MuxLink: Circumventing Learning-Resilient MUX-Locking Using Graph Neural Network-based Link Prediction

Lilas Alrahis, Satwik Patnaik, Muhammad Shafique, and Ozgur Sinanoglu

About

MuxLink is a link prediction-based attack on learning resilient logic locking. This repository contains the python implementation of MuxLink attack in addition to our own implementation of the deceptive logic locking scheme (DMUX).

Contact Lilas Alrahis (lma387@nyu.edu)

Setup

Step 1: Install the default GNN model (i.e., DGCNN)

$ git clone https://github.com/muhanzhang/pytorch_DGCNN
$ cd pytorch_DGCNN/lib
$ make -j4
$ cd ../..

Step 2: Install Required Packages

Install PyTorch
Install numpy, scipy, networkx, tqdm, sklearn, gensim

Usage

Attacking DMUX

1) Lock a design

Example, lock the c1908 ISCAS benchmark with key size of 32

$ cd ./MuxLink/DMUX_Locking
$ python3 convert_DMUX.py c1908 32 ../data/c1908_K32_DMUX

convert_DMUX.py is a Python script that reads a circuit in Bench format and locks it using DMUX. It will convert the design into a graph. It assigns unique numerical IDs (0 to N-1) to the nodes (gates). N represents the total number of nodes (gates) in the design.
It will generate a directory ../data/c1908_K32_DMUX which includes: -- The extracted features will be dumped in feat.txt. The ith line in feat.txt represent the feature vector of the node ID = the ith line in count.txt -- The existence of an edge i between two vertices u and v is represented by the entry of ith line in links_train.txt -- The links_test.txt and link_test_n.txt are created to identify the edges exclusive to the testing set. links_test.txt includes all the true MUX connections while link_test_n.txt includes all the false MUX connections -- The cell.txt file includes the mapping between node IDs and gate instances -- The c1908_K32.bench file represents the locked circuit

2) Train MuxLink

$ cd ../
$ python Main.py --file-name c1908_K32_DMUX --train-name links_train.txt  --test-name links_test.txt --testneg-name link_test_n.txt --hop 3  --save-model > Log_train_c1908_DMUX_K32.txt

3) Get the predictions

$ python Main.py  --file-name c1908_DMUX_K32 --train-name links_train.txt  --test-name links_test.txt --hop 3  --only-predict > Log_pos_predict_c1908_DMUX_K32.txt
$ python Main.py  --file-name c1908_DMUX_K32 --train-name links_train.txt  --test-name  link_test_n.txt --hop 3  --only-predict > Log_neg_predict_c1908_DMUX_K32.txt

The likelihoods for the links will be dumped in links_test_3__pred.txt and link_test_n_3__pred.txt. Here, 3 represents the hop size

4) Parse the predictions

$perl break_DMUX.pl c1908_K32_DMUX 0.01 3

Here, 0.01 is the threshold value (th) explained in the paper. it can be between 0 and 1.
3 represents the hop size.

Citation & Acknowledgement

If you find the code useful, please cite our paper:

MuxLink 2022:

@INPROCEEDINGS{muxlink,
  author={Alrahis, Lilas and Patnaik, Satwik and Shafique, Muhammad and Sinanoglu, Ozgur},
  booktitle={2022 Design, Automation Test in Europe Conference Exhibition (DATE)}, 
  title={MuxLink: Circumventing Learning-Resilient MUX-Locking Using Graph Neural Network-based Link Prediction}, 
  year={2022},
  pages={702-707},
 }

We owe many thanks to Muhan Zhang for making his SEAL code available.

lilasrahis/MuxLink