Search papers, labs, and topics across Lattice.
The paper introduces a hierarchical stream compaction pipeline implemented on FPGAs to address the challenge of processing sparse sensor data in real-time machine learning applications, specifically GNN-based first-level triggers in collider experiments. This pipeline rearranges data from numerous input FIFOs to fewer output FIFOs connected to the ML accelerator, effectively removing dynamic input sparsity and improving hardware acceleration. The authors demonstrate the concept with a GNN trigger for the Belle II detector's Electromagnetic Calorimeter and provide extensive performance evaluations, including latency, throughput, and resource utilization, for various configurations.
Squeezing sparse sensor data through an FPGA-optimized stream compaction pipeline unlocks efficient GNN acceleration for real-time applications like high-energy physics triggers.
Machine learning algorithms are being used more frequently in the first-level triggers in collider experiments, with Graph Neural Networks pushing the hardware requirements of FPGA-based triggers beyond the current state of the art. To meet the stringent demands of high-throughput and low-latency environments, we propose a concept for latency-optimized preprocessing of sparse sensor data, enabling efficient GNN hardware acceleration by removing dynamic input sparsity. Our approach rearranges data coming from a large number of First-In-First-Out interfaces, typically sensor frontends, to a smaller number of FIFO interfaces connected to a machine learning hardware accelerator. In order to achieve high throughput while minimizing the hardware utilization, we developed a hierarchical sparsity compression pipeline optimized for FPGAs. We implemented our concept in the Chisel design language as an open-source hardware generator. For demonstration, we implemented one configuration of our module as preprocessing stage in a GNN-based first-level trigger for the Electromagnetic Calorimeter inside the Belle II detector. Additionally we evaluate latency, throughput, resource utilization, and scalability for a wide range of parameters, to enable broader use for other large scale scientific experiments.
