Serial powering characterisation for the CMS Inner Tracker at the High Luminosity LHC

The CMS tracking system will be completely replaced in order to operate in the extreme conditions of the High Luminosity LHC. The new tracker will consist of two main subdetectors: the Inner Tracker, containing pixel modules, and the Outer Tracker, consisting of strip and macro-pixel modules. In the Inner Tracker, silicon sensors will be read out by the CMS Readout Chip (CROC). The chip, built in 65 nm CMOS technology, is able to withstand high radiation doses (500 Mrad) and hit rates (3 GHz/cm2 on the innermost tracking layer) during operation. Moreover, it must handle a fine sensor granularity (2500 µm2 pixels) and operate at low detection thresholds (1000 e-). The Inner Tracker will make use of the serial powering scheme to provide the ∼60 kW required by ∼4000 modular units. The Shunt-LDO (SLDO) regulator is the part of the CROC responsible for providing the power by draining a constant current in time, independently of the chip power consumption. Tests performed with Single Chip Cards (SCC) in serial configurations are reported in this paper.


The CMS Inner Tracker upgrade
The Phase-2 tracker of the CMS experiment [1], foreseen for the High Luminosity phase of the Large Hadron Collider (HL-LHC), has two major systems: the Inner Tracker, closer to the collision point and comprising pixel modules, and the Outer Tracker, comprising strip and macro-pixel modules [2].The Inner Tracker has stringent design requirements, such as a radiation tolerance of at least 500 Mrad, high rate capabilities (up to 3 GHz/cm 2 ), high granularity (50 × 50 µm 2 pixel cells) and low thresholds and noise.These requirements imply a very high power consumption, of the order of 60 kW.
The new CMS pixel ReadOut Chip (CROC) is developed inside the RD53 collaboration [3], a joint CMS and ATLAS effort to design the pixel electronics of the experiment upgrades for the HL-LHC.The CROC is designed in 65 nm CMOS technology, features 1.5 × 10 5 pixel channels and consumes about 8 − 10 µA per channel.By adding the power consumption of the chip periphery, the total needed current per chip is ∼ 1.5 A, with a supply voltage of 1.2 V.

The Inner Tracker serial powering scheme
A direct powering scheme for the Inner Tracker would require large cross-section power cables, which would dramatically increase the passive material in the tracker.A possible approach, based on the present pixel tracker, is based on DC-DC converters.However, these are affected by two problems.Firstly, they are not radiation resistant enough for the Inner Tracker environment and secondly, they are large objects that are difficult to place inside the tracker, while also adding significant passive material near the collision point.While this scheme is indeed used in the Outer Tracker, where the radiation environment is lower and space is less constrained, the Inner Tracker has opted for a serial powering scheme instead [4].Figure 1 shows the main differences of the possible powering options.
The Inner Tracker is organised in serial powering chains of up to ten modules.Each module in the inner (outer) layers will have a single sensor bump-bonded to two (four) CROCs, which are usually referred to as "double" ("quad") modules."Double" modules will be installed in the two innermost tracker layers, while "quad" modules will be installed in outer layers.
The CROCs within each module are connected in parallel, as shown in figure 2. The CROC includes a highly specialized circuit that combines the functionality of a current shunt and a Low-DropOut (LDO) regulator, thus referred to as Shunt-LDO (SLDO).The SLDO ensures that power and -1 -  current consumption are kept constant, independent of hit and trigger rates.Moreover, the SLDO is designed to ensure appropriate current sharing between multiple chips, powered in parallel.Thanks to this scheme, the serial chain presents itself as a constant load to the power supplies as the SLDO manages the CROC power consumption variations.Figure 3 shows the equivalent circuit of the SLDO, amounting to an effective resistance plus an offset voltage.For low current values there is a transient region where the chip is not operational.Once a high enough current is reached, the output voltage is kept constant regardless of the input current.
The current injected in the power loop must satisfy the highest possible load current, including an extra headroom to comply with fast dynamic current variations of the digital logic of the chip.Consequently, each serial powering chain will require an injected current of up to 8.0 A (for a "quad" module), out of which 6.0 A would be consumed by the pixel chips and 2.0 A would be the extra current headroom (25%).Since it is crucial that the digital current variations do not affect the sensitive analogue front-end part of the chip, two SLDOs are integrated on the chip, one per power domain (digital and analogue).
One downside of a serial powering scheme is that it is more susceptible to failures, since a malfunction in one module propagates through the chain.Therefore a careful study of the possible failure scenarios is necessary.
-2 -  Each module in a chain works with a different local ground, since the output of a module is the input of the following one.Since the bias voltage (needed to deplete the silicon sensors) is provided with a direct powering scheme (as shown in figure 4), with the modules in a chain connected in parallel, the applied bias voltage on the silicon sensors is different through the chain.There are two independent high voltage lines for a chain of ten modules therefore, assuming a maximum possible voltage drop of 2.5 V between two modules, the bias voltage difference between two modules in a chain could reach 10 V.This could lead to performance differences along the chain for some types of pixel sensors.
Two types of pixel sensor technologies will be used for the future Inner Tracker: traditional planar pixel sensors, where the electrodes are parallel to the sensor surface, and 3D pixel sensors, where the electrodes are orthogonal to the sensor surface [5].Due to a reduced drift distance, 3D sensors feature a higher radiation resistance, therefore they will be installed in the innermost tracker layer.Another key feature of 3D sensors is that the bias voltage needed to deplete them is smaller in comparison to planar sensors.Typically the depletion voltage of 3D sensors is as small as 5 V, therefore the effect of the bias voltage drop inside a serial powering chain is more pronounced for 3D sensors.The goal of this paper is to demonstrate that the performance of 3D modules is not affected by this bias voltage drop.

