In the CMS detector, a two-level trigger system is installed to select interesting collision events. The first level trigger (L1) is designed to take a decision synchronously with the 40 MHz bunch-crossing rate. The system is built in custon hardware and has been successfully operated in run-1 and run-2 of CMS data-taking. For the high-luminosity-LHC (HL-LHC) an upgrade of the L1 trigger system is foreseen to provide an even more efficient event selection within the bandwidth constrains in this very challenging environment. In oder to meet the harsh requirements, new algorithms of artificial intelligence (machine learning) are foreseen to be executed on field-programmable-gate-arrays (FPGAs). We are developing the respective algorithms, simulate their performance and test them on our local FPGA hardware, before being installed in the online environment of the CMS L1 trigger in the future.