CBC2: CMS microstrip readout for HL-LHC [D. Braga], G. Hall , M. Pesaresi, M. Raymond (Imperial College) D. Braga, L. Jones, P. Murray, M. Prydderch (RAL) D. Abbaneo, G. Blanchot, A. Honma, M. Kovacs, F. Vasey (CERN)

Background CMS upgrade under consideration for many years objective to reach 3000 fb-1 in next decade or so High Luminosity LHC (Phase II) requires new Tracker around 2023 Requirements: lower material budget increased granularity enhanced radiation tolerance tolerable power consumption affordable cost all compatible with physics objectives –some of which remain uncertain e.g. will new physics be discovered in next running period? but solid long term programme of Higgs & top studies, searches, etc… Geoff Hall HSDT9 Sep 2013

Evolution of objectives Original goal new – improved - tracker with similar angular coverage, constrained by re-using existing services provide some part of tracker data to L1 trigger to contain rate to 100 kHz More recent developments Baseline Tracker design now adopted “conventional” barrel-endcap layout looks optimal but CMS exploring enhancing forward region physics as well as standard physics programme uncertainty if L1-track triggering will reduce rate to 100 kHz in 6.4µs ideas (and detector requirements) not yet validated by simulations possible objective of L1 readout up to 1 MHz/10-20µs both approaches require on-detector data reduction Geoff Hall HSDT9 Sep 2013

Baseline tracker layout (pixels not shown) 2S short strip double-layers (~7500 modules) PS strip-strixel double-layers (~10,000 modules) Double layer readout compatible with trigger Geometry compatible with forward extension Geoff Hall HSDT9 Sep 2013

ASIC development Earliest developments foresaw conventional outer tracker, plus some dedicated trigger (and pixel) layers 128 channel CMS Binary Chip (CBC) produced and proven 2011 Readout architecture followed original tracker, using APV25 Analogue data abandoned for practical reasons digital optical link components now standard, ADC power, 130 nm CMS harder to implement analogue cells Subsequently, outer modules with trigger capability 254 CBC2 successfully developed – delivered 2012 optimised for new assembly method with mass production in mind Geoff Hall HSDT9 Sep 2013

Present CMS Tracker architecture 8.1 mm Analogue unsparsified readout APV25 in 0.25µm commercial CMOS synchronous occupancy independent data volume 2.7mW/chan for 10-20cm µstrips analogue data transmission to external ADC & zero-suppression, clusters semi-custom optical links @ 40Msps 1310 nm single-mode Fabry-Perot lasers very successful reflected state of technology at the time benefits simple and easy to debug/evaluate robust against noise pipeline 128x192 7.1 mm 128 x preamp/shaper APSP + 128:1 MUX bias gen. pipe logic control logic Analogue Optical Hybrid Geoff Hall HSDT9 Sep 2013

Basic trigger module concept Compare binary pattern of hit pixels on upper and lower sensors Pass Fail High pT tracks can be identified if hits lie within a search window in R-f (rows) in second layer R Upper Sensor 1-2 mm ~200μm Lower Sensor ~100μm f ~mm-cm f (rows) Sensor separation and search window determines pT cut z-segmentation determines vertex capability z (columns) Geoff Hall HSDT9 Sep 2013

2S PT-module with CBC2 Track & trigger µstrip module for outer tracker region CBC2 logic correlates hits on two sensors to reject those from low pT tracks low mass, high density interconnect layer two layers of sensors chips on top surface only signals from lower sensor via’d through substrate 8 x 254 channel chips bump-bonded to hybrid concentrator & controller ~5 cm ~5 cm 90 µm pitch strips sensors wire-bonded above and below 8 x 254 channel chips bump-bonded to hybrid Geoff Hall HSDT9 Sep 2013 8

