ITk Strips Barrel Modules Overview

• ITk Strips Barrel Modules are silicon microstrip detector units used in the ATLAS Inner Tracker upgrade, integrating sensors, electronics, adhesives, and stringent QC measures.
• They employ both short- and long-strip designs with precise geometries and readout configurations to enable effective 3D reconstruction and maintain low noise levels.
• Innovations such as reduced thermal cycling ranges and interposer use have mitigated thermo-mechanical stresses and cold noise, improving reliability under HL-LHC conditions.

Searching arXiv for recent and foundational papers on ITk strip barrel modules, thermal cycling, QC, sensor behavior, and readout ASICs. ITk Strips Barrel Modules are the square silicon microstrip detector modules used on the barrel staves of the ATLAS Inner Tracker (ITk) for the High-Luminosity LHC upgrade. They are the fundamental electromechanical units of the strip barrel, combining a silicon sensor, one or two readout hybrids, a powerboard, adhesive interfaces, and extensive wirebonding into a module that must satisfy strict geometrical and electrical criteria before integration onto local support structures. Their technical development has been defined not only by radiation-hard sensor and frontend design, but also by distributed quality control, thermo-mechanical reliability studies, and later mitigation of deformation, sensor fracture, and cold-noise phenomena (Liang, 2018, Tishelman-Charny, 2024, Salami et al., 5 Mar 2025).

1. Role within the ITk strip barrel

The ITk strip detector is the outer silicon microstrip component of the upgraded ATLAS Inner Tracker. In the design described for the Phase-II detector, it comprises four barrel layers within $|z|<1400~\mathrm{mm}$ and twelve endcap disks, covering a total sensitive area of approximately $165~\mathrm{m}^2$. Barrel modules are mounted on double-sided staves, with strips arranged at a stereo angle to enable 3D space-point reconstruction. Each stave holds 14 modules per side and is built around a carbon fibre honeycomb/foam core with embedded cooling pipes, carbon-fibre face-sheets, a co-cured kapton bus tape, and End-of-Structure cards for services and communication (Liang, 2018).

Two barrel module variants are used to balance occupancy against channel count. The inner barrel layers use short strips, while the outer barrel layers use long strips.

Barrel variant	Strip length	Readout layout
Short-strip (SS)	$24.1~\mathrm{mm}$ or $24.16~\mathrm{mm}$	Two hybrids, four strip sections
Long-strip (LS)	$48.2~\mathrm{mm}$	One hybrid, two strip sections

The module concept is therefore inseparable from the stave concept: the module supplies the sensing, amplification, discrimination, and local power-management functions, while the stave supplies precision placement, cooling, service distribution, and low-mass structural support. This modular decomposition is central to the large-scale assembly strategy for the strip barrel (Liang, 2018).

2. Sensor technology, frontend architecture, and local response geometry

The barrel sensor technology selected for ITk strips is $n^+$-in-$p$ float-zone silicon. The choice is motivated in the design studies by electron collection and the absence of radiation-induced type inversion. Full-size ATLAS18 short-strip sensors have a strip pitch of $75.5~\mu\mathrm{m}$, strip length $2.4~\mathrm{cm}$, four segments of $1280$ strips, and a $165~\mathrm{m}^2$0 thick $165~\mathrm{m}^2$1-type substrate with p-stop isolation, poly-silicon bias resistors, AC/DC pads, and punch-through protection structures. Earlier barrel prototypes such as ATLAS12 used a nominal pitch of $165~\mathrm{m}^2$2 and thickness $165~\mathrm{m}^2$3, establishing the sensor architecture later carried into production-oriented layouts (Fernandez-Tejero et al., 2024, Poley et al., 2016).

At module level, the electronics are mounted directly on kapton flex circuits bonded to the sensor with electronics-grade epoxy. A short-strip module tested in beam consisted of a single $165~\mathrm{m}^2$4-in-$165~\mathrm{m}^2$5 silicon sensor of $165~\mathrm{m}^2$6 thickness, two hybrids, and a powerboard; each hybrid carried $165~\mathrm{m}^2$7 ABCStar chips and one HCCStar. ABCStar provides $165~\mathrm{m}^2$8 binary-output channels per chip, with per-strip preamplifier and discriminator stages, buffering, event building, and cluster finding. The frontend is optimized for $165~\mathrm{m}^2$9 BX containment, approximately $24.1~\mathrm{mm}$0, and the tested operating mode included 01X suppression, requiring a hit in the triggered BX and zero in the previous BX (Arling et al., 2023, Liang, 2018).

