Pixelated 3D sensors for tracking in radiation harsh environments

The High Luminosity upgrade of the CERN Large Hadron Collider will be able to reach a peak instantaneous luminosity of 5E34/cm2 s. The innermost detectors of the CMS and ATLAS experiments will have to cope with unprecedented requirements on radiation hardness. At the end of the operation period, radiation levels are expected to reach values above 2.6E16 neq/cm2. Sensors based on 3D pixel technology, with intrinsic radiation tolerance, are the baseline option for the innermost layers of the vertex detectors of several HL-LHC experiments. This article gives an overview of the ongoing characterization of the pixelated 3D sensor technology, their performance and their current development status for tracking on radiation harsh environments.


Introduction
The High Luminosity Large Hadron Collider (HL-LHC) project [1] is going to push their experiments to very stringent radiation environments. The innermost detectors of the largest HL-LHC experiments, ATLAS and CMS, will have to cope with unprecedented radiation levels and with extremely high track densities. Their upgraded vertex detectors will have to be able to operate in a hostile radiation environment, reaching hadron fluences up to about 2.6 × 10 16 n eq cm −2 before being replaced [2,3]. Sensors based on 3D pixel technology are excellent candidates because of their inherent radiation tolerance.
A 3D sensor is built by inserting highly doped columns into the bulk to form the P-N junctions, decoupling in that way the inter-electrode distance from the active sensor thickness (see Fig. 1). The thickness is defining the amount of charge generated but is in the case of 3D sensors independent of the distance between the electrodes, respectively the depletion voltage. The inter-electrode distance can thus be optimized independent of the amount of charge deposited by an ionizing particle, as opposite to a planar sensor. The small interelectrode distance allows to reduce the charge carrier trapping probability after irradiation, acquire the induced charge faster, and use low voltages to fully deplete the sensor, among other benefits [4]. This low depletion voltage enables to operate the 3D sensors with an extremely low power consumption, especially compared with equivalent planar sensors.
However, the columnar structure also presents some disadvantages. The electric field is not uniform within the depleted region, and therefore the sensor response cannot be uniform either. Indeed, there are charge losses inside the electrodes and ohmic columns, as the columns themselves are not active regions. Null points between same type of electrodes will delay the created charges generated in there, as they have to diffuse first. As the inter-electrode Schematic simplified layout of a planar (left) and a 3D (right) sensors, assuming a n-in-p layer structure. Read out electrodes are bonded to the highly doped n + thin region (colored in green). Electrodes bonded to the ohmic layers p + (colored in yellow) are needed to polarize the sensor. The sensor thickness, ∆, determines how many charges are created int the sensor when a particle passes through, while the inter-electrode distance, L, defines the maximum distance a charge carrier should travel before being collected. It can be observed that in a planar sensor L = ∆, but in a 3D sensor L << ∆. distance is smaller, the capacitance of a 3D sensor is order of 3-5 times higher than an equivalent planar, and consequently a 3D sensor is noisier. The fabrication process is still in the development phase, being an expensive and time consuming process made by just a few R&D institutions or foundries (IMB-CNM in Barcelona, FBK in Trento or SINTEF in Oslo), especially compared to the manufacturing of planar sensors. There are many companies capable to produce planar sensors, and with a well-established production, even for this small-pitch detector.
Several of these drawbacks can be or have been overcome. For instance, the production processes have been simplified and improved over the past few years, in particular by using single-sided processes as opposed to the double-sided initially used for the first 3D batches [5]. The latest productions are mature enough to achieve overall good yields [6]. Another example: the response of the electric field can be uniformed by tilting the detector so that the particle passes through different electric field regions.
The 3D pixel technology has already been successfully used in some LHC experiments as position tracking detectors in radiation environments such as in the ATLAS IBL [7], the ATLAS Forward Proton (AFP) detector [8], and the CMS-TOTEM Precision Proton Spectrometer (PPS) detector [9]; proving to be a mature enough yet novel technology. For the HL-LHC, ATLAS already decided to include 3D pixel detectors with 50 × 50 µm 2 and 25 × 100 µm 2 pixel size in the innermost layer of its new silicon inner tracker [3]. At the time of this report, CMS is in the process of deciding the inclusion of this technology in the two innermost layers of the barrel and in the inner ring of the end-cap of its inner tracking detector.
The RD50 Collaboration has been involved in several activities related with the 3D pixel technology, such as their time performance in order to be used as timing detectors, the radiation hardness at extreme fluences larger than 1 × 10 17 n eq cm −2 , and the qualification and characterization of the small-pitch 3D sensor with the pre-production readout chip RD53A [10], as part of the sensor qualification process for the ATLAS and CMS Phase-2 upgrade programs. The RD50 collaboration has also played an important role to bring 3D sensors to industrial production. In this paper, a summary of the last line of work is being reported. Information and results of the other work packages can be found elsewhere [11][12][13].

