Does one do the “minimum” DFT just to avoid buying the “next” tester, does one do all the test “on chip” and cost reduce the ATE down to a battery and a data source and sink, or is silicon so precious that any use for non mission related function make the cost of the final die prohibitive. One can easily find proponents for each position and the data to back it up. This is because the answer is highly dependent on the particulars of the design of the device and the issues related to the business case that the device is going into.