Hybrid ATPG (Automatic Test Pattern Generation) algorithm

Zhang, Lizi1 ; Davoodi, Azadeh1

Published Aug 05, 2024 on Dryad. https://doi.org/10.5061/dryad.m0cfxppcm

Data files

Aug 05, 2024 version files 113.75 KB

circuits.zip

110.99 KB
README.md

2.76 KB

Abstract

The project "NN-for-ATPG" proposes a hybrid approach for Automatic Test Pattern Generation (ATPG) by integrating machine learning with the FAN algorithm. It includes implementations of two approaches, NN-Hyb and NN-All. It features two implementations, NN-Hyb and NN-All, which apply neural network models at selective and all circuit levels, respectively. Project files support the paper by providing the source code, circuit files, and scripts needed to reproduce key results, including comparisons for "all fault cases," "hard-to-detect faults," and runtime comparisons with and without acceleration. Minimal requirements include a C++ compiler and shell script execution capabilities.

README

File Structure

This repository includes the following directories and files (full file for download under Related works 'Software'):

README.md: instructions to compile and generate results in the paper
NN-Hyb: directory containing source code of our proposed approach
NN-All: directory containing soruce code of alternative approach compared in the paper
circuits: circuit files used in our experiments (also found on Dryad)
hard_faults: hard-to-detect faults per circuit used in results of Table II in the paper
script: scripts to generate results in Tables I, II, III in the paper

Installation

In the paper we have compared the following 3 approaches:

(1) Fan: base approach for ATPG used for comparison (2) NN-Hyb: our approach based on applying our neural network (NN) model during backtrace at select levels of the circuit during ATPG (3) NN-All: another alternative approach using our NN model, but at all levels of the circuit

To get an executable for (1), please download source code from https://github.com/hsluoyz/Atalanta and compile. (We did not include it because it is already published by another group.)

For (2) and (3), please compile (using the make command) from the NN-Hyb and NN-All directories, respectively.

Options

Once compiled, use the command ./atalanta [path to circuit file] from each directory. The options needed to generate the results of the paper are listed below:

-w: Disable lookup table (enabled otherwise if -w not specified)

-b n: Maximum number of backtracking (default: -b 10)

-f fn: Faults are read from the file fn. This options is only used to generate the results in Table II in the paper. Otherwise, if a fault file is not specified, the input is assumed to be all possible faults.

Run Experiments

The scripts in the "script" directory can be used to generate the results in the paper reported in Tables (1), (2) and (3). These are shown below for each case:

(1) Scripts for generating results in Table I: "All Faults Case":

AllFault_FAN.sh
AllFault_NN-All.sh
AllFault_NN-Hyb.sh

Table I reports fault coverage, ATPG effectiveness, number of backtracks and runtime. These quantities are reported by our program for NN-All and NN-Hyb.

For FAN's results, the tool does not directly report ATPG effectiveness. To compute it use this formula: (#redundant_faults + #detected_faults)/#total_faults

(2) Scripts for generating results in Table II: "Hard-to-Detect Faults":

H2D_FAN.sh
H2D_NN-Hyb.sh

(3) Scripts for generating results in Table III: "Acceleration with Lookup Approach":

Acceleration_NN-Hyb.sh

This script produces the results for NN-Hyb case, for both cases of with and without acceleration.