Sensor characterization for the HL-LHC experiments upgrade
The inner tracker upgrade projects of ATLAS and CMS require small-pitch, radiationhard detector candidates, to be assembled with the readout electronics and mechanics projected by the collaborations. The readout chip (ROC) developed for CMS and ATLAS, the RD53A, is a demonstrator ROC for CMS and ATLAS upgrades, based on 65 nm CMOS process [10]. This is a common layout design developed by the RD53 collaboration, which included several Analogs Front-Ends (AFE) to be tested. The CMS and ATLAS ROCs prototypes, the CROC and the ITkPixV1, respectively, have finally chosen to use different AFEs: the linear AFE for CROC and the differential AFE for ITkPixV1. Fig. 2 shows a picture of the RD53A ROC with the linear and differential front-end regions highlighted, and the single chip cards assemblies used for the characterization.

Module description
The sensors used have been produced either by IMB-CNM in Barcelona (Spain) or by the FBK in Trento (Italy). In both cases, the production was made with a single-sided process on p-type substrates Silicon on Silicon (SiSi) or Silicon on Insulator (SOI). The former has finally been chosen over the SOI, as it does not need the removal of the handle-wafer for biasing, which simplifies the production process and improves the yield. FBK used 6 inches Direct Bond Wafers (DBW) with a support wafer made of 500 µm thick Czochralski (CZ) silicon, with less than 1 Ω cm resistivity. The electrodes were implanted by a Deep Reactive Ion Etching (DRIE) process, in a Float Zone (FZ) wafer of high resistivity (higher than 3000 Ω cm) of 130 µm (production with Mask Aligner lithography) and 150 µm (production with Stepper lithography) thickness. The columns have about 5 µm of diameter.
For the case of the IMB-CNM, the p-type substrates were produced on 4 inches SOI wafers or 4 inches SiSi wafers with a handle wafer of 300 µm thick CZ silicon of low resistivity. The electrodes are implanted on an active region with a nominal resistivity between 1000-5000 Ω cm and 150 µm thick. On the SiSi wafers, there is no need to remove the handle wafer for bias, as the low-resistivity wafer is conductive. The columns have a diameter of about 8 µm. Fig. 3 summarizes schematically the described sensors. The sensors were manufactured with rectangular and square geometries, matching the layouts CMS and ATLAS are considering to use. The square 50 × 50 µm 2 has an inter-electrode distance of about 35 µm, and is the easiest to produce. Furthermore, they do not suffer from the cross-talk originated in the metallic routes [14], needed to match the RD53A 50 × 50 µm 2 bond matrix pattern. However, a rectangular geometry allows to increase the resolution in the short pitch side, and therefore were considered as well. Two rectangular geometries 25 × 100 µm 2 have been tested: with 2 electrodes (2E) per pixel cell with a very small interelectrode distance of 28 µm but difficult to produce, and 1 electrode (1E), less problematic to produce but with larger inter-electrode distance. The later has been proven to be enough radiation hard, even with its 52 µm of inter-electrode distance, and therefore the 2E flavour has already been discarded. Fig. 4 summarizes all the pixel geometries used in the results presented in this report. Some sensors were also irradiated in order to emulate the damage produced by irradiation after years of operation in the experiments. Irradiations have been carried out at several facilities, with different particles and energies, in several intermediate steps until reaching fluences up to 1 × 10 16 n eq cm −2 , the expected fluence of the innermost layers of CMS or ATLAS after the two first runs of the HL-LHC in the scenario of 3000 fb −1 . Details about the irradiations and methodology can be found in Ref. [15].

