A Data-Driven Fast Simulation Approach for MAPS-based Detectors and their Optimization

This paper presents a fast, data-driven parametric simulation tool for pixel sensors that bypasses the need for proprietary manufacturing details by using measurement inputs, demonstrating its efficiency and validation on the MALTA2 sensor to guide performance optimizations for the upcoming MALTA3 redesign.

The Gist

Imagine you are trying to build a super-advanced camera to take pictures of tiny, invisible particles zooming through space at nearly the speed of light. These particles are the building blocks of the universe, and to study them, scientists need detectors that are incredibly fast, precise, and tough.

This paper is about creating a super-fast "virtual camera" to test how these detectors work before they are actually built.

Here is the breakdown of the story, using some everyday analogies:

1. The Problem: The "Secret Recipe" Dilemma

Usually, to design these particle detectors, engineers use a tool called TCAD. Think of TCAD like a high-end cooking simulator. To make a realistic virtual cake, the simulator needs the exact secret recipe from the bakery: the precise temperature, the exact brand of flour, the humidity in the room, and the chef's specific hand movements.

The problem? The companies that make the chips (the "bakeries") often keep these recipes secret (proprietary). Without the full recipe, the simulation is slow, expensive, and sometimes impossible to run.

2. The Solution: The "Data-Driven" Shortcut

The authors of this paper invented a new way to simulate the camera. Instead of needing the secret recipe, they decided to just watch the cake bake and measure the results.

• The Old Way: "Let's simulate the chemistry of the dough." (Requires secret data).
• The New Way: "Let's shine a laser on the sensor, see how the electricity flows, and write a simple rulebook based on what we see."

They created a parametric simulation. This is like a smart rulebook that says, "If a particle hits here, the signal looks like X; if it hits there, it looks like Y." They built this rulebook using real measurements from a sensor called MALTA2.

Why is this cool? It's incredibly fast. It's like switching from a slow-motion, frame-by-frame animation of a car crash to a quick, accurate sketch. This allows scientists to test thousands of design ideas in the time it used to take to test one.

3. The Sensor: The "Pixelated Grid"

The sensor they are testing (MALTA2) is a grid of tiny squares (pixels), like a digital camera sensor, but much smaller and faster.

• Charge Sharing: When a particle hits the sensor, it doesn't always hit the dead center of a square. Sometimes it hits the edge, and the "signal" (electric charge) spills over into the neighboring squares.
• The Analogy: Imagine dropping a drop of water on a tiled floor. If it lands right in the middle of a tile, that tile gets wet. If it lands on the grout line, the water spreads to four tiles. The simulation has to figure out exactly how much water goes to which tile.

4. The Glitch: The "Traffic Jam" at the Exit

The biggest challenge with these sensors is how they send data out. The MALTA sensor uses an asynchronous system.

• The Analogy: Imagine a busy highway where cars (data) can leave at any time, not just when a traffic light turns green.
• The Merging Problem: If two cars try to exit the highway at the exact same time (within 1.6 nanoseconds—a blink of an eye), the exit gate has to merge them into a single lane. Sometimes, the gate gets confused. It might think two cars are one, or it might send them to the wrong exit lane.
• The Result: This causes "hit loss" (missing data) or "displaced hits" (data appearing in the wrong spot).

The paper shows that their new simulation can perfectly mimic this traffic jam. They proved that their "rulebook" predicts exactly where the traffic jams happen and how bad they get.

5. The Optimization: Fixing the Traffic

Once they had a working virtual model, they started playing "What If?" to make the next version of the sensor (MALTA3) better.

• Idea 1: Widen the Merge Window. What if the exit gate waited a tiny bit longer before merging cars?
  • Result: They found that making the gate faster (a shorter time window) reduced the traffic jams significantly.
• Idea 2: Change the Parking Lots. The sensor groups pixels into little blocks (currently 2x8). What if they made the blocks bigger (like 8x8)?
  • Result: Bigger blocks meant fewer "border crossings" where the traffic jams happened. It's like having fewer intersections in a city; traffic flows smoother.

6. The Two Main Jobs

The team tested this simulation for two different jobs:

1. Tracking: Like following a car's path through a city. They need to know exactly where the particle went. The simulation showed that by fixing the "traffic jams," they could get the tracking accuracy up to 99.8%.
2. Calorimetry: Like weighing a pile of sand by counting the grains. They need to count how many particles hit the sensor to measure energy. The simulation showed that with the new design, they could count particles accurately even for very high-energy crashes (up to 50 GeV).

The Bottom Line

This paper is a blueprint for a faster, smarter way to design particle detectors.

Instead of waiting years to build a physical prototype and test it, scientists can now use this "virtual sandbox" to try out new designs instantly. They found that by tweaking how the sensor groups its pixels and how fast it merges data, they can build a next-generation detector that is almost perfect at its job.

It's the difference between trying to fix a car engine by taking it apart and guessing, versus having a perfect 3D hologram of the engine where you can swap parts and see the results instantly.

Technical Summary

1. Problem Statement

The accurate simulation of silicon sensors is critical for designing next-generation particle physics detectors, particularly for tracking and vertexing systems near the beam pipe.