A persistent theme in barrel-module studies is that the effective responding area of a strip is not set solely by the nominal strip pitch. High-resolution X-ray and particle-beam measurements showed that in bond-pad regions the responding width can increase to approximately $24.1~\mathrm{mm}$1 for a strip with a bond pad, while adjacent strips narrow to approximately $24.1~\mathrm{mm}$2, producing a $24.1~\mathrm{mm}$3 variation in cluster counts relative to unmodified regions. In the earlier ATLAS12 bond-pad studies, the effective sensitive width was found to follow the local geometry around p-stops and bond pads rather than the nominal implant pitch. Subsequent measurements and TCAD-supported interpretation attributed the dominant effect to the bond-pad geometry, with p-stop modifications alone producing only minor changes. About $24.1~\mathrm{mm}$4 of the strip length is in bond-pad regions and about $24.1~\mathrm{mm}$5 in architecture-modified regions more broadly, so the effect is localized rather than global, but it is relevant for simulation and reconstruction (Poley et al., 2016, Poley et al., 2016).

3. Assembly sequence and distributed quality control

The ITk strip module programme is organized as a globally distributed production effort. Approximately $24.1~\mathrm{mm}$6 strip modules in eight geometries are assembled and tested at $24.1~\mathrm{mm}$7 institutes on four continents, with identical procedures across sites apart from additional stitch-bonding for some multi-sensor endcap modules. For barrel modules, the standardized assembly path proceeds from hybrid preparation to full module qualification (Tishelman-Charny, 2024).

Hybrid assembly begins with ASIC gluing onto the hybrid flex, followed by weighing, metrology, wirebonding, electrical test, and burn-in. The metrology criteria cited for hybrid assembly are ASIC placement within $24.1~\mathrm{mm}$8 in $24.1~\mathrm{mm}$9, glue height between $24.16~\mathrm{mm}$0 and $24.16~\mathrm{mm}$1 with nominal $24.16~\mathrm{mm}$2, and tilt angle below $24.16~\mathrm{mm}$3. Electrical acceptance at this stage uses gain and noise thresholds: a channel is bad if gain is outside $24.16~\mathrm{mm}$4–$24.16~\mathrm{mm}$5, noise outside $24.16~\mathrm{mm}$6–$24.16~\mathrm{mm}$7 ENC, or gain exceeds $24.16~\mathrm{mm}$8 from the mean; an ASIC is bad if gain $24.16~\mathrm{mm}$9 or noise $48.2~\mathrm{mm}$0 ENC. Burn-in lasts $48.2~\mathrm{mm}$1 hours, with a hybrid failing if more than $48.2~\mathrm{mm}$2 of channels fail or a given channel is bad in more than $48.2~\mathrm{mm}$3 of tests (Tishelman-Charny, 2024).

Module assembly then adds the HV-tab, sensor I-V qualification, hybrid and powerboard gluing, final strip wirebonding, metrology, electrical characterization, and thermal cycling. The HV-tab stage requires no breakdown above $48.2~\mathrm{mm}$4, preserving margin relative to the nominal $48.2~\mathrm{mm}$5 operating point and the $48.2~\mathrm{mm}$6 end-of-life flexibility requirement. Full module qualification includes I-V scans up to $48.2~\mathrm{mm}$7 with $48.2~\mathrm{mm}$8 current measurements averaged at each voltage step, timing and pedestal extraction, injected-charge noise measurements, and channel-quality criteria. A module fails if more than $48.2~\mathrm{mm}$9 of channels are bad or if eight consecutive channels are bad (Tishelman-Charny, 2024).

Thermal cycling is itself part of quality control rather than a separate reliability campaign. In the thermal-cycling setup, modules are screwed onto aluminum chucks in a cold box, temperatures are monitored with thermistors, and shape metrology is performed on an approximately $n^+$0 grid over the sensor surface. The original electrical characterization at the temperature extrema included I-V measurements for early breakdown and leakage current, with a pre-irradiation specification of less than $n^+$1 at $n^+$2. The resulting metrology data are condensed into a shape coefficient derived from the out-of-plane profile (Salami et al., 5 Mar 2025).

