Another thought, FWIW (just thinking out loud):
For more analog-oriented circuits, a similar but different tool could be made by stimulating the DUT with a relatively fast step function, then capturing the response waveform (either on the same pin, or another point in the network).
About 40 years ago, me and a colleague wrote up a patent application on a similar idea. He experimentally showed that with a sampling scope, there were small differences that could be seen in the step-response when a component was modified in a network. But associating with a particular component was difficult. My contribution was to make a neural network to analyze the step-response waveform, and thus point to a *specific* component that was 'not correct'. We trained the network on many simulated circuits (each with a component defect), then verified sample cases with real hardware. It was for a manufacturing process, to be used as a tool for debugging many identical boards (analog read-chain circuits for disk drives). The company (DEC) chose not to pursue the patent because it could simply be 'kept secret' in our manufacturing process.
Pete
saipan59 (Pete)