I am an Assistant Professor in the School of Electrical Engineering and Computer Science at Washington State University. I obtained my Ph.D. in Computer Engineering from Arizona State University in 2020 (Advisor: Prof. Umit Y. Ogras). Before ASU, I was a senior software engineer at Samsung Research and Development Institute, Bangalore, India.
I am primarily interested in the design, optimization and application wearable IoT devices, heterogeneous mobile devices, flexible hybrid electronics, and health monitoring. Some of the topics in these areas that I am working on include energy management, energy harvesting, human activity recognition, dynamic thermal and power management, and resource management. I am also interested in applications of machine learning, dynamic programming, and convex optimization in real-world problems.
November 2021: Two papers accepted for publication in DATE 2022. Congratulations to Nuzhat and Dina!
Serving as technical program committee member in DAC 2022
Ferdinand A. Stanchi Fellowship, Arizona State University, 2019
Engineering Graduate Fellowship, Arizona State University (2018, 2019)
Wearable IoT Devices
Wearable IoT devices have the transform multiple facets of our life by enabling applications such as unibuitious sensing, smart healthcare, and robotics. In my research, I focus on design and optimization of wearable IoT devices for health monitoring applications. In particular, my current research projects in this area include:
Flexible Hybrid Electronics Devices
The emerging Flexible Hybrid Electronics (FHE) devices combine the form-factor benefits of pure flexible substrates and performance benefits of rigid CMOS ICs. Using the FHE technology we desinged the experimental protytpe shown on the right. It combines a TI MCU, Invensense Motion Processing Unit, energy harvesting using photovoltaic cells, and communication using BLE. The device can be easily worn as a patch on the body, thus enabling continuous monitoring of the user motion.
We have used this prototype in OpenHealth, our open-source hardware/software platform for health monitoring.
We also implemented applications such as gesture recogniton and human activity recognition.
In the future, we plan to use the derivatives of this board to conduct experiments with movement disorder patients. Relevant papers:D&T'19,
Energy Management for Wearable IoT Devices
Limited battery capacity is one of the major challenges for the widespread adoption of wearable IoT devices.
Larger batteries make the device heavy and bulky, while smaller batteries provide limited lifetime.
Therefore, we use solar energy harvesting to achieve energy-neutral operation for wearable device.
Our approach, shown on the left side, first models the energy that can be harvested over a given period.
Then, we use a dynamic programming approach to optimally allocate the harvested energy over the day.
This ensures that the total energy consumed is equal to the harvested energy, thus eliminating battery charging requirements.
Our algoritm also accounts for variations in the harvested energy and usage at runtime so that unforseen circumstances can be accounted for.
We continue to work in this area by adding energy harvesting modalities and developing new approaches for energy management.
Human activity recognition aims to identify daily human activities, such as walking, sitting, and jogging.
In this research, we used a wearable setup consiting of a stretch sensor and accelerometer to identify seven activities and transitions between them.
A demonstration of our application is shown in the video on the right.
We also developed an online learning approach to fine-tune the weights of our classifier for new users.
The datasets of this work are available online on my GitHub page.
We plan to extend this approach to monitor the symptoms of movement disorder patients.
Dynamic Thermal, Power, and Resource Management in Mobile Devices
Mobile devices are being used in multiple aspects of our lives including communication, healthcare,
entertainment, and education. Modern mobile systems integrate mutiple CPU types, GPUs, and accelerators to
achieve competitive performance and power consumption. The usage multiple resources introduces challenges in the thermal, power,
and resource management on these devices. My research projects in mobile devices try to address these issues.
Dynamic Thermal and Power Management
The power consumption and temperature form a positive feedback loop, which causes a continuous increase in
both until a steady state is reached. More specifically, the dynamic power consumption of the mobile device
increase the temperature that increases the leakage power consumption. In a stable system, this feedback loop
converges to a stable steady state temperature, while a thermal runaway occurs in an unstable system.
A thermal runaway causes permanent damage to the device, therefore we need to analyze the stability of the system at runtime.
My work addresses this question by developing closed-form conditions for the stability of the system.
Modern mobile devices increase the operating frequency of resources to meet the performance needs of the user.
The increase in frequency causes an increase in the skin and device temperatures. Therefore, the devices
need runtime algorithms to maintain skin and device temperatures under safe limits.
Naively reducing the frequency of the resources is not sufficient as it leads to performance degradation.
In my research, we developed predictive algorithms to maintain the temperature of the device within safe limits with minimal
impact on the performance.
We continuously predict the temperature in the future and take action whenever we predict that the temperature will rise
beyond a given threshold. Moreover, we use optimization techniques to ensure that our actions cause minimal performance impact.
Our algorithm reduces the number of thermal violations by one order.
We continue to work in this area by developing algorithms that aim to reduce the frequency for the applications that cause
Illustration of the prediction of thermal steady state using simulations (black lines), experimental measurements (red line), and analytical prediction (red triangles)
Illustration of the power modeling methodology used in the thermal management work.
Dynamic Resource Management using Imitation Learning
Increase in the number of resources in mobile devices leads to a large number of possible configurations (number and frequency of
active cores) for running applications.
Each application has unique resource requirements for its optimal execution. Furthermore,
the optimal configuration changes at runtime as a function of the change in the application characteristics.
The default algorithms in mobile system use utilization to make resource management decisions.
However, utilization alone does not provide insight into the application characteristics.
Therefore, in our research we use performance counters to make runtime resource management decisions.
Using these counters, we design imitation learning policies to choose the resource configurations at runtime.
We plan to extend these methods to work on embedded high performance computing nodes. Relevant papers:TODAES'20,
EE 524 / CptS 561 Advanced Computer Architecture - Fall 2020
CptS 260 Introduction to Computer Architecture - Spring 2021
Nuzhat Yamin, Ph.D. Student, Fall 2020 - present
Dina Hussein, Ph.D. Student, Spring 2021 - present
ECO: Enabling Energy-Neutral IoT Devices through Runtime Allocation of Harvested Energy
Yigit Tuncel, Ganapati Bhat, Jaehyun Park, Umit Y. Ogras IEEE Internet of Things Journal (Accepted, in press), 2021 arXiv
MGait: Model-Based Gait Analysis Using Wearable Bend and Inertial Sensors
Sizhe An, Yigit Tuncel, Toygun Basaklar, Gokul Krishnakumar, Ganapati Bhat, Umit Y. Ogras ACM Transactions on Internet of Things (Accepted, in press), 2021 arXiv