Sensor operating point calibration and monitoring of the ALICE Inner Tracking System during LHC Run 3

This paper presents the calibration methods and monitoring strategies developed to ensure the stable operation of the ALICE ITS2, the largest-scale application of Monolithic Active Pixel Sensors (MAPS) at the LHC, which comprises 24,120 sensors and 12.6 billion pixels.

The Gist

Imagine the ALICE experiment at the Large Hadron Collider (LHC) as a massive, high-speed camera trying to take a picture of the universe's smallest building blocks smashing together. For the 2021 "Run 3" of the collider, ALICE installed a brand-new, super-sensitive inner camera called ITS2.

This camera isn't made of glass lenses; it's a giant, cylindrical blanket of 24,120 tiny silicon sensors (called ALPIDE chips) covering an area of about 10 square meters (roughly the size of a small apartment). It contains a staggering 12.6 billion pixels. To put that in perspective, if every pixel were a grain of sand, they would cover a football field several inches deep.

However, building a camera with 12.6 billion pixels is like trying to tune a piano with 12.6 billion keys. If even a few keys are out of tune, the music (the data) sounds terrible. This paper explains how the scientists "tuned" this massive camera and kept it playing in perfect harmony.

The Problem: A Billion Tiny Keys Need Tuning

Think of each pixel on the sensor as a tiny microphone listening for a whisper (a passing particle).

• The Threshold: The microphone needs to be set to a specific volume level. If it's too quiet (low threshold), it hears the wind and static (noise) and thinks it's a whisper (a fake hit). If it's too loud (high threshold), it misses the actual whispers (missed particles).
• The Goal: The scientists needed to set every single one of the 12.6 billion microphones to the exact same volume so they could hear the whispers clearly without the static.

The Solution: The "Calibration" Concert

The paper describes a series of tests (called "scans") the team performed to tune these microphones.

1. The "Digital and Analogue" Check-up
Before tuning the volume, they checked if the microphones were even plugged in. They sent a test signal to every pixel.

• Dead Pixels: Some microphones were broken and wouldn't speak at all.
• Noisy Pixels: Some were screaming static even when no one was talking.
• Stuck Columns: Sometimes, a whole row of microphones would get stuck in a loop, sending the same message over and over. The team had to identify these "broken rows" and turn them off (mask them) so they didn't clog up the system.

2. The "Volume Knob" Tuning (Threshold Scans)
This is the main event. The team used a special tool to inject a tiny, known "charge" (a simulated whisper) into the pixels.

• They slowly turned the volume knob up and down.
• They watched to see exactly at what point the pixel said, "I heard something!"
• They found that the "sweet spot" for hearing a particle without hearing noise was between 100 and 150 electrons (a tiny unit of charge).
• The Challenge: Because the LHC is a harsh environment, radiation from the collisions acts like a slow-acting poison. Over time, it changes the sensitivity of the silicon, making the microphones either too sensitive or too dull. The team had to re-tune the volume knobs roughly once a year to compensate for this "aging."

3. The "Noise" Hunt
Even with the volume set perfectly, some pixels are just naturally "crazy" and fire randomly. To catch them, the team ran the camera in a "dark room" (with no particle collisions).

• They counted how many times each pixel fired on its own.
• If a pixel fired too often (more than once in a million frames), it was labeled "noisy" and permanently silenced (masked).
• Result: They only had to silence about 0.01% of the pixels. That's like finding one bad apple in a truckload of 10,000.

The "Brain" Behind the Tuning

You might wonder: "How do you tune 12.6 billion pixels without taking a century?"
The answer is a massive computer farm.

• The Orchestra: The sensors send data to 192 "Readout Units" (like conductors).
• The Musicians: These units send data to a farm of 350 powerful computers.
• The Magic: These computers process the data while the scan is happening (on-the-fly). They calculate the perfect settings for every chip in real-time. If a problem is found, they can adjust the settings for the next run almost instantly.

Why Does This Matter?

Imagine trying to find a specific needle in a haystack, but the haystack is moving at the speed of light.

• If your "needle detector" (the sensor) is too sensitive, you think every piece of hay is a needle. You get lost in false alarms.
• If it's not sensitive enough, you miss the needle entirely.

By keeping the "volume" (threshold) perfectly calibrated and silencing the "crazy" pixels, the ALICE team ensures that when they record data, they are seeing real physics, not just electronic noise.

The Bottom Line

This paper is a success story of engineering and patience. It shows that even with a detector the size of a house, containing billions of tiny, fragile components, we can:

1. Tune them all to work together perfectly.
2. Monitor them daily to catch any drifts caused by radiation.
3. Fix them quickly if they start acting up.

Thanks to this "calibration orchestra," the ALICE experiment can continue to take the clearest, most detailed pictures of the universe's most violent collisions, helping us understand how the world began.

Technical Summary

1. Problem Statement

The ALICE experiment at the LHC upgraded its Inner Tracking System (ITS) to ITS2 for Run 3 (starting 2021) to handle the increased luminosity. The new detector utilizes 24,120 silicon Monolithic Active Pixel Sensors (ALPIDE chips), creating a massive system with 12.6 billion pixels and a total sensitive area of ~10 m².

The primary challenges addressed in this paper are:

• Scale and Complexity: Calibrating 24,120 chips and 12.6 billion pixels represents the largest-scale application of Monolithic Active Pixel Sensor (MAPS) technology in high-energy physics.
• Operating Point Stability: The detector must maintain a specific charge threshold (typically 100–150 $e^-$) to ensure >99% detection efficiency while keeping the fake-hit rate below $10^{-6}$ hits/event/pixel.
• Radiation Effects: Accumulated radiation dose (Total Ionizing Dose and Non-Ionizing Energy Loss) alters transistor properties and leakage currents, causing threshold drifts and noise increases over time.
• Power Distribution: The complex power scheme involving long cables (7m) and voltage drops requires precise compensation to ensure stable chip operation.
• Operational Constraints: Calibration must be performed rapidly (within the ~45-minute window between LHC fills) to avoid delaying physics data taking.

