Vertex 2005 Nikko Manfred Pernicka HEPHY Vienna Contents

Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Contents Some general remarks about occupancy reduction An advanced method using the APV 25 front-end chip How it could be realised for the upgrade of SVD 2 @ Belle Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

What are the problems and effects with high OCCUPANCY ? • BELLE definition: every strip above threshold plus 2 neighbours for every cluster, divided by the total number of strips • Effect depends on the number of layers and the spatial resolution. • Above a certain size, occupancy creates fake tracks. • Track finding becomes more complicated and the results are of lower quality. The detection of track points is reduced. Therefore an increase of beam intensity ( with included background) can reduce the performance of a detector, thus the number of useful events and finally also the physic results. Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 3

Possibilities to reduce the effects of the occupancy: A. reduction of the sensitive area/channel of the detector. Finally you will end up with a Pixel detector with an area of e. g. 100*150 µm 2 like CMS. The occupancy at CMS will be around 0. 1%. The sensitive time window for a trigger will be around 25 -30 ns. Threshold for internal trigger around 2500 electrons T CMS Pixel ASIC would not be optimal for continuous beam (because it is designed for bunched LHC beam clock synchronised ). Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 4

B. with the reduction of the shaping time. The APV 25 offers a range of 450 ns down to 35 ns (designed for 50 ns). Threshold Even a short peaking time (~50 ns) creates a wide sensitive time window (~160 ns) above threshold. Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 5

Some more details about the APV 25 • Single or multi peak mode • Pipeline 192 cells • Multi event buffer for 30 event-triggers with single hit read out, multi hit mode 10 event-trigger. • Calibration system, 8 time steps per clock, max charge 25 f. C • Readout speed: clock frequency or half clock frequency, for data multiplexing at CMS • Output levels adjustable • 0. 25µm CMOS: radiation hard (>100 MRad) • Works up to 80 MHz (samples tested but not guaranteed) • 1. 25, 2. 5 Volt supply • Clock, trigger/cal/reset and output are differential signals Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 6

Analogue data processing APV 25 Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 7

C. A further improvement: Reduction of the sensitive time area with analogue data processing. You need 3 analogue values of the shaped signal. These signals are combined with different weights, resulting in a pulse with reduced width. The APV 25 has this possibility called “deconvolution mode”. + The time sensitive window is reduced by a factor of ~ 4 - Can only be used for beam intersection every 25 ns 40 MHz not for continuous beam! - S/N reduced measured Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 8

D. External Pulse Shape Processing The APV 25 has the possibility to store 3 consecutive samples (spaced by the system clock) of a signal with one trigger (multi-peak mode). With a second trigger just 3 clocks later, you can get the next 3 time samples of the shaped input signal. We can use this mode to determine the peak time of the shaped detector SIGNAL relative to the CLOCK. We can also measure the time between the TRIGGER and the (APV 25) CLOCK. (Continuous beam: trigger can come at any time related to the clock. ) Comparing those 2 timing values we can select if the hit belongs to the event or reject as background if it is out of the selected time window (~20 ns). Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 9

Eliminating background using time information Time of the hits t=B t=A t=A t=B t=A + Time of the trigger to the clock t=A = t=B t=A Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 10

12 time slices of a shaped input signal created by 4 triggers spaced by 3 clocks (Beam test CERN August 2004) We can use 3 or more pulse height values for time calculation. Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 11

Unsynchronised trigger Trigger D Q Clock synchronised trigger 40 MHz clock Sampling clock Clock The time between these 2 signals must be measured TDC Clock S 1 S 2 S 3 Calculate peak time to clock . Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 12

M. C. signal with noise Accepted as signal above threshold ~230 ns As signal detected time width ~230 ns Threshold Selected with S 1, S 3<=S 2 <50 ns Selected S 1<S 3<=S 2 and S 3<S 1<=S 2 ~25 ns Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna ~25 ns 13

First sample 1' 3 a 4 1 2 3 b Peak time Peak Shaping curve Peak time is measured from first sample 3 a 4 1' 1 2 3 b Unsync'd Trigger delay is measured between unsync'd and sync'd triggers Trigger delay + Peak time = constant = distance of Unsync'd Trigger to Peak 1 despite of clock phase wrapping 2 Clock phases 3 4 1' Sync‘d Trigger Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

…and with the knowledge of the shaping curve, it is possible to obtain the peak time with ns precision. Comparison of APV 25 output waveforms (Tp=50 ns): • Simulation (ideal=Dirac current pulse), • Simulation (real=realistic detector current pulse), • Measurement (internal calibration), • Measurement (extracted from particle signals) Knowledge of shaping curve important for the precision of the time measurement. Expect that every input channel of the APV 25 has the same shaping curve. Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 15

