Although it is possible to implement some popular algorithms of FPGA, there is a problem. It may require a custom and expensive PCB in order to match the GPU performance on existing algoritms.
Actually thats not exactly true FPGAs require not many external components and for mining you dont use all of the pins so less tracks. Back in the days when mining of bitcoin was possible on an FPGA most of the boards was 4 layers of copper and people managed to produce them quickly, for comparison a typical GPU board is 10-12 layers of copper which is out of scope for a lot of PCB manufacturers still, but anything that is 4 or 6 layers is cheap nowadays.
Also Monero is minable on FPGAs its highly probable that people are mining monero on FPGAs as we speak, of course i'm not a dev and cannot speak for the profitability of the FPGAs
Those days are almost gone. ASICs are destroying all profitable coins step by step, from the other hand GPUs are getting better.
I think it is necessary to use external memory to make FPGA-miner profitable. FPGA without external memory is severely limited. It is also good to implement the PCI Express interface. As a result, the PCB will not be very simple.
The problem will not be in the complexity of the PCB, but in the price of the FPGA. You need a huge discount for a massive purchase in order to get a payback.
As a FPGA developer, I'm considering only high-performance solutions. They are compact, profitable and scalable. Perhaps I underestimate the potential of inexpensive modules, because I usually do a fully unrolled pipelines.But algorithms like nist5 require a folded implementation when small chips are used. So low-cost cards may be profitable.
But IMO the most interesting is solution based on top FPGA like Virtex Ultrascale+.
I will support a new thread dedicated to the FPGA.