4. Thermal cycling, module bow, and thermo-mechanical stress

The most consequential reliability issue identified for barrel modules in pre-production was not an intrinsic sensor defect, but a deformation induced by the quality-control procedure itself. Under the original ten-cycle protocol between $n^+$3 and $n^+$4, modules that initially satisfied the shape specification often showed permanent post-cycling bow. The shape coefficient was defined as

$n^+$5

with the accepted pre-installation range given as between $n^+$6 and $n^+$7. After cycling to $n^+$8, the average shape coefficient increased by about $n^+$9, or approximately $p$0 in the paper’s quantitative summary, and $p$1 of modules exceeded the allowed threshold post-cycling (Salami et al., 5 Mar 2025).

The observed bow was bowl-like for positive shape coefficient and hill-like for negative shape coefficient. Direct bake tests strengthened the causal link to elevated temperature: holding a module at at least $p$2 produced permanent increases such as $p$3 at $p$4 and $p$5 at $p$6. Metrology excluded correlations with pre-assembly sensor bow and with the loading process itself, directing attention to the multilayer adhesive structure of the module (Salami et al., 5 Mar 2025).

The thermo-mechanical mechanism was traced to the epoxy Eccobond F112 and to differential thermal expansion between silicon and the copper/polyimide flexes. The manufacturer-quoted glass-transition temperature was $p$7 for standard cure, but differential scanning calorimetry on room-temperature-cured material, which matched production conditions, gave an actual $p$8. During hot cycling, locally powered flex regions could exceed $p$9 even when the chuck was held at $75.5~\mu\mathrm{m}$0, so the adhesive approached or surpassed its actual transition regime. The resulting mismatch can be expressed as

$75.5~\mu\mathrm{m}$1

with typical coefficients $75.5~\mu\mathrm{m}$2, $75.5~\mu\mathrm{m}$3, and polyimide up to $75.5~\mu\mathrm{m}$4–$75.5~\mu\mathrm{m}$5. The interpretation advanced in the study is that glue softening and post-curing at high temperature permit a new stress state to form, which becomes locked in on re-solidification; the highest stress was localized between the hybrids and the powerboard (Salami et al., 5 Mar 2025).

This understanding led to a procedural change: the warm ceiling for module thermal cycling was reduced from $75.5~\mu\mathrm{m}$6 to $75.5~\mu\mathrm{m}$7. Under the revised range of $75.5~\mu\mathrm{m}$8 to $75.5~\mu\mathrm{m}$9, $2.4~\mathrm{cm}$0 of modules showed negligible shape change and all remained within specification; the paper reports that the permanent deformation issue was eliminated and that no post-cycling increase in stress or shape coefficient was observed thereafter. An independent extreme-cycling study of four representative LS modules supported this conclusion from a different direction: one module failed within the first ten cycles because of a pre-existing glue-on-guard-ring defect and would not have passed nominal QC, while the three survivors completed $2.4~\mathrm{cm}$1 cycles with only about a $2.4~\mathrm{cm}$2 noise increase, but cycling up to $2.4~\mathrm{cm}$3 produced deformations up to $2.4~\mathrm{cm}$4 and directly motivated adoption of the $2.4~\mathrm{cm}$5 warm limit (Tishelman-Charny et al., 2024, Salami et al., 5 Mar 2025).

5. Electrical performance, beam tests, and radiation tolerance

Beam-test measurements on an unirradiated short-strip module established the baseline electrical and tracking performance of the barrel-module architecture at both warm and cold operating points. At $2.4~\mathrm{cm}$6, the mean input noise was approximately $2.4~\mathrm{cm}$7, corresponding to $2.4~\mathrm{cm}$8; at $2.4~\mathrm{cm}$9 it was approximately $1280$0, or $1280$1. The warm-to-cold increase was approximately $1280$2, consistent with circuit-model expectations. The most probable cluster charge was essentially unchanged with temperature, with corrected values of $1280$3 at $1280$4 and $1280$5 at $1280$6, in very good agreement with Allpix-Squared/GEANT4 simulations. The measured signal-to-noise ratio was $1280$7 at $1280$8 and $1280$9 at $165~\mathrm{m}^2$00, while detection efficiency remained above $165~\mathrm{m}^2$01 for thresholds up to approximately $165~\mathrm{m}^2$02 at both temperatures and noise occupancy stayed well below $165~\mathrm{m}^2$03. Delay-scan reconstruction yielded a pulse FWHM of $165~\mathrm{m}^2$04, matching the $165~\mathrm{m}^2$05 circuit-model prediction (Arling et al., 2023).

