You dont. You know they will happen and it will be a non trivial percentage. Could be anywhere from 5 to 50% on a first run. But thats not the question. The question is if you want to discover all those bad chips before you spent time and money putting the defective chips in a fully populated PCB and assemble it to a functional testable miner, possibly with components that are harder to score than the asics (ask BFL) , or if you are going to do what is done with probably every other asic ever produced, test the dies and/or packaged chips before you waste those components,  time and money.

Im nt sure you realize those chips might be among the cheapest components and with the highest defect rate by far. Thats the one you want throw away, not the rest.