Projects

This project led by Ashutosh Ghimire was accepted for Lecture Presentation in IEEE NAECON conference held on 2023 in Dayton OH. It presents a novel approach to detect Hardware Trojans by combining unsupervised machine learning and side-channel analysis, leveraging unique features from on-chip ring-oscillator networks to identify anomalies through unsupervised clustering. Evaluation on FPGA chips with Trojan insertion demonstrated exceptional accuracy, surpassing alternative methods with a 99% accuracy rate. The centroid-based clustering model exhibited superior performance, with a slight edge in false positive rate and an f1 score. This project contributes to enhancing trust in semiconductor IC supply chains by offering a fresh perspective on Hardware Trojan detection, eliminating the dependency on golden chips, and providing a reliable and practical solution for identifying compromised ICs.

SRAM PUF Manufacturer Identification

This project led by Harshdeep Singh received the Best Poster Award at the IEEE VLSID 2024 conference held in July 2024 in Knoxville, TN. It addresses the critical issue of identifying counterfeit SRAM memory chips manufactured in untrustworthy facilities, which compromise the integrity and security of the supply chain for microelectronic systems. By developing an AI-assisted hardware-oriented solution, this project leverages

Physical Unclonable Functions (PUFs) and specifically SRAM PUF technology to authenticate devices and generate secret keys. The innovative framework uses machine learning models to exploit process variations in SRAM memory cells, enabling efficient identification of microelectronic device manufacturers without cumbersome procedures. Validated through a study involving 345 SRAM chips from five reputable manufacturers, the AI-driven technique achieved high accuracy and an F1 score of 97% in identifying device manufacturers, significantly enhancing the reliability and trustworthiness of the IC supply chain.

Adversarial Attack Resilient ML-Assisted Hardware Trojan Detection

The project led by Mohammed Alkurdi and Ashutosh Ghimire was presented at the IEEE VLSID 2024 conference held in July 2024 in Knoxville, TN. It aims on improving the resilience of hardware Trojan detection machine learning models against adversarial attacks. The methodology being followed is to create a surrogate machine learning mode that has the same accuracy as recent research. Afterwards, perform a feature-space adversarial attack on the model, to generate adversarial examples, which are then used in the retraining of the model to increase its resiliency against similar attacks. The threat model of our current project is that an adversary can modify his Trojan, then insert it into the chip design, afterwards check if the Trojan is detected by the machine learning model. If the trojan is detected, then modify the Trojan again and insert it again, else we can consider the adversarial attack a success.

Object detection with Tiny ML

The project “Object detection with tiny ML” is led by Mounika Thatikonda. It aims to detect Vehicle Number plate and Helmet . It is implemented using Tiny ML Yolov8 model with the help of Raspberry pi through optimization techniques like neural architecture search and model compression, focusing on perceptible metrics like inference latency and energy consumption.  To verify the performance of the YOLOv8, the experiments are conducted on a  RHNP dataset, which contains four categories: rider, helmet, no-helmet, and license plate. The accuracy(mAP) of the Yolo V8 model without optimization is 56%. The future work includes adding security to this.

Explainable AI and Chronic Kidney Diseases

The research study was led by K M Tawsik Jawad which largely focuses on interpreting the decision-making process of the ensemble tree models for the prediction of chronic kidney disease. This research is organized to aid the physicians and patients in general to make them understand the factors affecting the kidneys for the chronic kidney disease. The end goal is to explain to them from the biological features on why this disease has been diagnosed in their body and what factors have played stronger roles in the occurrence of this disease. The research findings are analyzed, validated with nephrologists and also compared with existing methodologies.