CBC main features IBM 130nm CMOS process binary, unsparsified architecture retains chip and system simplicity but no pulse height data designed for ~2.5 - 5cm µstrips < ~ 10 pF 128 channels, 50 mm pitch wire-bond either polarity input signal not contributing to L1 trigger powering test features: 2.5 -> 1.2 DC-DC converter LDO regulator (1.2 -> 1.1) feeds analogue FE fast (SLVS) and slow (I2C) control interfaces 4 mm 2.5->1.25 DC-DC converter 256 deep pipeline + 32 buffers 7 mm amplifiers & comparators 128 channel input bias generator linear drop out regulator Geoff Hall HSDT9 Sep 2013

CBC measured performance 115k 80f 1p VPLUS Vdda 16k 200k 100f 60k 92k VCTH 2k 4k 8k 500k preamp postamp comparator simulation: open circles simulation: open circles analogue 130 + (21 x C [pF]) µW/chan digital < 50 µW/chan total 180 + (21 x C[pF]) µW/chan e.g. < 300 µW /channel for C = 5 pF Geoff Hall HSDT9 Sep 2013

CBC comparator thresholds adjusted satisfactorily timewalk within spec events above threshold comparator global threshold [mV] 128 channels before tuning after threshold uniformity VCTH thresholds adjusted satisfactorily timewalk within spec timewalk: threshold at 1 fC VDDA postamp O/P O/S adjust 8-bit value (per channel) 16k VCTH hysteresis 2k 4k 8k 500k postamp O/P comparator Geoff Hall HSDT9 Sep 2013

Beam telescope Based on CMS Tracker DAQ readout hardware, software (used for UA9 studies) FED, APV25s, 100m fibres, custom control system, multi-core PCs 50 kHz data taking during 10s spill, 10 kHz to disk, spatial resolution: 6.8-7.0 µm angular resolution: 5.2 µrad Goniometer p+ beam upstream downstream XY Plane 1 XY Plane 2 XY Plane 3 UV Plane 4 XY Plane 5 Crystal Granite Table 10,289mm 295 mm 416 214 9,960mm Trigger Scintillators Geoff Hall HSDT9 Sep 2013

CBC beam tests 2011 CERN H8 beam line 400 GeV/c protons beam profile APV plane CBC sensor 5 mV / division beam profile CERN H8 beam line 400 GeV/c protons 5 cm p-on-n 150 mm pitch 320 µm thick 64 strips bonded to 5 cm p-on-n sensor no pitch adaptor 150 µm pitch 320 µm thickness fan shaped Geoff Hall HSDT9 Sep 2013

CBC performance in beam Successful operation Digital logic works well no pipeline errors no CBC errors in > 30M events better than pitch/√12 due to contribution from 2 strip clusters use telescope to select events at CBC module single track events only (pileup eliminated) incident on CBC sensor (transverse to strips) incident in 3mm along strips (const p=134um) events within 7ns of sampling clock close to binary resolution from 1 strip clusters measure resolution of CBC module from residual using telescope for track extrapolation factoring out telescope spatial resolution resolution: 29.4 um Geoff Hall HSDT9 Sep 2013

CBC -> CBC2: New features 5 mm CBC -> CBC2: New features inter- chip signals DC-DC 250µm pitch C4 layout aim for commercially assembled module some gains in bond inductance back edge wire-bond pads for wafer probe 254 channels for 127 + 127 strips correlation logic for stub formation between top & bottom strips vetoes wide clusters Test pulse no time to implement on CBC & other minor circuit improvements Improved DC-DC (CERN) received Jan 2013 – fully functional 254 inputs pipeline + buffering 11 mm CWD, offset correction and colleraltion logic 254 amplifier/comparator channels bias gen. bandgap inter- chip signals LDO Geoff Hall HSDT9 Sep 2013

CBC2 C4 wafers (shared with RAL XFEL ASIC) reticle notch wafer name: A4PNFAH Geoff Hall HSDT9 Sep 2013

Stub finding Logic Stubs shift register Individual mask for noisy channels →254b from I2C reg. (can be also used to inhibit coincidence logic) Need to be able to inhibit stub shift register operation →1b EN from I2C reg. 254-OR of channel outputs to signal any activity on chip 127-OR of stubs to control the stubs SR readout latch @40MHz