The irradiation programme supports the view that this performance envelope has sufficient margin for HL-LHC operation. Prototype and production-oriented studies reported that at approximately $165~\mathrm{m}^2$06 bias the collected signal charge of irradiated sensors drops by about a factor of two at the relevant HL-LHC fluence, but remains sufficient for high-efficiency tracking. Test-beam and laboratory measurements on irradiated modules yielded post-irradiation $165~\mathrm{m}^2$07 and efficiency above $165~\mathrm{m}^2$08 at thresholds compatible with noise occupancy below $165~\mathrm{m}^2$09 per channel (Liang, 2018).

Full-size ATLAS18 short-strip sensors with final layout were irradiated over the range from $165~\mathrm{m}^2$10 and $165~\mathrm{m}^2$11 up to $165~\mathrm{m}^2$12 and $165~\mathrm{m}^2$13, including the $165~\mathrm{m}^2$14 safety factor. Pre-irradiation full depletion occurred at $165~\mathrm{m}^2$15, satisfying the specification of below $165~\mathrm{m}^2$16. At fluences above $165~\mathrm{m}^2$17, the depletion voltage rose to above $165~\mathrm{m}^2$18, but leakage current, inter-strip capacitance, inter-strip resistance, coupling capacitance, bias resistance, and punch-through protection behavior all remained within ATLAS specifications throughout the studied range, and no breakdown was observed to $165~\mathrm{m}^2$19 after irradiation. A non-trivial feature of the heavily irradiated sensors was the strong frequency dependence of bulk-capacitance measurements: only low-frequency C-V probes traced the deep-trap response and the true depletion point, whereas high-frequency capacitance remained flat with voltage. The study’s interpretation was that this does not degrade frontend noise because the tracker readout is sensitive only to the high-frequency capacitance (Fernandez-Tejero et al., 2024).

6. Cold noise, sensor fracturing, and interposer-based mitigation

A second major reliability issue emerged specifically in short-strip barrel modules operated below about $165~\mathrm{m}^2$20: so-called Cold Noise (CN). CN appeared as localized clusters of strips with very high noise, spatially correlated with regions overlapping the powerboard. The underlying mechanism was correlated with mechanically induced vibrations, primarily from the DC-DC converter, transmitted through a glued flexible PCB into the silicon sensor. In the beam-based performance study of CN-affected modules, the effect did not reduce the collected charge itself, but instead narrowed or eliminated the threshold operating window in which both efficiency and noise-occupancy requirements could be satisfied. In the non-irradiated module, high-CN regions had fewer than $165~\mathrm{m}^2$21 failed strips; in the irradiated module, the failed-strip fraction rose to about $165~\mathrm{m}^2$22 in low-CN regions and about $165~\mathrm{m}^2$23 in high-CN regions. The study concluded that CN can be tolerated before irradiation by raising thresholds to around $165~\mathrm{m}^2$24, but becomes a serious lifetime issue in high-fluence barrel locations (Affolder et al., 8 Jan 2026).

The principal design mitigation for both thermo-mechanical bow and CN was the introduction of an interposer between the sensor and the overlying flex electronics. In stand-alone pre-production module tests, the interposer stack was composed of a $165~\mathrm{m}^2$25 silicone gel layer (DOWSIL SE-4445) and a $165~\mathrm{m}^2$26 Kapton film, forming a Flex $165~\mathrm{m}^2$27 Silicone $165~\mathrm{m}^2$28 Kapton $165~\mathrm{m}^2$29 Epoxy $165~\mathrm{m}^2$30 Sensor stack-up. Detailed quality-control measurements found no adverse impact on adhesion, no increase in leakage, no increase in noise or cold-noise phenomena, and only a minor increase in on-module temperatures, which remained within acceptable limits. Under extended cycling, non-interposer modules exhibited a mean permanent bow increase of $165~\mathrm{m}^2$31 when the maximum cycling temperature was raised from $165~\mathrm{m}^2$32 to $165~\mathrm{m}^2$33, and four of five fractured after cycling to $165~\mathrm{m}^2$34. By contrast, interposer modules showed no significant bow change, $165~\mathrm{m}^2$35, after ten cycles between $165~\mathrm{m}^2$36 relative to $165~\mathrm{m}^2$37, and two modules completed $165~\mathrm{m}^2$38 thermocycles up to $165~\mathrm{m}^2$39 without fracturing (Fomin et al., 16 Jul 2025).

