Creating scalable neural networks with maximal FPGA resources