GTU Architecture

1
ORI diagnostics TRAP online software new
features, open questions Corrupted data Data
reader analysis Conclusions
2
Diagnostics of the dead links in SM II
Fortunately our laser driver chip has many
diagnostic features. We can read the mean current
through the diode and the mean photocurrent in
the monitor diode. The optical power is linearly
dependent on the two currents and can be
estimated with an error of -50..100. More
diagnostics compare the EEPROM and LTC, read
status flags. Resuts the 4 ORIs in SM II have no
internal errors, the estimated optical power is
500..800 µW.
3
TRAP online software (1) maj. ver.
6
5
4
3
2
1
0
TP
ZS
DT
RS
O2
O1
O0
Options
Reserved
Disable Tracklets
Zero Suppression
F2
F1
DE
Disable Empty Hd
Test pattern mode
321 send full every 1024/256/128 event
P2
P1
P0
Pattern Id
No TP, no ZS
-
-
SM
Statistics Mode
With these 7 bits on the same address in all
TRAPs we can control the readout type
4
TRAP online software (2) statistics mode
No TP, no ZS
-
-
SM
Statistics Mode

• Online by the TRAP CPUs
• Accumulate for each channel in sumch (32 bit)
  the ADC values
• Accumulate for each channel in sum2ch (64 bit)
  the squares of the ADC values
• Offline
• Read by SCSN the sum and sum2 from DMEM
• Using the total number of accumulated samples
  nNeventsNsamples calculate the mean and RMS

The fastest way to monitor the noise even without
readout!
5
TRAP online software (3) test patterns

• Four patterns implemented so far. Briefly the
  properties of each pattern
• 0 oldest, based on rob, mcm, cpu and a word
  counter. Not recommended, as 2 and 3 are better
• 1 ADC data replaced by pseudorandom 10 bit
  data. Recommended when searching for bit errors
• 2 based on a part of the event counter, the
  full position of the mcm in detector (sm,
  plane, chamber, rob, mcm) and a word counter
• 3 like 2, but with more bits from the event
  counter and without a word counter
• With patterns 2 and 3 one can identify the source
  of each data word.

6
TRAP online software (4) e/pi LUT
In the TRAP chip we can accumulate online the
cluster charges in two programmable windows. As a
part of the tracklet (8bit) the TRAP sends the
electron probability.
Q1
Q0
We have free memory in the TRAP for 11 ? 8 bit
LUT. The 11 input bits are composed from the
binary scaled Q0 and Q1, eventually normalized by
the total number of hits. We need somebody to
find the optimal windows and table.
7
Corrupted data
Break in the raw data, tracklet endmarker of the
next link without raw data endmarker of the
present link.
0x83004119 h0 0x7BA25E8D h1 0x8C0000AC MCM
header, MCM12 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x8D0000AC
MCM header, MCM13 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x8E0000AC
MCM header, MCM14 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x8F0000AC
MCM header, MCM15 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x880000AC
MCM header, MCM 8 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x890000AC
MCM header, MCM 9 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x8A0000AC
MCM header, MCM10 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x8B0000AC
MCM header, MCM11 ROB 0 Event cnt 10 Link
9 0xFE00000C 0x7F ADC_mask "1100" 0x10001000
tracklet endmarker? 0x10001000 0x8300411D
h0 0x7BA25E8D h1 0x9C0000AC MCM header,
MCM12 ROB 1 Event cnt 10 Link
10 0xFE00000C 0x7F ADC_mask "1100"

run748_ldc_daq09.001.raw
Always after 8 MCMs
Swap the optical cables Use the latest TRAP
software and test patterns. Read clear the
counters in ORI after each event Use all GTU
diagnostics features word counters, timers (from
L0), link monitor with 4 bit error code,
endmarker counters etc.
8
Data reader analysis

• Currently we have two readers of Mateusz and
  of Jan. I have a simple reader indented more for
  single link usage ROB test
• Up to date with the TRAP software!
• Automatic check analysis of the data taken in
  test pattern mode
• - Find an error and skip the rest of the link
  data or of the event is not enough
• - Count the bit errors for each bit, each link,
  each MCM if necessary
• - Count the breaks in the data structure, again
  for each link, storing the position of the break

9
Conclusions
It is not efficient to search for errors in old
data, taken with out of date software
hardware Shorten the turn around time from
the change in sources (TRAP configuration, test
software, firmware update etc.) to the usage on
the supermodule. Shorten the time from the
measurement to the data analysis More reliable
configuration, software to check the configuration
10
Spares
11
Wafer test results
Production run
Tested Usable
Preproduction (25 wafers) 14324 5056
35 Feb 2006 (25 wafers) 14400 8143
57 Jun 2006 (49 wafers) 28006
24018 86 Sep 2006 (49 wafers) 28212
24600 87 Jan 2007 (3 wafers) 1728
1568 91 Mar 2007_1 (25 wafers) 14381
11044 77 Mar 2007_2 (25 wafers) 14366
11927 83 All
115417 86356 75
We have about 40,000 unbonded TRAPs (2007.09)