Intermediate Silicon Tracker (INTT)

• Intermediate Silicon Tracker (INTT) is a two-layer barrel silicon strip detector that delivers precise spatial and timing measurements for robust track reconstruction.
• It bridges the fine-granularity MVTX and the outer TPC, improving pattern recognition and correctly associating tracks to the corresponding RHIC bunch crossings.
• Beam tests and commissioning demonstrate high hit efficiency (>99%), excellent timing resolution, and reliable performance in dense heavy-ion collision environments.

The Intermediate Silicon Tracker (INTT) is the silicon-strip middle layer of the sPHENIX tracking system at the Relativistic Heavy Ion Collider, positioned between the innermost MVTX pixel detector and the outer Time Projection Chamber (TPC). It is a two-layer barrel silicon strip tracker whose principal functions are to provide two spatial measurement points per charged track, improve pattern recognition and global track reconstruction robustness, and supply timing information precise enough to associate reconstructed tracks with the correct RHIC bunch crossing, thereby reducing out-of-time pileup. Together with the sPHENIX full barrel calorimeter system, the broader tracking system enables heavy-flavor jets and upsilon-state identification (Shih, 4 Aug 2025).

1. Placement in the sPHENIX tracking system

Within sPHENIX, the tracking system consists of the MVTX, the INTT, the TPC, and the TPOT. The INTT occupies the intermediate radial region of this stack, surrounding the collision point azimuthally at about $10$ cm from the beam line and bridging the gap between the fine-granularity MVTX and the large-radius TPC. In detector hierarchy, it is the only sPHENIX tracking detector with sufficiently good timing to associate reconstructed tracks with the correct RHIC bunch crossing (Shih et al., 31 Aug 2025).

The detector is therefore both a geometric and a temporal bridge. Geometrically, it provides two spatial points per track candidate so that tracks can be linked from the inner pixel system into the TPC more reliably. Temporally, its fast strip readout permits bunch-crossing assignment and helps eliminate pile-up events caused by misidentifying bunch crossings. This dual role is central in the sPHENIX environment, where dense heavy-ion events and overlapping collisions make pattern recognition and event association nontrivial (Shih, 4 Aug 2025).

A common simplification is to describe the INTT merely as an additional silicon layer. That description is incomplete. The supplied timing information is one of its defining functions, and the data block states explicitly that this capability is crucial for associating reconstructed tracks with individual RHIC bunch crossings and for reducing out-of-time pileup (Shih et al., 31 Aug 2025).

2. Barrel geometry, ladders, and silicon sensors

The INTT is a barrel-type detector with two cylindrical layers and a total of $56$ silicon ladders: $24$ in the inner barrel and $32$ in the outer barrel. The corresponding radii are about $7.5$ cm for the inner barrel and $10$ cm for the outer barrel, and adjacent ladders are staggered in $\phi$ to avoid azimuthal dead space and provide full azimuthal coverage (Shih et al., 31 Aug 2025).

Each ladder contains Type-A and Type-B silicon sensors, FPHX readout chips, HDI cables, and a carbon fiber composite stave for support and heat removal. A full ladder has $6{,}656$ channels and an active area of $45.6 \times 2$ cm$^2$. In the more detailed ladder description, an INTT ladder is built from $56$0 silicon sensors, $56$1 FPHX chips, $56$2 HDI, and $56$3 carbon-fiber composite stave per module, with two modules mounted longitudinally on the stave to form one ladder; each module has $56$4 channels, giving $56$5 channels per ladder (Akiba et al., 12 Mar 2025).

The silicon sensors are two single-sided, AC-coupled strip sensor types, both fabricated by Hamamatsu and based on FVTX technology. They are p-implant sensors on a $56$6m thick n-type substrate with strip pitch $56$7, strip implant width $56$8, readout aluminum width $56$9, biasing through $24$0 polysilicon resistors, and nominal operating voltage $24$1 V. Type-A sensors have physical size $24$2 mm $24$3 $24$4 mm, active area $24$5 mm $24$6 $24$7 mm, segmentation $24$8 blocks, strip length $24$9 mm, $32$0 blocks, and $32$1 channels. Type-B sensors have physical size $32$2 mm $32$3 $32$4 mm, active area $32$5 mm $32$6 $32$7 mm, segmentation $32$8 blocks, strip length $32$9 mm, $7.5$0 blocks, and $7.5$1 channels. Together, Type-A and Type-B provide $7.5$2 mm $7.5$3 $7.5$4 mm active area and $7.5$5 channels per module. The active fractions are $7.5$6 for Type-A and $7.5$7 for Type-B (Akiba et al., 12 Mar 2025).

