...

I finished reading completely:

•  Chapter 4
•  Chapter 5
•  Chapter 15

Chapter 3: The Basic Applications

...

PUFs can reduce costs by reducing fabrication complexity and by providing the ID from the intrinsic properties of the chip.

Since the output of PUFs is noisy, typically the identification system has to be error tolerant. Depending on the tolerance of the system, error correction codes can even be omitted. The measure of the tolerance for binary output data is called Hamming distance.

To increase the probability of correct ID assignment the number of output bits can be raised.

...

Approach: Reducing Errors Using More Than One PUF Cell → IMPORTANT!!!

Example: three cells are used to produce one bit of output. The output is determined by a majority decision: If there are more ones than zeros (011, 101, 110, 111) in the bit string, a “1” appears at the output. Otherwise, if there are more zeros than ones (000, 001, 010, 100), a “0” is generated. There are two bit strings (000, 111) where two bits have to change in order to change the output value. Furthermore, there are six bit strings in which one bit has to change in order to change the majority decision.

...

All complex EECs become difficult to handle as soon as the amount of energy is highly limited as it is for example on RFID tags. In such cases other solutions are required. For this reason the repetition code, a combination of repetition and Hamming code, and a combination of repetition and BCH code is analyzed. For their implementation simplicity and correction capabilities, such codes are an alternative to highly complex codes.

...

k is the number of bits in the original signal and l is the number of bits of the encoded block

Chapter 15: Using the SRAM of a Microcontroller as a PUF

...

It can be seen for example, that a repetition code with repetition factor of 31 reduces the initial error rate of 10% to an error rate (BER) of 6.85E−07%. The correction capabilities of the repetition factors 21 and 11 can be seen in the BER after correction: 1.35E−04% and 2.96E−02% respectively.

...

Measurement Results

Here, 16 kB were assumed to be sufficient to provide enough bits for data and repetition code.

The following parameters are determined: mean value, error rate, the temperature behavior(error rate at different temperature corners), correlation between bits, correlation between chips and memory effect. The analysis results are presented in the following sections.

...

To test the implementation the key generation was repeated 1,000 times at different temperatures in the range between 0 ◦C and 80 ◦C. No errors were produced by the PUF in all of the measurement runs after error correction. This is the expected result, since the theoretical error rate of a PUF with 8% BER of a single bit, 2048 bit key, and a repetition factor of 31 can be determined to be 6.85E−07%.

→ check the article https://ieeexplore.ieee.org/abstract/document/6060013/ for more details about this experiment.

Summary

The analysis showed that before implementing a PUF on an internal SRAM the SRAM has to be checked carefully toward its feasibility as a PUF: both, the inter-chip and the intra-chip Hamming distance have to be within a predefined range. Furthermore, the memory effect of cells should be checked, especially if the cells are also used during regular usage of the SRAM.

...