• Limitations of Current Tools: The industry standard, Technology Computer-Aided Design (TCAD), offers high predictive power but relies heavily on proprietary manufacturing process details. This makes TCAD simulations slow, computationally expensive, and difficult to adapt for rapid design iterations or large-scale detector systems.
• The Challenge: There is a need for a fast, efficient simulation tool that can predict sensor performance (specifically for Monolithic Active Pixel Sensors, or MAPS) without requiring access to confidential chip design data. This is essential for optimizing digital read-out architectures and large detector geometries for experiments like the High-Luminosity LHC (ATLAS) and the Electron-Ion Collider (EIC).

2. Methodology

The authors developed a parametric, data-driven simulation framework structured as a GEANT4 suite combined with a C++ offline analysis package. The approach bypasses detailed TCAD physics modeling in favor of parametrizing sensor behavior based on empirical measurements.

• Charge Sharing Model:
  • Derived from Edge Transient Current Technique (E-TCT) and test beam data.
  • Modeled using a convolution of an error function (describing charge collection at pixel edges) and a Gaussian distribution (accounting for laser beam width).
  • Validated as a 2D symmetric model where charge is distributed among the "seed" pixel and its three nearest neighbors, summing to unity.
• Time Walk Modeling:
  • Parametrized the charge-dependent timing shift (time walk) observed in fixed-threshold pixel sensors.
  • Measured using an oscilloscope with varying laser intensities to establish a relationship between signal amplitude and arrival time delay.
  • Includes a scaling law to extrapolate time walk behavior for different threshold settings.
• Digital Hit Merging Logic:
  • Simulates the specific asynchronous read-out architecture of the MALTA2 sensor, where hits within a 1.6 ns window are merged via a bitwise OR operation.
  • Models the "hit merger" circuitry, which encodes pixel positions into 40-bit words. This logic can cause displaced hits (incorrect coordinate assignment) or hit loss when multiple hits occur in different pixel groups simultaneously.
• Validation Platform:
  • The simulation was validated against experimental data from the MALTA2 sensor (a 180 nm CMOS DMAPS) using a 120 GeV/c hadron beam at the CERN SPS.
  • Two application scenarios were simulated: Tracking (using a particle gun and passive silicon planes) and Digital Calorimetry (using a 3.2 cm tungsten block to induce electromagnetic showers).

3. Key Contributions

1. Fast, Proprietary-Free Simulation Tool: A GEANT4-based package that simulates MAPS performance using only measurement-derived parameters, eliminating the need for proprietary process data.
2. Comprehensive Validation: The tool was rigorously validated against real test beam data for efficiency, cluster size, and timing, demonstrating high fidelity.
3. Optimization of Digital Read-Out: The simulation was used as a "testbed" to explore redesigns for the upcoming MALTA3 sensor (65 nm TPSCo process), specifically focusing on the digital front-end.
4. Quantification of Merging Inefficiencies: The study isolated the "hit merger" circuit as the primary source of detection inefficiency and coordinate displacement, providing a quantitative basis for architectural changes.

4. Key Results

• Validation Accuracy:
  • Efficiency: The simulation matched experimental data with a relative residual of 0.005% (within 1σ). At the nominal 200 e⁻ threshold, both data and simulation achieved ~99.24% efficiency.
  • Cluster Size: The average cluster size matched within 1.67% deviation (Data: 1.48 pixels; Simulation: 1.45 pixels).
  • Timing: Qualitative agreement was found, though quantitative differences existed due to simplifications in the timing model (Data: 3.12 ns center-to-corner difference; Simulation: 1.40 ns).
• Tracking Optimization:
  • Merging Window: Reducing the merging time window from the current 1.6 ns to 0.5 ns significantly improved efficiency (reaching ~99.8%).
  • Pixel Grouping: Increasing the pixel group size (e.g., from 2×8 to 8×8 pads) reduces the perimeter-to-area ratio, minimizing the probability of hits falling on group boundaries where merging errors occur. The 8×8 configuration showed the highest efficiency.
  • Orientation: Vertical pixel group alignment was found to be superior to horizontal alignment due to the specific bit-encoding structure of the MALTA read-out.
• Calorimetry Optimization:
  • Digital calorimetry suffers from hit-counting saturation at high energies due to merging.
  • Increasing the pixel group size (e.g., to 8×8) and reducing the merging window (0.5 ns) restored linearity in hit counting up to 50 GeV for electromagnetic showers.
  • Higher thresholds (e.g., 1000 e⁻) further improved linearity at high energies by reducing the average cluster size and thus the frequency of merging, though this trades off low-energy resolution.

5. Significance

• Accelerated Sensor Design: This approach allows for rapid iteration of digital read-out architectures (like the upcoming MALTA3) without waiting for costly TCAD runs or new fabrication cycles.
• Scalability: The computational efficiency of the parametric model makes it feasible to simulate large-scale detector systems (e.g., full tracker upgrades) that would be prohibitively expensive with TCAD.
• Performance Gains: The study provides concrete design guidelines (smaller merging windows, larger pixel groups, optimized orientation) that are expected to push sensor efficiency above 99.8% and extend the linear dynamic range of digital calorimeters.
• Future Applications: The framework is designed to be generalizable to other MAPS technologies, provided equivalent charge collection and timing data are acquired, supporting future experiments at the HL-LHC and EIC.