Test beam characterization
Electrical, TCT, TCT-TPA, or other characterizations techniques were used to study the sensors, and results have already been extensively presented elsewhere [13,16,20]. The sensors described in the previous section were bump-bonded to the RD53A demonstrator chip and wire bonded to PCB cards as explained before, and tested under a beam of particles. Two main facilities were used: CERN [17], with hadron beams of 120 GeV/c, and DESY [18], where electrons of 5.6 GeV/c are delivered to the beam lines. Test beam experiments are useful to study the behaviour and response of a sensor under similar conditions it will face when it is mounted in the real experiment. The sensor response is measured with the help of a telescope [19], where the incident particles are identified and reconstructed. Telescope reconstructed particles are used to identify the position in the sensor under study where a hit should appear, allowing to measure hit-efficiency, charge, resolution, etc. The first important observable reported is the efficiency per pixel cell, measured for 25 × 100 µm 2 and 50 × 50 µm 2 fresh sensors. The two plots in Fig. 5 show the hit efficiency drop in the ohmic columns at the four corners when the incident particles enter perpendicular to the sensor plane. This drop is the result of the incident particle passing inside the nonactive region of the columns. However, there is no significant drop observed in the n-column at center, as the charge generated below the column (see Fig. 3) is enough to create a hit (thresholds are tuned around 1000 electrons).
The drop in the n-column appears in the irradiated sensor as consequence of two effects: thresholds are increased to deal with a noisier detector, and trapping, lowering the amount of generated charge below the column. Fresh sensors are exhibiting a 98% overall efficiency, including the ohmic column losses. When tilted, a 99.5% overall efficiency is recovered with very low bias voltages. As it can be observed from Fig. 6, fully depletion is reached with a few volts for fresh sensors, and with a few hundred of volts for sensors irradiated up to 1 × 10 16 n eq cm −2 .
The resolution of the sensor attached to the RD53A before and after irradiation up to 1 × 10 16 n eq cm −2 , has been recently measured in the short pixel direction of the 25 × 100 µm 2 geometry [22]. The sensor shows a resolution of 3 µm and there is almost no degradation after irradiation, with a resolution of 5.7 µm evaluated at the optimal angle. These results complement the previously obtained for the 50 × 50 µm 2 geometry, with a resolution of 5 µm, evaluated at the optimal angle and before irradiation.
A parameter of interest, especially important for the commissioning and operation of the modules in the targeted experiment, is the power consumption. The ATLAS specifications are requiring to be below 10 mW/cm 2 , while CMS requires less than 1 W/cm 2 (included sensor and chip consumption). The power consumption of the tested sensors after being irradiated up to 1 × 10 16 n eq cm −2 , has been measured below the ATLAS specifications in the full operational range of the sensor, defined with an overall efficiency larger than 97%. Consumption has been evaluated operating the sensors at around −25°C, and for both geometries 50 × 50 µm 2 and 25 × 100 µm 2 . Fig. 8 summarizes the results detailed in [20,23].

Conclusions
Small-pitch 3D pixel sensors bump-bonded to the pre-production ROC RD53A have proven to have an excellent radiation tolerance up to 1 × 10 16 n eq cm −2 , keeping hit efficiency larger than 97% at particle normal incidence with operational bias voltages below 200 V. This allows to fulfill the power dissipation specifications required by the ATLAS and CMS detectors.
Both geometries being considered, 25 × 100 µm 2 and 50 × 50 µm 2 , show similar performance, in particular the 25 × 100 µm 2 with 1 electrode is able to deal with the radiation tolerance demanding conditions, therefore there is no need for the 2 electrodes design. Their performance meets the baseline requirements to be qualified for the HL-LHC inner tracker upgrades of ATLAS and CMS.
A new set of test beam measurements is being planned for samples already irradiated to 2 × 10 16 n eq cm −2 , while further irradiation are planned to reach the expected fluences for the LS5, i.e. 2.6 × 10 16 n eq cm −2 , when is planned the replacement of the modules of the innermost layers of CMS. New productions of sensors bump-bonded to the experiment-dedicated ROCs, the ItkPixV1 (ATLAS) and the CROC (CMS) have already started.