2. Methodology

The paper outlines a comprehensive framework for calibration, monitoring, and dynamic adjustment of the ITS2 operating parameters.

A. Calibration Scans

A suite of automated scans was developed to tune and monitor the detector:

• Digital/Analogue Scans: Identify dead, inefficient, and noisy pixels, as well as faulty double-column readout circuits.
• Threshold Scans (VCASN & ITHR):
  • VCASN: Coarse tuning of the charge threshold.
  • ITHR: Fine tuning.
  • Short vs. Full Scans: A "short" scan samples ~2.1% of pixels to determine global chip settings, while a "full" scan (all pixels) is performed annually as a reference.
  • Analysis: Instead of slow curve fitting, a numerical derivative of the hit-vs-charge S-curve is used to extract the threshold and temporal noise in ~1–2 µs per pixel.
• VRESETD Scan: Monitors the reset voltage working point, which shifts due to radiation damage.
• Pulse-Shape Scan: Measures the Time of Arrival (ToA) and Time-over-Threshold (ToT) to characterize the analogue signal response.
• Noise Scan: Performed without beam collisions to measure the fake-hit rate and generate masks for noisy pixels.

B. Computing Infrastructure

• Hardware: 192 Readout Units (RUs) connect to 22 Common Readout Units (CRUs) and 13 First Level Processors (FLPs).
• Processing: Data is processed on-the-fly by a farm of 350 Event Processing Nodes (EPNs).
• Workflow:
  • Raw data is decoded and analyzed in real-time.
  • Calibration parameters (thresholds, noise masks) are extracted and stored in the Condition and Calibration Database (CCDB).
  • Quality Control (QC) tasks run immediately to validate data integrity and fake-hit rates.
• Efficiency: A full re-calibration (threshold tuning + masking) can be completed in ~45 minutes using 40–80 EPNs.

C. Dynamic Adjustment Strategy

• Threshold Tuning: Performed annually or when radiation-induced drifts occur. The system adjusts global DAC settings (VCASN/ITHR) to target a mean threshold of 100 $e^-$.
• Masking: Noisy pixels and faulty double columns are masked to prevent readout stalls and combinatorial explosion during track reconstruction.
• Voltage Compensation: An iterative procedure compensates for voltage drops across power cables, ensuring the chip receives the target 1.8 V.

3. Key Contributions

1. Scalable Calibration Framework: Developed a method to calibrate 12.6 billion pixels efficiently, utilizing statistical sampling (short scans) for daily operations and full scans for annual references.
2. Real-Time Processing: Demonstrated the capability to process, analyze, and store calibration data on the ALICE computing farm within the inter-fill time window, enabling immediate detector optimization.
3. Radiation Monitoring: Established a clear correlation between accumulated radiation dose and detector performance degradation (threshold drift and noise increase), validating the need for periodic retuning.
4. Voltage-Threshold Correlation: Identified and corrected a linear correlation between the analogue supply voltage (AVDD) and the pixel threshold, reducing short-term fluctuations from ~5 $e^-$ to ~1 $e^-$.

4. Key Results

• Threshold Tuning: The system successfully tuned all 24,120 chips to a target mean threshold of 100 $e^-$.
  • Post-tuning standard deviation of mean chip thresholds: 3.8 $e^-$.
  • Pixel-to-pixel variation within a chip: ~20 $e^-$.
• Detector Health:
  • Bad Pixels: ~0.10% of total pixels are "bad" (non-functional or noisy).
  • Non-working Chips: 105 chips (0.43% of total pixels) are non-functional due to assembly damage or readout issues.
  • Noisy Pixels: Masking affects <0.01% of pixels per chip (approx. 50 pixels/chip).
• Stability Over Time:
  • Thresholds remained stable between 85–105 $e^-$ over 18 months (March 2023–Nov 2024).
  • A 15% decrease in Inner Barrel (IB) thresholds was observed after heavy-ion collisions due to radiation, necessitating retuning.
  • Outer Barrel (OB) thresholds showed a slight increase, consistent with lower radiation doses.
• Fake-Hit Rate:
  • The fake-hit rate remained consistently below the design limit of $10^{-6}$ hits/event/pixel.
  • After retuning in June 2024, the rate stabilized between $10^{-7}$ and $10^{-8}$.
• Signal Characteristics:
  • Mean Time of Arrival (ToA): 431 ns.
  • Mean Time-over-Threshold (ToT): 6.7 µs (for 300 $e^-$ charge).
  • Temporal noise: 5–8 $e^-$ (increasing slightly with radiation dose).

5. Significance

This work represents a milestone in high-energy physics instrumentation, proving that large-scale MAPS detectors can be operated stably in the harsh environment of the LHC.

• Operational Viability: It demonstrates that complex, multi-billion-pixel detectors can be calibrated and monitored dynamically without interrupting physics data taking.
• Radiation Hardness Management: The study provides critical data on how ALPIDE sensors degrade under LHC Run 3 conditions, informing future detector designs and operational strategies for Run 4 and beyond.
• Data Quality: By maintaining low fake-hit rates and high efficiency through precise calibration, the ITS2 ensures the high-quality tracking data required for ALICE's physics goals (e.g., heavy-flavor physics and quark-gluon plasma studies).
• Framework Reusability: The software and hardware infrastructure developed for ITS2 calibration serves as a blueprint for future large-scale pixel detector experiments.