Gaussian Random Number Generator and Generative Adversarial Networks Implementation on FPGA for Generating CMS Events at 40MHz

To develop and test the upgraded CMS systems for long shutdown 3, one would need to generate a continuous stream of simulated CMS events at a 40 MHz rate, which is unfeasible with standard techniques as even generation and simulation software takes several seconds per event. The project explored the alternative route of using a machine learning approach called Generative Adversarial Networks -GANs- to generate events at 40MHz. Then this algorithm was run in a prototype trigger board FPGA and generated realistic data that can be feeded into the prototypes of the scouting system. The project has two main challenges: 1. Implementing a normally distributed random number generator on FPGA to feed the generator of GANs . 2.Resource utilization challenge of GRNG comes together with the input noise dimension of the GANs. One should construct a generative adversarial network generator that can achieve high accuracy with the small sized input noise.