In recent years, neural networks have proven to be algorithms that can solve a wide variety of problems. They are used in particular when there is a lot of data but no formal description of the problem. The mathematical structure of neural networks allows for a high degree of parallelization and the use of bi-width optimizations. They are therefore well suited for implementation on FPGAs. This means that the advantages of AI can also be used in the embedded sector.

The process for implementing a neural network in the FPGA includes, among other things:

• Data preparation
• Prototyping and network design with python machine learning frameworks
• Offline training of the network with data records
• Transferring the architecture into an FPGA design
• Interface specification

With the help of FPGAs, you can also use machine learning algorithms where low latencies and performance requirements are needed.