3D sensor performance in a serial powering chain
In order to test the performance of a 3D module in a serial powering environment, a serial powering chain of ten modules was prepared.The modules were made by coupling a sensor to a CROC, which were then mounted onto Single Chip Cards (SCCs).The SCC is designed for testing, and is useful to check the behaviour of the chip before assembling full modules.The ten modules included a mixture of both 3D and planar pixel sensor types.Figure 5 shows a photograph of the setup.Four fans were placed near the modules in order to cool down the system.
One 3D module of the chain was taken as reference and measured under several conditions.Figure 6 shows the IV curve of the 3D module under test, where the total input current was increased from 0 A to 4 A, passing through the working point of 2 A (1 A per channel).The IV curve reports the current per (analogue or digital) domain, assuming a 50% sharing between them.It can be observed -4 -  that the input voltages (VinD and VinA for the digital and analogue domains respectively) linearly increase with the current, similar to figure 3. Several voltages are measured on the SCC during the IV scan, in particular, the LDO-regulated digital and analogue output voltages (VddD and VddA respectively), which are both measured to be ∼ 1.2 V, as expected.This IV curve was performed both inside the chain and in a standalone setup (i.e.individually powered) and no differences were observed.
The 3D module under test was first tested in a standalone setup, with a bias voltage of 30 V, and was tuned to an average pixel threshold of 1000 e − .The 3D module was then tested in three different positions of the serial powering chain.It should be noted that the 30 V of bias voltage is referred to -5 - ground (which corresponds to the last position of the chain).Therefore, the effective bias voltage of the 3D module varies on its position inside the chain.Figure 7 and figure 8 show the threshold and noise distributions of the 3D module under test for three different positions in the chain: first (in blue), sixth (in red) and tenth (in green).No significant differences can be observed in the three positions.The threshold and noise distributions have average values of 1000 e − and 100 e − respectively, compatible with values measured within the standalone setup.
In conclusion, this test showed that the performance of a 3D pixel module is not affected by the serial powering scheme.

Offset voltage sharing
The two (four) CROCs in a "double" ("quad") module have their respective offset voltages shorted together.In this way small variations between the CROCs are corrected and the current sharing between the chips is improved.However, this configuration opens a possible failure scenario -if one of the chips stops working, the other one (three) in parallel will have a lower offset voltage, and therefore an higher current is needed to turn on the chips.
This scenario was tested by switching off one CROC while keeping the other three CROCs on in parallel.Figure 9 shows the IV curve of the three CROCs in parallel, where it can be seen that the VinD/A values of the CROCs are perfectly superimposed.The ohmic behaviour of the SLDO starts at 8 A, which is the working point for a "quad" module.Therefore, the system can sustain this failure scenario.

Conclusions
In this paper two failure scenarios of the serial powering scheme of the CMS Inner Tracker were studied.It was shown that a 3D sensor in a serial powering chain does not suffer a deterioration of its performance due the different effective bias voltage.The threshold and noise distributions are compatible between different positions of the module in the chain.

Figure 1 .
Figure 1.Comparison of three different powering scheme options, with the total power consumptions.The serial powering scheme was chosen for the CMS Inner Tracker.

Figure 2 .
Figure 2. Schematic representation of a serial powering chain.The chips inside a module are powered in parallel.

Figure 3 .
Figure 3. Equivalent circuit of SLDO.The CROC is seen as an effective resistance plus an offset voltage.

Figure 4 .
Figure 4. Illustration of pixel modules in a serial powering chain.The bias voltage to deplete the silicon sensors is provided in parallel.

Figure 5 .
Figure 5. Photograph of the experimental setup, showing a serial powering chain of 10 CROC modules (with 3D or planar pixel sensors).

Figure 6 .
Figure 6.IV curve of the reference 3D CROC module under test inside the serial powering chain.The behaviour is the same as when operated in a standalone setup.Several chip parameters are monitored during the scan, such as the LDO-regulated digital and analogue voltages (VddD and VddA respectively), which are around 1.2 V (as expected).Vofs_in is half the offset voltage and VrefD/A is half of VddD/A.Two linear fits are performed for the input voltages VinD and VinA, and are perfectly superimposed.The slope is the effective resistance of the SLDO, while the intercept is the offset voltage.

Figure 7 .
Figure 7. Pixel threshold distribution in ΔV cal units (where 1 ΔV cal corresponds to approximately 5 electrons) of the 3D CROC module under test in three different positions of the serial powering chain: blue is the first position, red is the sixth position and green is the tenth position.No significant differences can be observed.

Figure 8 .
Figure 8. Pixel noise distribution in ΔV cal units (where 1 ΔV cal corresponds to approximately 5 electrons) of the 3D CROC module under test in three different positions of the serial powering chain: blue is the first position, red is the sixth position and green is the tenth position.No significant differences can be observed.

Figure 9 .
Figure 9. IV curves of three CROCs in parallel with a disabled CROC, where offset voltages are shorted together.Each colour (blue, orange and green) is referred to one of the three CROCs.The VinD/A values of all three CROCs are perfectly superimposed.The current working point for a "quad" module is highlighted.