Because the strips are much shorter in the transverse dimension than along the beam direction, the INTT has good spatial resolution in the azimuthal or polar projection, but its hit position resolution along the beam axis $7.5$8 is comparatively coarse. This is why it is not a conventional vertex detector, even though it can be used for vertex-related studies through tracklet combinations (Shih, 4 Aug 2025).

3. Front-end ASICs, interconnects, and long readout cables

The front-end ASIC is the FPHX, a custom $7.5$9-channel readout chip originally developed for the PHENIX FVTX detector and reused in sPHENIX. The chip is $10$0 mm $10$1 $10$2 mm, operated at $10$3 V, dissipates as little as $10$4 mW per chip, and outputs digitized data at up to $10$5 MHz. Architecturally, it comprises an analog front-end with integration and shaping, a $10$6-bit ADC stage, and a digital back-end using a triggerless data-push architecture. It processes up to four hits within four RHIC beam crossings, sparsifies and zero-suppresses hits, and outputs a hit record containing a $10$7-bit timestamp, a $10$8-bit channel ID, and a $10$9-bit ADC value. These words are serialized on two LVDS lines in alternating order at up to $\phi$0 MHz (Akiba et al., 12 Mar 2025).

The HDI is the flexible circuit board that connects the sensors and FPHX chips and carries power, bias, slow control, and signal routing. It is a $\phi$1-layer FPC with dimensions $\phi$2 mm $\phi$3 $\phi$4 mm in the sensor area, connector-end width $\phi$5 mm, total thickness $\phi$6m in the sensor area, $\phi$7 signal lines, $\phi$8m / $\phi$9m line and space, and designed $6{,}656$0 differential LVDS impedance. The signal layers are sandwiched between ground and power layers to suppress EMI, and the HDI includes $6{,}656$1 pairs of output data lines, $6{,}656$2 pairs of slow-control clock LVDS lines, and $6{,}656$3 calibration pulse injection lines (Akiba et al., 12 Mar 2025).

Because the detector sits inside the TPC, raw data cannot be processed nearby. Instead, signals travel from the ladder through the HDI, then through a bus extender (BEX) cable, then through a conversion cable, to the readout card (ROC) outside the TPC. The signal chain is longer than $6{,}656$4 m and must carry high-density, high-speed LVDS data while meeting constraints of length, density, flexibility, impedance control, and barrel geometry. The BEX is a specially developed $6{,}656$5 m long flexible printed circuit using liquid crystal polymer as dielectric. It has $6{,}656$6 layers, dimensions $6{,}656$7 m $6{,}656$8 $6{,}656$9 mm, connector-region width $45.6 \times 2$0 mm, total thickness $45.6 \times 2$1m, dielectric thickness $45.6 \times 2$2m, bonding sheet thickness $45.6 \times 2$3m, $45.6 \times 2$4 signal lines, and measured differential characteristic impedance $45.6 \times 2$5. At $45.6 \times 2$6 MHz, the measured insertion loss is $45.6 \times 2$7 dB and the return loss is $45.6 \times 2$8 dB. The conversion cable uses $45.6 \times 2$9-coax technology, specifically I-PEX CABLINE-UX II, with AWG $^2$0, $^2$1 differential characteristic impedance, $^2$2 harnesses per bundle, three bundles, and total length $^2$3–$^2$4 cm depending on cable type (Akiba et al., 12 Mar 2025).

Power delivery also reflects the long cable chain. The maximum current per half ladder is $^2$5 A analog and $^2$6 A digital, while the voltage drop through the readout chain is $^2$7–$^2$8 V. To compensate, the ROC regulators were upgraded to $^2$9 V analog and $56$00 V digital so that about $56$01–$56$02 V reaches the FPHX chips at the detector. The total thickness in the sensor region is $56$03–$56$04 mm and the total radiation length is $56$05 (Akiba et al., 12 Mar 2025).

