As for reliability, you raise a good point, but I don’t really know the answer to your question. The lifespan of a PUF will really depend on the way it is implemented, be that on an FPGA, ASIC, or something else completely.

As for your security comment, I think I might have explained the PUF incorrectly. The PUF doesn’t actually store any keys at all. A PUF generates responses by running a large amount of ring oscillators through several multiplexers and seeing which oscillators are fastest, which determines the bits of the response. The actual response is never actually written down in the PUF; it is dynamically produced each time. As such, it would be nearly impossible for an attacker to duplicate the PUF, since the circuitry used in the design is usually in the nanometer range. Any type of measurement at this scale will most likely be either inaccurate or simply destroy the PUF.

You also mentioned that given enough time, one could guess the “challenge-response” pairs, but I disagree. For the PUFs that I worked on, these were 128 bit inputs and outputs, which is a very large number indeed.

- Reliability. Will it reliably work in 1, 2, 5 or 10 years in the product?
- Security. PUFs are used as secure key storage. There are succesfull attacks to crack PUFs and extract the key. You can within reasonable computational time guess the PUF values. The attacker can sucessfully attack the rest of PUF logic and extract the key. It is difficult but not impossible!