Stub logic features Cluster width Offset correction and correlation exclude clusters wider than 3 strips Offset correction and correlation programmable window, selects pT up to +/-8 channels programmable offset, adjust lateral displacement up to +/-3 channels p = ∞ IP R-f view offset n n+1 n-1 1/2/3 strip cluster on channel n channel comparator outputs Geoff Hall HSDT9 Sep 2013

first wafer probed manually selection of chips for module assembly Geoff Hall HSDT9 Sep 2013

final yield for 1st wafer bad chip 112 reticles 108 good chips 4 bad chips bad chips due solely to physical damage from probe card CBC2 reticle no defective channel found on any of 112 chips tested on this first wafer => 100% yield perhaps not too surprising if overall wafer yield high CBC2 is relatively small area of reticle & significant fraction of CBC2 area not occupied by active circuitry supply current - all chips Geoff Hall HSDT9 Sep 2013

2S module Development with CMS team Substrate development mainly by CERN Hybrid procured and assembled commercially First version: 2 chip hybrid electrical validation mount capacitors here to simulate interstrip capacitance small pitch adapter board (wire-bond pitch to 1.27mm connector) calibrated charge injection (both sensor layers) Geoff Hall HSDT9 Sep 2013

Result with test pulse 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 channel on chip on layer 1 on layer 2 8 test groups Proves basic functionality but need real data for better test Geoff Hall HSDT9 Sep 2013

2S mini-module Prototype assembled with two sensors Now under test in lab Geoff Hall HSDT9 Sep 2013

scope in persistence mode Sr-90 source scintillator signal beta source hits in the data stream data frame width CBC2 trigger output scope in persistence mode Sr-90 source scintillator signal Geoff Hall HSDT9 Sep 2013

scintillator/PM signal CBC2 triggered data output hits in the data stream CBC2 trigger output scintillator/PM signal Geoff Hall HSDT9 Sep 2013

=> track passed through strips directly above each other 50 ns CBC2 triggered data output 2 consecutive hits => track passed through strips directly above each other 50 ns Geoff Hall HSDT9 Sep 2013

single strip clusters offset by 1 strip 75 nsec Geoff Hall HSDT9 Sep 2013

single strip clusters offset by 3 strips 175 nsec Geoff Hall HSDT9 Sep 2013

single strip clusters offset by 4 strips 225 nsec Geoff Hall HSDT9 Sep 2013

2 strip cluster in one plane correlates with 1 strip cluster in the other Geoff Hall HSDT9 Sep 2013

2 strip cluster in one plane correlates with 1 strip cluster in the other Geoff Hall HSDT9 Sep 2013

2 strip cluster in lower sensor correlates with 1 strip cluster 375 nsec 2 strip cluster in lower sensor correlates with 1 strip cluster in the other, offset by 7 strip Geoff Hall HSDT9 Sep 2013

CBC3 - the “final prototype” 2 strip cluster centred on n and n+1 next version of chip should incorporate all features required for HL-LHC final choices for front end ½ strip cluster resolution 2 strip cluster position assigned to mid-point stub data definition 8 bits address (for ½ strip resolution) of cluster in bottom layer 5 bit bend information address of correlating cluster in top layer stub data formatting & transmission to concentrator 13 bit / stub, up to 3 stubs/BX => 39 bits +1 bit unsparsified L1 triggered readout data => 40 bits / 25 nsec e.g. 10 lines at 160 Mbps (per chip) other useful features e.g. slow ADC to monitor bias levels … n 1 or 3 strip cluster centred on channel n n-1 127 5 bits to describe correlating cluster address in top layer window top bottom 8 bits to describe cluster address in bottom layer CBC CBC S1 B1 S2 B2 S3 B3 R 25 ns CBC data to concentrator CBC CBC CBC 25 ns CBC 10 lines / CBC CBC CBC concentrator Geoff Hall HSDT9 Sep 2013 33

