Abstract:We present a smart pixel prototype readout integrated circuit (ROIC) designed in CMOS 28 nm bulk process, with in-pixel implementation of an artificial intelligence (AI) / machine learning (ML) based data filtering algorithm designed as proof-of-principle for a Phase III upgrade at the Large Hadron Collider (LHC) pixel detector. The first version of the ROIC consists of two matrices of 256 smart pixels, each 25$\times$25 $\mu$m$^2$ in size. Each pixel consists of a charge-sensitive preamplifier with leakage current compensation and three auto-zero comparators for a 2-bit flash-type ADC. The frontend is capable of synchronously digitizing the sensor charge within 25 ns. Measurement results show an equivalent noise charge (ENC) of $\sim$30e$^-$ and a total dispersion of $\sim$100e$^-$ The second version of the ROIC uses a fully connected two-layer neural network (NN) to process information from a cluster of 256 pixels to determine if the pattern corresponds to highly desirable high-momentum particle tracks for selection and readout. The digital NN is embedded in-between analog signal processing regions of the 256 pixels without increasing the pixel size and is implemented as fully combinatorial digital logic to minimize power consumption and eliminate clock distribution, and is active only in the presence of an input signal. The total power consumption of the neural network is $\sim$ 300 $\mu$W. The NN performs momentum classification based on the generated cluster patterns and even with a modest momentum threshold, it is capable of 54.4\% - 75.4\% total data rejection, opening the possibility of using the pixel information at 40MHz for the trigger. The total power consumption of analog and digital functions per pixel is $\sim$ 6 $\mu$W per pixel, which corresponds to $\sim$ 1 W/cm$^2$ staying within the experimental constraints.

What problem does this paper attempt to address?

The paper primarily explores a novel prototype Readout Integrated Circuit (ROIC) for "smart pixels," aiming to address the technical challenges faced by silicon pixel detectors in future high-energy colliders by implementing Artificial Intelligence (AI)/Machine Learning (ML) algorithms within the pixels to filter sensor data. Specifically, the research goals include: 1. **Improving data processing efficiency**: Given the massive amount of data generated per second, efficiently reducing the data volume becomes crucial. By integrating AI/ML algorithms within the pixels, data filtering can be performed at the source, thereby reducing the amount of data transmitted to peripheral systems for further processing. 2. **Adapting to high-radiation environments**: Future high-energy collision experiments will generate a large number of particle events that need to be accurately detected and recorded in high-radiation environments. Therefore, designing detectors that can operate stably in such environments is essential. 3. **Meeting stringent bandwidth and latency requirements**: To ensure real-time data analysis and discovery, future trackers must operate under limited bandwidth and latency conditions. Implementing data filtering within the pixels helps reduce the dependency on external data processing systems, thereby meeting these stringent requirements. Key technical points in the paper include: - **First-generation ROIC**: It includes an analog front-end for collision event detection and digitization, capable of synchronously digitizing sensor charges and demonstrating good performance metrics in tests. - **Second-generation ROIC**: Building on this, it integrates a fully connected two-layer neural network (NN) to process information from a cluster of 256 pixels and determine whether these patterns correspond to high-momentum particle tracks, thereby deciding whether to select and read out this information. In this way, the research team not only achieved effective data filtering but also significantly reduced the overall data transmission volume, which is of great importance for data processing in future high-energy physics experiments.

Smart Pixels: In-pixel AI for on-sensor data filtering

A Reconfigurable Convolution-in-Pixel CMOS Image Sensor Architecture

Smart pixel sensors: towards on-sensor filtering of pixel clusters with deep learning

Intelligent Pixel Detectors: Towards a Radiation Hard ASIC with On-Chip Machine Learning in 28 nm CMOS

Smartpixels: Towards on-sensor inference of charged particle track parameters and uncertainties

Pixel-Level Clustering Algorithm for Data Reconstruction in ILC

Design of a Pixel Readout ASIC Using Super Pixel Circuits with a Built-in LDO Regulator for Hybrid X-ray Imaging Detectors

A Full Functional Monolithic Active Pixel Sensor Prototype for the CEPC Vertex Detector

A Sensor with Analog and Digital Pixels in 0.15 micron SOI Technology

M.i.p. detection performances of a 100 us read-out CMOS pixel sensor with digitised outputs

A Self-Triggered Column-Level ADC for CMOS Pixel Sensors in High Energy Physics

Charged particle detection using a CMOS active pixel sensor

Pixel detector R&D for the Compact Linear Collider

High data-rate readout logic design of a 512 × 1024 pixel array dedicated for CEPC vertex detector

Beyond the CMOS sensors: the DoTPiX pixel concept and technology for the International Linear Collider A

Low-power In-pixel Computing with Current-modulated Switched Capacitors

The SparkPix-S ASIC for the sparsified readout of 1 MHz frame-rate X-ray cameras at LCLS-II: pixel design and simulation results

Characterization of the first prototype CMOS pixel sensor developed for the CEPC vertex detector

Pixel front-end development in 65 nm CMOS technology

Design and Performance of HV CMOS Sensors for Future Colliders by the RD50 Collaboration

A single-photon sampling architecture for solid-state imaging