Development of a system for testing full-size CMS LGAD sensors

This paper presents a modular, automated probe card system designed for the scalable electrical characterization of full-size pixelated LGAD sensors, demonstrating its capability to perform rapid I-V and C-V measurements with minimal leakage current to address the quality control needs of large-scale particle physics experiments.

The Gist

Imagine you are a quality control inspector for a massive, high-tech city made of 256 tiny, independent power stations. Each station is a LGAD sensor pixel, a microscopic silicon chip designed to act like a super-fast stopwatch for particles in physics experiments. These chips are incredibly sensitive; if even one is broken or misbehaving, it could ruin the data for the whole city.

The problem? Checking these 256 stations one by one by hand is like trying to test every lightbulb in a skyscraper by climbing a ladder and unscrewing each one individually. It's slow, tedious, and prone to human error.

This paper describes a new automated robot system built by researchers in Korea to solve this problem. Here is how their system works, explained in everyday terms:

1. The "Finger" Array (The Probe Card)

Instead of a human hand with one finger, the team built a specialized "finger" array called a probe card.

• The Analogy: Imagine a giant, spring-loaded comb with 256 tiny, bouncy pins (called pogo pins).
• How it works: When you press this comb onto the sensor chip, all 256 pins land perfectly on their matching spots simultaneously. Because they are spring-loaded, they stay connected even if the chip wobbles slightly, ensuring a firm handshake with every single pixel at once.

2. The "Traffic Controller" (The Switching Board)

Once the pins are connected, you need to test them. You can't plug all 256 into your measuring tools at once; you need to check them one by one (or in groups).

• The Analogy: Think of the switching board as a massive, high-tech traffic control center or a switchboard operator.
• How it works: This board has 256 lanes. When the computer wants to test "Pixel #42," the switchboard instantly connects only Pixel #42 to the measuring machine and sends all the other 255 pixels to "ground" (a safe, quiet resting state). This prevents noise or interference from the neighbors from messing up the test.
• The Bonus: It's not just for one-by-one testing. The switchboard is smart enough to group pixels together. You can test an entire row of 16 pixels at once to get a quick "health check" of that whole line, or even test the connection between two neighbors.

3. The "Robot Arm" (Mechanics and Alignment)

To make sure the spring-loaded comb lands perfectly on the tiny chips, the system uses a precise mechanical rig.

• The Analogy: Imagine a camera-guided robot arm that can move in every direction (up, down, left, right, and even tilt).
• How it works: The system uses cameras to look at the sensor and the probe card. It adjusts the position until the pins align perfectly with the tiny pads on the chip. It also keeps the whole setup in a dark box because these sensors are so sensitive that even a little bit of light can confuse the measurements (like trying to hear a whisper in a noisy room).

4. The "Brain" (Software)

All of this hardware is controlled by custom software.

• The Analogy: This is the conductor of an orchestra.
• How it works: The software tells the robot where to move, tells the switchboard which pixel to test next, and tells the measuring instruments what voltage to apply. It runs automatically, so a human doesn't need to touch anything once the process starts. It can also be controlled remotely from a different computer.

The Results: Fast vs. Detailed

The researchers tested this system on a 16x16 grid of sensors and found it worked beautifully:

• The "Speed Run": They tested the sensors row-by-row (16 pixels at a time). The whole 256-pixel chip was scanned in about 20 minutes. This is great for a quick "is everything working?" check.
• The "Deep Dive": They then tested every single pixel individually, one by one, from 0 to 300 volts. This took about 340 minutes (nearly 6 hours). This is necessary to find tiny defects that the speed run might miss.
• The "Silent Partner": They checked if the switchboard itself added any "noise" (leakage current) to the measurements. They found the noise it added was tiny (less than 1 nanoampere), which is so small it's like a single drop of water in a swimming pool compared to the normal signal of the sensor. It didn't ruin the test.

Why This Matters

In the past, testing these chips was slow and manual. This new system is like upgrading from a hand-cranked generator to a high-speed power plant. It allows scientists to check thousands of these sensors quickly and reliably, ensuring that the massive detectors used in particle physics (like those at the Large Hadron Collider) are working perfectly before they are installed.

In short: They built a robotic, automated "comb" and a "traffic controller" that can test 256 tiny, sensitive chips in minutes, ensuring they are all ready for the big physics experiments.

