Majority voting is an effective technique when PUF responses are characterized by low or transient noise. It is realized through the collection of a significant number of responses: if they are repeatedly extracted from the same PUF, the technique is defined as temporal majority voting, vice versa if they are taken from multiple PUFs at once, the method is defined as spatial majority voting .
As for the spatial majority voting, its discrete nature impedes to reach good values of bits uniformity in responses. As for the temporal majority voting, the technique needs for a number of repeated measures which exponentially increase with the desired noise reduction, due to the Chernoff bound.
To be used as keys, PUF responses have to guarantee perfect distribution of bits and noise-free responses.
Contrary to majority voting algorithms, fuzzy extractor schemes involve an error- correction code algorithm to set PUF responses free from the noise and require to collect only one sample per single response [12, 26]. Typically, a fuzzy extractor scheme requires two main phases. The first one, namely generation phase, the PUF is enrolled and the obtained response is securely stored. Moreover, an additional bit string, called helper data, is generated. The second one, called reproduction phase, exploits the previously defined helper data in order to recover noisy version of the enrolled response, The helper data are not critic for the scheme and can be publicly exchanged, as they do not weaken secrecy of the PUF response. If the noise which affects the response is small enough, the reproduction phase guarantees that the reproduced response perfectly matches the PUF response extracted in the first phase. In particular, depending on the error-correction design parameters, fuzzy extraction is able to recover a response if the amount of error bits is under a certain threshold. Figure 10.10 illustrates every step for generation and reproduction phases.
Besides the error correction, a fuzzy extractor scheme comprises also an additional step, called privacy amplification. First of all, it is used to extract random bits such that the PUF response turns out enhanced in uniformity (see Sect. 10.3.3), gaining information entropy, which is necessary whenever responses have to be used as secure keys [1, 15].
Fig. 10.10 The fuzzy extraction algorithm scheme. a Generation phase, to accomplish during the PUF enrollment in order to extract the key and the helper data. b Reproduction phase, which recover a noise version of the response and outputs the same key extracted in generation