAT board layouts need modification• Also ~3 months, according to our estimate
– Includes thickening up the 1/2-oz ground and power planes and necessary changes to details of vias, thermal reliefs, etc.
– Includes “cleaning up” the data pathway from RAT to TMB.
• Revised “loop-back” backplane used for self-testing– will add proper RAT connection for production testing (previously RAT-TMB
connectors inserted by hand, subject to bent pins)
• More type-A connectors and backplane shrouds need to be ordered• Backplane pins have long lead-time, should be ordered immediately
15Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB/RAT Production Schedule• Mezzanine card production can begin next week
• No problems seen with Virtex-II mezzanine card
• Re-layout of RAT and TMB:• RAT: 2 weeks• TMB: ~10 weeks• Testing loop-back backplane: 1 week
• Quick prototype cycle in February.• Production immediately after that.• Testing goes quickly, more than 4 boards per working day.• Deliveries of working TMB and RAT pairs:
• 25% by April 30 (130 of 520 boards, including 10% spares)• 50% by June 30 (260 of 520 boards)• 75% by August 31 (390 of 520 boards)• 100% by October 31 (all 520 boards)
• Boards will be shipped as soon as they are tested and acceptable space (not boxes) exists for them at CERN
16Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Happier Matters• Active pulsing of CFEB front ends• Active pulsing of test strips• DAQ error checking by CRC technique• New TMB counters
17Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
DAQ Error Checking by CRC• CRC = Cyclic Redundancy Check• Special “checksum” (22 bits) added to end of data
block to represent pattern of 1’s and 0’s sent• This checksum can be re-calculated from the 1’s and
0’s sent, then compared to the checksum received.• ALCT and TMB both create CRCs for their data.• TMB also checks ALCT CRC on the fly
• TMB adds CRC-error bit to last ALCT trailer word.• TMB counts & stores in VME register the number of CRC errors from
ALCT.• DDU now separates CRC errors coming from ALCT, CLCT/TMB.
• CRC bench test:• Special ALCT firmware creates dummy buffer of 400 words and responds
to external trigger• CRC can be checked at arbitrary rate with pulse generator
18Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB Counters• VME registers recently added for real-time
counts of 23 various interesting things:• ALCTs
• CLCTs
• Matched LCTs
• L1 Accepts
• ALCT readout
• CLCT readout
• MPC accepts (muon 0, muon 1)
• ALCT CRC errors
• Etc. etc.
• Very useful diagnostics at test beam
19Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Additional Slides Follow
20Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Active Pulsing of CFEB Front Ends• How it’s done:
• All channels have capacitors with 4 levels of charge that can be preset (0,1,2,3) within Buckeye ASIC
• “left half-strip” puts strip charges at ….,0,0,0,2,3,1,0,0,0…• “right half-strip” puts strip charges at …,0,0,0,1,3,2,0,0,0…• Single VME command to DMB pulses all channels simultaneously• Patterns give e.g. 6-layer CLCTs• First vary 40 MHz clock phase from TMB to comparators until
patterns correctly found• Then patterns can be swept across entire chamber and checked
• Advantages:• Finds optimum phase of 40 MHz clock to comparators• Checks all Buckeye, comparator, and CFEB-TMB Skewclear
cable channels• Test takes a few minutes• No gas, HV, etc. needed
21Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Active Pulsing of Test Strips• How it’s done:
• CCB provides 500ns gate to make the ALCT test strip pulses• Capacitive coupling between test strips and anode wire groups
gives pulses on leading and trailing edges of 500ns pulse• All channels fire on leading edge• Hot wire mask on ALCT board ALCT patterns• Like CFEB, vary 40 MHz receive and send phases until optimum
data transmission from TMB to and from ALCT• Find optimum in 2D matrix of receive/send clock phases• Scan patterns across chamber to find bad AFEB/ALCT channels
Advantages:• Set up phases of 80 MHz clock TMB to and from ALCT with high
reliability without HV, gas, or cosmic ray data• Check all AFEB, ALCT, and Skewclear channels through the
system• What previously took days now takes minutes
22Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
More on TMB and RAT Power
23Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB Ground minus RAT Ground
100 kHz ALCT
600 mV P-P
1 MHz ALCT
700 mV P-P
No ALCT triggers
300 mV P-P• 40 MHz noise depends
on ALCT trigger rate:
24Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Other Noise Measurements• RAT GND versus
RAT +3.3v (okay) 100 mV/division
GND vs +3.3v
Clock off
GND vs +3.3v
Clock ON
GND vs +3.3v
Clock ON
ALCT 1 MHz
25Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Other Noise Measurements• TMB GND versus
TMB +3.3v (okay) 100 mV/division
GND vs +3.3v
Clock off
GND vs +3.3v
Clock ON
GND vs +3.3v
Clock ON
ALCT 1 MHz
26Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
Other Noise Measurements• TMB “logic 1” signal as seen at
RAT, 2v per division
• Big noise seen, as expected from GND(TMB) – GND(RAT) measurements
Logic 1
Clock off
Logic 1
Clock ON
Logic 1
Clock ON &
ALCT 1 MHz
!
27Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
LVTTL (from +3.3v) Logic Levels• Typical outputs 0.4v low, 2.4v high, transition ~1.5v• Worst case output supposed to be 0.8v low, 2.0v high• Measurements show violation of that
28Endcap Muon meeting: FNAL, Oct. 29-30, 2004 J. Hauser UCLA
TMB, RAT Power Supply Filtering Capacitor Design
• TMB base board• Power sources for +3.3v and +1.8v: 2x100uF, 4x4.7uF, 1x1uF, 1x0.1uF
• Power destinations: hundreds of 0.1 uF capacitors across board, typically several for each driver
• Power sources for +3.3v and +1.5v for RAT: 1x100uF, 1x4.7uF, 1x1uF, 1x0.1uF
• RAT board• 3x4.7uF, 12x0.1uF
• Many 0.1 uF capacitors distributed across board, several for each driver
• Virtex-II mezzanine board• 2x330uF, 10x47uF, ~20x0.1uF, ~40x0.022uF distributed according to
Xilinx spec.
• Connections to TMB base board with 56 ground pins, 19 +1.5v pins, 17 +3.3v pins
• Comment: looks OK