Moving NN Triggers to Level-1 at LHC Rates

Although the use of neural network hardware for level 2 triggering at hadron colliders has been established for many years, there has been little investigation of neural techniques for level 1 triggers at LHC experiments. In addition, commercially available NN solutions with appropriate timing constraints do not exist. Presented here is a new FPGA-based method for implementing multilayer perceptrons with hundreds of neurons in a 25 ns pipeline structure having only 600 ns latency. Possible implementation in an LHC level 1 trigger scenario is discussed.