The same mitigation translated to module-on-stave conditions. In tests of barrel modules loaded onto local support structures, an in-built additional kapton layer proved highly effective: $165~\mathrm{m}^2$40 of $165~\mathrm{m}^2$41 modules survived testing down to $165~\mathrm{m}^2$42, and $165~\mathrm{m}^2$43 modules survived $165~\mathrm{m}^2$44 thermal cycles to $165~\mathrm{m}^2$45 with no degradation in I-V or noise performance. Earlier alternative mitigations, such as a stiffer adhesive or increased hybrid–powerboard gap, shifted the failure onset but did not deliver comparable headroom. The interposer result therefore altered the engineering interpretation of the module problem: what initially appeared as a procedural issue in thermal cycling became a more general question of stress transmission through a rigid adhesive path, and the compliant interposer addressed that path directly (D'Amen et al., 25 Aug 2025).

Taken together, the interposer studies suggest a unification of two previously separate failure classes. Thermally induced bow and fracture were linked to coefficient-of-thermal-expansion mismatch, while CN was linked to vibration transfer through the glued flex path; both are reduced by mechanically decoupling the flex electronics from the silicon sensor (Fomin et al., 16 Jul 2025, Affolder et al., 8 Jan 2026).

7. Production diagnostics and readout evolution

As production matured, barrel-module quality assurance expanded beyond conventional metrology and noise screening to include failure modes that can bias the diagnostic system itself. One example is the silicon pinhole defect, in which the dielectric between strip implant and metal electrode is breached, creating a DC connection. Pinhole formation was associated most strongly with wirebonding damage, particularly wedge contact without wire and rebonding after bond failure, but could also be introduced by sensor cracking. On completed modules, pinholes affect the AMAC leakage-current measurement by providing a current path from the ABC input into the HV return even at zero bias. The practical consequence is that I-V curves with powered ABCs can become flat, offset, or saturated, potentially masking early breakdown. The adopted procedural modifications were to power down the ABCs during IV scans and, when needed, use a baseline at $165~\mathrm{m}^2$46 rather than $165~\mathrm{m}^2$47. Additional localization methods included AMAC feedback-resistance changes, chip-by-chip BVREF scans, light-induced per-channel gain tests, and optical microscopy. With these modifications, the study concluded that pinholes do not impede module testing or performance (Affolder et al., 30 Sep 2025).

A different production-level issue arose in the ABCStar ASIC itself. During wafer testing, yields that were expected to be about $165~\mathrm{m}^2$48 good chips fell to as low as $165~\mathrm{m}^2$49 on some wafers because of a timing issue in the synthesized logic controlling re-used SRAM blocks in the L0 Buffer and Event Buffer. The failure was sensitive to process variation, core voltage, and clock duty cycle. The principal mitigations were to raise the core operating voltage from $165~\mathrm{m}^2$50 to $165~\mathrm{m}^2$51 and to increase the effective high phase of the clock seen by the ABCStar, implemented in practice by swapping the positive and negative wires of the SLVS differential clock. On the worst wafers, the voltage change alone raised yields from about $165~\mathrm{m}^2$52 to above $165~\mathrm{m}^2$53; the final Category A (+B) yield reached approximately $165~\mathrm{m}^2$54 ($165~\mathrm{m}^2$55). Validation covered temperature down to $165~\mathrm{m}^2$56, irradiation up to $165~\mathrm{m}^2$57, and multiple digital stress tests, so the mitigation was accepted without redesign or foundry-process changes (Ashmanskas et al., 21 May 2026).

These developments show that the barrel module is not a static sensor-plus-electronics assembly but a tightly coupled production object whose acceptable operation depends on the interaction of sensor geometry, adhesive mechanics, support-structure interfaces, high-voltage diagnostics, and ASIC timing margins. The module programme therefore evolved through iterative closure between beam tests, metrology, wafer probing, thermal cycling, and stave integration, with each stage feeding back into the final qualified design (Tishelman-Charny, 2024, Ashmanskas et al., 21 May 2026).