Developing trigger algorithms for the CMS Upgrade at the HL-LHC

从2027年起，大型强子对撞机的重大升级将使质子碰撞的速率提高到前所未有的每秒100亿次。一个复杂的数据处理系统-“L1触发器”-将处理每一个事件，以确定感兴趣的碰撞，供储存和进一步分析;未被选择的事件将永远丢失。L1触发器由使用高速光链路和FPGA器件的定制硬件处理器构成。它可以以60 Tb/s的速度接收数据，并且必须在不到10秒的时间内处理每个事件。在这个项目中，我们将使用标准技术和机器学习来开发触发算法，以识别有趣的碰撞，例如CMS L1触发器的一对希格斯玻色子的产生。这些网络的物理性能将使用蒙特-卡罗模拟进行研究和优化。将针对一系列网络类型研究物理性能与推理延迟和资源使用之间的权衡。最终，候选算法将在FPGA中实现，并在真实的处理器板上演示。

A major upgrade of the Large Hadron Collider from 2027 will increase the rate of proton collisions to an unprecedented 10 billion per second. A sophisticated data processing system - the "L1 Trigger" - will process each event to identify interesting collisions for storage and further analysis; events not selected are lost forever. The L1 Trigger is constructed of custom hardware processors using high-speed optical links and FPGA devices. It can receive data at 60 Tb/s and must process each event in less than 10s. In this project we will develop trigger algorithms using both standard techniques and employing machine learning to identify interesting collisions, such as production of a pair of Higgs bosons, for the CMS L1 Trigger. The physics performance of these networks will be studied and optimised using Monte-Carlo simulation. The trade-off between physics performance and inference latency and resource usage will be studied for a range of network types. Ultimately, candidate algorithms will be implemented in FPGAs and demonstrated in real processor boards.