Reduction factor comparing VA 1 TA (current Belle SVD readout ASIC) and APV 25: VA 1 TA shaping time 800 ns 2000 ns above threshold = accepted as signal APV shaping time 50 ns 160 ns Reduction factor ~12. 5 160 ns above threshold Using time information of signal and trigger to clock 160 ns ~20 ns Reduction factor ~8 Cost: something needs an APV 25 readout system No extra Cost: requires some extra processing power Total reduction ~100 Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 16

Correlations with double-sided readout • For matching hits in p and n sides, the following data are correlated and can be used to create a fast trigger: 1) time of the hit 2) pulse height 3) cluster size (if particle has same angle against both strips, like the UV-striplet detector proposed for Super-Belle) • These correlations can be used to remove ambiguities, eliminate ‘ghost hits’ and thus reduce background • A requirement on the cluster size can be used to reject hits from tracks with don’t originate from vertex Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 17

With time information it is possible to determine the hit point(s) and judge if a cluster contains one or more hits in most cases. t=A Using 6 time slices t=A and t=B and t=C same amplitude same time same cluster width t=C t=A t=a Vertex 2005, t=B Nikko Manfred Pernicka, HEPHY Vienna 18

To get a impression of the possible time resolution of the processed 3 time slices of a hit, we use the p and n side data of a UV-striplet detector with APV 25 readout obtained in a beam test. The trigger time has no influence on the result. Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Correlation plots u-v pulse height time information cluster width (vertical incidence) Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

(const-TDC) vs. fitted peak time Residuals RMS=2. 16 ns (including scintillator trigger jitter) Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Beam only run 056 Source only run 048 Beam + Source run 050 Time in ns Top row: peak time distribution Bottom row: Signal distribution ~same statistics for beam and source Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 23

Beam + Source + Peak time cut run 050 Beam only run 056 Removal of source (background) by peak time cut Sensitive time window reduced to 20 ns Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 24

Fitted peak time residuals vs. SNR Time resolution (RMS residuals as as function of S/N) P-side N-side As expected this method suffers with low S/N Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

A trigger with good time resolution in combination with an APV 25 multi-hit readout system dramatically reduces the computing time and opens new possibilities at extremely high luminosity ideal trigger with little jitter (~3 ns) Jitter ~20 ns + 3 ns of trigger We need more time slices when the trigger has more time jitter and the method is compromised. trigger with jitter With a reduced trigger time information we can still find hits which fit to a track but the selection quality to an event is reduced. Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 26

3 requirements for this system: A. Trigger for the APV 25 readout system with good time resolution. RMS as small as possible, ideally <3 ns B. Delay for trigger <160 clock cycles C. Good S/N Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 27

Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

A lot more data has to be handled. Therefore we propose a signal processor on the ADC module. • 3 -6 times more data in multi-hit compared to single mode. Therefore every input should have its own data pipeline processor for hit finding. The processor could handle a trigger rate up to 30 - 40 k. Hz and an occupancy of 3 -4%. Limited mainly by the output connection to the DAQ. • It is open how far we want to go in hardware data processing on the module. Anyhow it is foreseen to switch between three different modes: - data above a threshold with coordinates - transparent and processed data together in a block - hit data with peak time information We can realize that using a Nikko modified CMS Vertex 2005, HEPHY Vienna Pixel read. Manfred out. Pernicka, module

9 channels are processed at the moment in one FPGA, this number could be reduced, because the FPGA also has its limits. Reorder : result 1 to 128 Exists and working Hit finding exists and working Time calculation hardware exists Hit finding on strips + neighbours with 2 steps of common mode correction Looks for the time between clock and signal D. P. M Channel x m u x De m ux D. P. M Reordered Memory + delay of 128 clocks + delay of 4*128 clocks + delay of …. . Still some work to optimise… 0, 32, 64, 96 | 8, 40, 72, 104 | 16, 48, 80, 112 | 24, 56, 88, 120 || 1, 33, 65, 97 | 9, . . . O, 1, 2, 3, 4, … 128 Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 32

The prototype FADC + Processor Electrical input board instead of optical receiver Clockcontrol signal Mezzanine card for LAN output and further data processing Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Summary: We are confident that we can eliminate occupancy problems of the BELLE Vertex detector for several years Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

Backup slides Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna

PSI beam test (August 2005): high intensity and inclined detector Operation with beam synchronous 50 MHz clock Amplitude (e) pileups Each colour = signal of a cluster Time (x 20 ns clock frequency) Vertex 2005, Nikko Manfred Pernicka, HEPHY Vienna 37