Sensor Co-design for $\textit{smartpixels}$

TL;DR

This work investigates sensor-level co-design for smartpixels, aiming to perform on-detector data reduction with a neural-network classifier integrated into the sensor ASIC. By systematically varying sensor geometry, Lorentz drift, irradiation, and noise using PixelAV-based simulations, the study quantifies impacts on signal efficiency and data reduction, guiding design choices. Key findings show that smaller y-pitch and greater thickness improve discrimination, Lorentz drift enhances low-$p_T$ performance, and retraining is essential to counteract radiation damage and noise, highlighting the need for reprogrammable on-sensor weights. The results provide concrete guidance for deploying on-detector filtering in future collider detectors and lay groundwork for further ASIC development and prototype studies.

Abstract

Pixel tracking detectors at upcoming collider experiments will see unprecedented charged-particle densities. Real-time data reduction on the detector will enable higher granularity and faster readout, possibly enabling the use of the pixel detector in the first level of the trigger for a hadron collider. This data reduction can be accomplished with a neural network (NN) in the readout chip bonded with the sensor that recognizes and rejects tracks with low transverse momentum (p$_T$) based on the geometrical shape of the charge deposition (``cluster''). To design a viable detector for deployment at an experiment, the dependence of the NN as a function of the sensor geometry, external magnetic field, and irradiation must be understood. In this paper, we present first studies of the efficiency and data reduction for planar pixel sensors exploring these parameters. A smaller sensor pitch in the bending direction improves the p$_T$ discrimination, but a larger pitch can be partially compensated with detector depth. An external magnetic field parallel to the sensor plane induces Lorentz drift of the electron-hole pairs produced by the charged particle, broadening the cluster and improving the network performance. The absence of the external field diminishes the background rejection compared to the baseline by $\mathcal{O}$(10%). Any accumulated radiation damage also changes the cluster shape, reducing the signal efficiency compared to the baseline by $\sim$ 30 - 60%, but nearly all of the performance can be recovered through retraining of the network and updating the weights. Finally, the impact of noise was investigated, and retraining the network on noise-injected datasets was found to maintain performance within 6% of the baseline network trained and evaluated on noiseless data.

Figures (8)

• Figure 1: A schematic diagram of two arbitrary charged particles traversing sensors in the barrel and endcap regions. The diagram highlights the definition of $\alpha$ and $\beta$ angles in both these regions for arbitrary incident particles, alongwith the magnetic field direction. The change in orientation of the sensors in the endcap regions with respect to sensors in the barrel region requires defining a coordinate system for each regions to ensure the definition of incident angles remain the same.
• Figure 2: Distributions of p$_T$ (left), $\alpha$ (center), and $\beta$ (right) across the various datasets produced. A symmetric distribution of $\alpha$ values in the center plot is not seen in the endcap regions because only a section of the endcap disk was simulated, ignoring the section diametrically opposite due to radial-symmetry considerations.
• Figure 3: Distribution of cluster size along the Y-direction for sensor geometries in the barrel region, using the physical-p$_T$ dataset. The distributions in left and right correspond to sensors with varying $y$-pitch and thickness, respectively.
• Figure 4: Performance results of sensors in the barrel region for four sensor geometries as a function of p$_T$ boundary. The legend describes the pixel dimensions along the $x$ (length), $y$ (width), and $z$ (thickness) directions, respectively.
• Figure 5: Performance results of sensors in barrel region for sensor geometries varying in pitch and thickness. The X-axis labels indicate unit pixel dimensions along the $x$ (length), $y$ (width), and $z$ (thickness) directions, respectively. The solid, dashed, and dotted lines correspond to geometries with varying $y$-pitch, $x$-pitch, and thickness, respectively.