Technical Summary

Technical Summary: Development of a System for Testing Full-Size CMS LGAD Sensors

Problem Statement
Low-Gain Avalanche Diodes (LGADs) are critical for precision timing in high-luminosity particle physics experiments, such as the High-Luminosity Large Hadron Collider (HL-LHC) era of ATLAS and CMS. These experiments require thousands of sensors with varying pixel geometries. As the scale of sensor deployment increases, conventional manual characterization techniques become inadequate for large-scale quality control (QC). Previous automated systems, such as those developed for the ATLAS experiment, are often optimized for sequential single-pixel measurements on smaller arrays (e.g., 5×5 or 15×15) and lack the flexibility for diverse test scenarios like grouped, row-wise, or inter-pixel measurements required for larger, more complex arrays. There is a need for a scalable, modular system capable of automated electrical characterization of large pixelated LGAD sensors (specifically 16×16 arrays) with arbitrary pixel selection.

Methodology
The authors developed a modular probe card system designed for the automated electrical characterization of 16×16 LGAD arrays. The system architecture comprises four primary subsystems:

1. Probe Card: A custom probe card was fabricated to establish simultaneous electrical contact with all 256 pixels of the LGAD array. It utilizes a pogo-pin array with a 1.3 mm pitch matching the sensor geometry, plus 15 additional pins for guard ring pads. Pogo pins were selected over cantilever probes for their mechanical robustness and repeatability over many measurement cycles. The card connects to the switching system via flexible flat cable (FFC) connectors.
2. Mechanics and Alignment: A dedicated mechanical jig ensures precise alignment between the probe card and the sensor. The system employs a vacuum chuck for the sensor and multi-axis precision motion stages (XY, Z, tilt, rotation) with 10 µm resolution. Alignment is monitored using a combination of top-view and side-view cameras to correct lateral position, rotation, and tilt, ensuring the pogo pins compress within their 0.5 mm stroke without damaging the sensor. The entire setup operates within a light-tight enclosure to prevent photo-induced currents.
3. Switching Board: A 256-channel modular switching matrix was developed to route signals from selected pixels to measurement instruments while grounding all non-selected pixels. The system consists of a motherboard hosting 16 unit boards, each containing four TMUX1134 analog switches (providing 16 channels per board). This architecture supports four independent measurement buses, enabling flexible configurations including single-pixel, grouped, row-wise, and column-wise measurements. Control is managed by a Raspberry Pi Pico microcontroller communicating via I2C with PCF8575 I/O expanders on each unit board.
4. Software and Instruments: The system is controlled by a server-client software architecture. A C++ daemon running on a host PC coordinates the switching matrix and measurement instruments (Source-Measure Unit, picoammeter, and LCR meter) via USB serial interfaces. The software provides both a web-based GUI (React) and a command-line interface (CLI) for automated scan sequences, real-time monitoring, and remote operation.

Key Contributions

• Modular Architecture: The system introduces a scalable hardware design where the switching matrix is composed of interchangeable unit boards, simplifying maintenance and future adaptation to different sensor layouts.
• Flexible Measurement Modes: Unlike previous systems focused solely on sequential single-pixel access, this design supports arbitrary pixel selection, grouped measurements, and rapid row-wise or column-wise scans.
• Two-Stage QC Workflow: The system enables a practical workflow combining rapid pre-screening (row-wise scans) with detailed inspection (pixel-by-pixel scans), significantly improving throughput for large arrays.
• Low-Noise Switching: The design minimizes signal degradation, with the switching matrix introducing negligible leakage current compared to the intrinsic leakage of functional LGAD pixels.

Results
The system was validated using a 16×16 LGAD array:

• Throughput: A full row-wise I–V scan of the 16×16 array was completed in approximately 20 minutes. A full pixel-by-pixel I–V scan (0 to 300 V, 1 V step) required approximately 340 minutes.
• Measurement Quality: The switching matrix was tested for noise and leakage. In a conservative worst-case configuration (all 256 outputs connected to a single channel), the total leakage current contribution was less than 1 nA. The switching matrix introduced an offset of ~0.46 nA and increased the standard deviation of the current measurement from 0.066 nA to 0.19 nA, values considered small relative to normal LGAD pixel leakage.
• Characterization: The system successfully generated I–V and C–V maps for the entire array. Pixels were classified into "normal," "warning," and "broken" categories based on breakdown behavior and depletion characteristics. The study noted that I–V and C–V status maps do not always coincide, as they probe different electrical characteristics (breakdown vs. depletion behavior).

Significance
The paper claims that the developed system provides a practical and scalable solution for the quality control of large-scale LGAD sensors. By combining a modular hardware design with flexible switching and automated software, the system addresses the limitations of conventional manual techniques and previous automated systems. It is presented as a viable tool for distributed testing environments and future large-scale production campaigns for high-granularity timing detectors, enabling efficient identification of defective pixels and ensuring the uniformity required for next-generation particle physics experiments.