SparkPix-RT

(LCLS/Argonne)

SLAC’s strategic detector development framework will allow data handling at the point of generation to be transformative and fully integrated as needs evolve. Algorithms hosted on the edge must be particular to the scientific question and must extract contextually relevant information before passage down the analysis chain. This is particularly challenging for light sources such as LCLS, where the experiments change on a daily or weekly basis in response to the breadth of the user program. SLAC is working closely with its users to identify the high-impact science cases and flexible technical solutions that will allow extraction and compression techniques to be implemented in Edge computing and specialized detectors.

During the past year, substantial progress has been made toward intelligent X-ray detectors with real-time information extraction capabilities. Two extraction engines have been successfully prototyped and are being integrated into a new family of detectors named SparkPix. The first two instances of this family will address the needs of the atomic and molecular physics program targeting coincidence measurements at LCLS in experiments to reconstruct the earliest stages of chemical reactions and in experiments designed to study stochastic rare events such as materials failure and heterogeneous catalysis. In addition, following a co-design approach with the data system, the hardware framework for on-detector edge-computing employing ML algorithms instantiated on FPGAs is now under development.

This proposal aims at developing the underpinning microelectronics technologies for next generation intelligent detectors required for future ultrafast ultra-bright multi-probe Free Electron Laser and Storage Ring facilities (BES, FES, ASCR) while addressing more broadly needs across different fields (HEP, NP). We envision detector systems derived from and co-designed with next generation modeling and analysis tools that incorporate AI and ML technologies for feature extraction, anomaly detection, and tagging, which will give us the capabilities to explore new materials and better control fabrication processes of next- generation microelectronics technologies.

Extracting new physics from instruments producing large amounts of data requires an expanded co- design paradigm with a focus on massive-scale data analytics. As a first objective, we intend to co-design an architecture with a multi-tiered pipeline of hardware and software spanning the detectors, data acquisition, computing systems, and their algorithms. An architecture capable of real-time feedback built into the analysis pipeline, enabling autonomous experiments that tighten the coupling between experimental analysis and data acquisition by incorporating ML/AI techniques that allow to optimize and prioritize the acquisition of data maximizing scientific return. Real time science extraction will allow us to realize our second objective: design next generation x-ray detectors with higher spatial resolution, temporal resolution at tenths of picosecond level for time-resolved experiments, dynamic range, contiguous/gap-less form-factors, and frame rates from megahertz in continuous mode to gigahertz in burst mode, with an expanded sensitivity range toward the very-soft and the very hard x-ray range. To keep up with the natural production rate of the data, we intend to apply ML/AI techniques capable of producing information that is actionable during the experiment.

The project team will pursue four strongly interconnected research thrusts: (1) sensor optimization for fast timing and efficiency over a large range of energies, (2) ultra-high-rate front-end ASIC with high position and timing resolution, dynamic range, and information extraction capabilities, (3) edge computing FPGA based architectures and DAQ, (4) ML/AI based workflows for dynamic real-time experiment operation. The goal of the proposal is to develop a technology demonstrator whose success will constitute the basis for future x-ray detectors matching the characteristics of next generation light sources.  We have provided datasets for evaluation of FPGA-based architectures and bit-level compression algorithms.