4. Tracking, timing, and vertex reconstruction

The INTT provides two intermediate spatial points needed to connect short, precise MVTX tracks with longer TPC trajectories. This improves track finding, pattern recognition, global track reconstruction robustness, and overall tracking precision, especially in the dense environment of heavy-ion collisions (Shih, 4 Aug 2025).

Its timing role is more distinctive. The data block states that the INTT has a timing resolution finer than the RHIC bunch-crossing frequency and is the best timing detector in the sPHENIX tracking system. As a result, it can associate tracks with the correct bunch crossing and help eliminate pile-up events caused by misidentifying bunch crossings. A concise way to represent this detector function is

$56$06

This timing information is also important for streaming operation, where events are recorded as long as the INTT sees particle hits (Shih, 4 Aug 2025).

Although the strip geometry does not make the INTT a conventional vertex detector, the detector can reconstruct the collision $56$07-vertex using tracklets. The procedure described in the data block is: form inner–outer barrel cluster pairs, keep pairs with small azimuthal angle difference, and use each pair to infer a probability distribution for the collision point along $56$08. Because a particle can hit a strip anywhere along its length, a single tracklet gives a trapezoidal probability distribution in $56$09. Stacking many tracklets in one event yields a Gaussian-like distribution, from which the event vertex is extracted by fitting. The qualitative relation given is

$56$10

Simulation studies showed that the INTT $56$11-vertex resolution can reach $56$12 for events with more than about $56$13 clusters. The reconstructed $56$14-vertex was compared with the Minimum Bias Detector result in data, and a positive correlation with a slope close to unity was reported (Shih, 4 Aug 2025).

This establishes an important distinction. The coarse single-hit $56$15 granularity from $56$16 mm and $56$17 mm strip lengths does not preclude useful event-level vertexing. A plausible implication is that the INTT extracts substantial $56$18-vertex information from multiplicity and geometric redundancy rather than from fine per-hit longitudinal segmentation.

5. Beam-test characterization and detector performance

A beam test of preproduction INTT ladders and the readout chain was conducted at the Research Center for Accelerator and Radioisotope Science, Tohoku University, Japan. The setup used a dedicated telescope consisting of four preproduction INTT ladders placed evenly spaced inside a dark box, with trigger scintillators upstream and downstream and a coincidence of the two trigger scintillators providing the trigger. The beamline angle was $56$19, the beam particle was a positron, the momentum was $56$20, and the beam was generated by a gamma-ray beam interacting with a $56$21m tungsten production target. The test used the right halves of the ladders; one upstream ladder was excluded because of a bias-voltage issue, and the remaining ladders were labeled $56$22, $56$23, and $56$24 (Shih et al., 31 Aug 2025).

The analysis excluded $56$25 noisy channels in ladder $56$26, grouped vertically adjacent blocks into a cell, built clusters from adjacent hits in a cell, and determined cluster $56$27-position using the energy-weighted average

$56$28

For track studies, the residual for ladder $56$29 was defined using a straight-line interpolation between ladders $56$30 and $56$31: $56$32 with measured offset $56$33 mm and therefore $56$34 mm (Shih et al., 31 Aug 2025).

The signal-to-noise ratio was defined as

$56$35

where $56$36 is the most probable energy deposition of a minimum-ionizing particle and $56$37 is the RMS of the noise distribution. Because the FPHX readout has only a $56$38-bit ADC, the analysis used an $56$39-step DAC scan with overlapping threshold ranges to reconstruct the full energy-deposit spectrum. The most probable signal amplitude was $56$40 (stat.) $56$41 (syst.) DAC, the noise width was $56$42 DAC, and the resulting signal-to-noise ratio was greater than $56$43, exceeding the conventional requirement of $56$44 for silicon tracking detectors. The spectrum was fit using Landau $56$45 Gaussian for the signal and two Gaussians at the origin for the noise (Shih et al., 31 Aug 2025).