2S-Pt prototype module studies 2S-Pt final module studies Rough road map CBC1 - analogue front end in 130nm wirebonded binary logic - L1 triggered readout only, non-sparsified CBC2 C4 bump-bonded full hit correlation logic L1 triggered non-sparsified readout, fast trigger OR CBC3 full readout architecture defined additional correlation logic CBC4 optimisation final version 2011 2012 2014 2016 2S-Pt prototype module studies 2S-Pt final module studies start production 2017 2015 2013 Geoff Hall HSDT9 Sep 2013

Summary & conclusions Two successful iterations of new outer Tracker ASIC First prototype version of 2S module in hand functions well in lab environment first beam tests foreseen late 2013 Road map for future developments detailed schedule depends on complete upgrade plan and HL-LHC approval process Geoff Hall HSDT9 Sep 2013

Backup slides

CBC power features Geoff Hall HSDT9 Sep 2013 DC-DC Vout vs. load DC-DC switched capacitor converter (CERN) converts 2.5 -> ~ 1.2 works well: ~ 90% efficiency but switch noise produces difference between internal and external grounds => interference depending on CEXT improved circuit on CBC2, and bump-bonding should help Cf CEXT GNDINT LDO dropout GNDEXT 40 mV vnoise LDO linear regulator provides clean, regulated rail to analog FE (uses CERN 130 nm bandgap) ~ 1.2 Vin, 1.1 Vout dropout ~ 40 mV for 60 mA load provides > 30dB supply rejection up to 10 MHz Geoff Hall HSDT9 Sep 2013

digital logic no correlation/trigger logic in this iteration - triggered readout only (~270kHz max sustained) hit detect logic in two modes – single/variable output from each channel stored in a 256 sample deep pipeline RAM (6.4us latency max) 32 trigger FIFO buffer unsparsified readout (140-to-1 shift register) comparator output 40MHz clock hit detect - variable mode hit detect - single mode

CBC performance in beam cluster width, resolution and efficiency will be affected by threshold setting - expecting to operate at 1fC level (~640mV) important to get rough estimations for tuning stub/tracklet simulations study of noise occupancy with threshold required 2 strip clusters originate from interstrip region efficiency drop due to charge sharing

CBC2 architecture nearest neighbour signals FE amp comp. pipeline shift reg. 4 4 11 11 254 40 MHz diff.clock vth 256 deep pipeline + 32 deep buffer fast control T1 trigger fast reset test pulse I2C refresh vth cluster width discrimination stub shift register vth offset correction & correlation all signals in blue are single-ended -only travel short distance on hybrid 1 chan. mask vth pipe. control test pulse bias gen. OR_stubs trig’d data out stub shift reg. O/P OR_254 Ck trigger O/P 4 4 11 11 slow control I2C reset nearest neighbour signals front end, pipeline, L1 triggered readout, biasing ~ same as prototype (some bug fixes) twice as many channels new blocks associated with Pt stub generation channel mask: block problem channels (not from L1 pipeline) cluster width discrimination: exclude wide clusters > 3 offset correction and correlation: correct for phi offset across module and correlate between layers stub shift register: test feature - shift out result of correlation operation at 40 MHz trigger O/P: in normal operation 1 bit per BX indicates presence of high Pt stub test pulse charge injection to all channels (8 groups of ~32), programmable timing and amplitude

@160MHz (@160MHz) From F.Vasey: Electronic System for 2S-Pt modules System Architecture and Data Formats, CMS Tk Week

42

Comparison logic Modules are flat, not arcs Compensate for Lorentz drift Orientation of module => position dependent logic d z offset h dependent search window to allow for luminous region and quantization => 3 pixels (if not tiny) p = ∞ IP R-f view ~200µm ~12mm h = 2.5 R-z view Luminous region (large!) Family of modules with offsets in z Geoff Hall HSDT9 Sep 2013