This repository provides the code and example data of the paper "Efficient Modeling Attack on MPUFs via Kronecker Matrix Multiplication"

To run the default EMBARK attack process, simply excute

python embark.py

To run EMBARK+COMB, excute

python integrated_embark.py

Our program will automatically generate CRPs, and then conduct attack process. If you need to import training and testing datasets separately from external files, uncomment the lines with loadTrainData and loadTestData.

# train_loader = loadTrainData("./data/5_64_MPUF_50k_train.csv", config.batch_size, config.device)
# valid_loader, test_loader = loadTestData("./data/5_64_MPUF_50k_test.csv", config.batch_size, config.device)

To conduct attack for $(n,k)$-MPUF, you only need to specify the data filename in the very first line of the main function in embark.py and integrated_embark.py. Our program will automatically prase the arguments in the filename.

The filename should be formatted like {k}_{n}_{PUF type}_{data size}.csv. For example, 5_64_MPUF_50k.csv leads to an attack for $(64, 5)$-MPUF with a dataset including 50k CRPs. For noisy environments, format filename like {k}_{n}_{PUF type}_{a}_{data size}.csv. The argument a indicates a noise level of $1/a$.

The following hyperparameters can be set in /utils/config.py.

• learning parameters: learning rate, batch size, loss function, etc.
• algorithm parameters: the coefficients of integrated EMBARK loss.

See /utils/config.py for more specifices.

/utils/model.py provides the implementation of models includeing

• PUF models: MPUF, cMPUF, rMPUF models used to generate CRPs
• attack models: EMABRK and integrated EMBARK models for MPUF, cMPUF and rMPUF