Residual distributions showed a symmetric distribution with steep tails, with the tails attributed to the low material budget. Simulation and data agreed well after applying the measured ladder alignment correction, indicating good control of the geometry and good understanding of the material model. For spatial resolution, because multiple Coulomb scattering contributes to the residual width, the paper did not extract resolution directly from the raw residual width. Instead it used the relation

$56$46

convolved the simulated multiple-scattering residual distribution with Gaussian smearing of varying width, and performed a $56$47 comparison to data using the central six bins, which contain more than $56$48 of entries. The best Gaussian smearing was $56$49, corresponding to INTT spatial resolution $56$50 (Shih et al., 31 Aug 2025).

Hit-detection efficiency was defined as

$56$51

where ladders $56$52 and $56$53 formed the track and ladder $56$54 was the detector under test. A hit was counted as detected if a cluster was found in ladder $56$55 and $56$56 mm. The integrated hit-detection efficiency was $56$57 (stat.) with $56$58 (syst.), and the efficiency stayed above $56$59 over the measured vertical range from $56$60 mm to $56$61 mm. Comparing the nominal cable configuration with the BEX configuration showed efficiencies differing by less than $56$62, supporting the reliability of the readout-cable chain (Shih et al., 31 Aug 2025).

A special multiple-track run used a $56$63 cm-thick lead plate placed $56$64 cm upstream, with thickness $56$65, to generate a particle shower. Using all $56$66 cells, the tracking algorithm looped over cluster combinations across the three ladders, chose the most linear combination by straight-line fit, accepted a candidate if $56$67 mm, removed used clusters, and repeated the search iteratively. Up to $56$68 tracks per event were reconstructed, and the measured multiplicity distribution agreed well with simulation (Shih et al., 31 Aug 2025).

6. Commissioning, operation, and relation to other inner-tracker concepts

In Au+Au running, sPHENIX began taking data in May 2023, and the INTT was commissioned shortly afterward in triggered-readout mode. A key commissioning check was the correlation between MBD total charge and INTT cluster multiplicity; a clear positive correlation was observed, showing that INTT and MBD were time-aligned, both subsystems were recording reliable data, and the detector response was behaving as expected in the heavy-ion environment. The INTT-based $56$69-vertex reconstruction was also applied to data and compared to MBD, with an observed linear correlation near unit slope (Shih, 4 Aug 2025).

In April 2024, sPHENIX began collecting $56$70 data and the INTT transitioned to streaming readout mode. This operating mode is explicitly connected in the data block to the heavy-flavor physics program, because the relevant objects are often not energetic enough to satisfy standard calorimeter triggers. The BCO-index distribution showed a highlighted trigger-associated bin and additional spikes corresponding to non-triggered collision events recorded by the streaming system, demonstrating successful streaming operation. Further commissioning checks included a positive correlation between inner-barrel and outer-barrel cluster counts, standalone INTT track reconstruction with magnetic bending curvature clearly visible in event displays, and an observed dependence of measured energy deposition on incidence angle consistent with the expectation that smaller incident angle implies longer path length through silicon and hence larger deposited energy (Shih, 4 Aug 2025).

The INTT’s role in the sPHENIX physics program follows directly from these detector functions. Heavy-flavor jets require precise tracking close to the interaction point and reliable association of tracks to the correct collision; the INTT contributes by improving track seeding and pattern recognition, providing intermediate spatial constraints, enabling timing-based event association, and supporting streaming readout. The combination of the tracking system and full calorimetry also enables upsilon-state identification. Even though its $56$71-resolution per strip is limited by the long-strip geometry, the combined barrel information and timing make it a powerful subsystem for full-event reconstruction (Shih, 4 Aug 2025).

The term “inner tracker” can invite comparison with substantially different detector philosophies. The ALICE ITS3 project, for example, is described as “the first truly cylindrical inner tracker” and is based on bent, wafer-scale monolithic pixel sensors fabricated with stitching, with layers consisting practically only of the silicon sensors and cooled by forced airflow (Yüncü, 2022). By contrast, the INTT is a two-layer barrel silicon strip detector assembled from ladders, HDIs, cables, and carbon-fiber composite staves, with $56$72 ladders total and a readout architecture centered on FPHX strip electronics (Akiba et al., 12 Mar 2025). Conceptually, both systems pursue precise tracking close to the interaction point, but the INTT represents a conventional silicon-strip intermediate tracker optimized for timing, bunch-crossing association, and bridging between MVTX and TPC, rather than a wafer-scale, bent-sensor, nearly all-silicon cylindrical barrel.