Electrical and electronics engineering student Sueda Taner tests her sonar project at the Duke Reclamation Pond

Summer Undergraduate Research Program

Spend your summer conducting research with Duke ECE faculty

The Duke ECE Research Experience for Undergraduates (REU) brings students from around the world into our faculty research laboratories for nine weeks each summer.

These students work with faculty members and their research groups to tackle an innovative research project.

Admitted students receive:

  • Competitive monthly research stipend
  • On-campus housing
  • Travel allowance

The application period opens in the fall preceding the REU summer.

Read an article about students from a recent Duke ECE REU »


How to Apply

2020 Program

Click this link to apply for the 2020 Program!


Eligibility

The program is designed for undergraduates who are third-year students (juniors) in the spring before their REU summer. Exceptional second-year students (sophomores) may receive consideration.

Participating students should be majoring in electrical and computer engineering or a related relevant discipline.

Domestic U.S. and international students are invited to apply. 

Important Note for International Applicants

Accepted international applicants must be issued the appropriate visas to participate. You should be aware of the requirements and policies associated with your visa.

Applicants requiring J-1 visa entry can find more about J-1 policies (including the possible requirement of two years of home-country physical residence) through:


Dates and Stipend

  • Program Dates: May 25- July 24, 2020 (9 weeks)
  • Stipend: $500 per week
  • Travel Reimbursement: Up to $600
  • Housing: Shared rooms on the Duke campus, TBD

Research Opportunities

These projects are available in the summer of 2020:

Smart Garment for Automatic Stress and Anxiety Detection and Reduction 

This is an open-ended project that involves designing and building a working prototype of a wearable system that can detect and reduce stress and/or anxiety. One example of such a system could be a sensor (or sensor network) that can detect stress/anxiety and in linked to a compression garment that gets automatically triggered when the person is stressed/anxious.
 
Student work will include reviewing the literature, building a prototype using sensors, actuators, and a microprocessor/Arduino, while using machine learning techniques to detect stress/anxiety from sensor data.
 
Experience in machine learning and embedded systems is recommended.
 
Faculty Advisor: Prof. Rabih Younes (rabih.younes@duke.edu)

Edge Computing-Assisted Next-Generation Mobile Augmented Reality

 

Mobile augmented reality (AR), which overlays digital content on the real-world scenes surrounding a user, is bringing immersive interactive experiences where the real and virtual worlds are tightly coupled. However, while already immersive, modern AR experiences suffer from multiple limitations, including high resource consumption, limitations in multi-user experiences, and general lack of adaptiveness and intelligence. This project is focused on creating next-generation augmented reality with the aid of edge computing, a paradigm where computing resources are moved from the cloud closer to the end users. We will explore the use of edge computing for improved multi-user AR experiences and improved scene understanding.   https://maria.gorlatova.com/current-research/

 

Student work will include the development and performance evaluation of next-generation augmented reality with Unity, mobile phones, and/or augmented reality headsets. It will also include development and performance evaluation of assistive client-server architectures for next-generation augmented reality.

 

The project requires general computing skills and coursework in networking and communications. Experience with Unity, Android development, client/server architectures would be advantageous but is not required.

 
 

Faculty Advisor: Prof. Maria Gorlatova (maria.gorlatova@duke.edu)


Interactive GUI for Composing Accelerators for the Cloud

 

Designing and deploying accelerators in this post-Moore's Law era is still cumbersome and time-consuming. In this project, we want to provide users with an easy-to-use, interactive GUI to compose accelerators by moving the hardware blocks and connections around visually to form communication channels amongst the accelerators, as well as between the accelerators and the memory subsystem. The students will design the GUI and translate the movements the users make in the GUI into already defined hardware constructs in Chisel (a Scala-based Hardware Description Language). The goal of this project is to move towards the vision of making designing and using hardware-accelerated systems almost as easy as it is to write and use software.  https://apexlab-duke.github.io
 
The students will design the GUI and translate the movements the users make in the GUI into already defined hardware constructs in Chisel (a Scala-based Hardware Description Language).
 
Programming Language: Java, Javascript, and/or other languages that can be used to design an interactive GUI. Chisel or Verilog knowledge desired but not necessary.

 

Faculty Advisor: Prof. Lisa Wu Wills (lisa@cs.duke.edu)


Accelerating Genomic Simulation in the AWS Cloud

 

Genomic simulation is increasingly popular as a way to gain an understanding of specific genomic datasets or validating constructed biological models. An example of such simulation, which formed a novel analysis framework processing data collected on an international scale had led to significant findings in Cholangiocarcinoma (bile duct cancer). This analysis framework integrates genomic and epigenomic analyses and incorporates experimentally derived protein- DNA binding affinities ad pathway information to identify new driver genes, noncoding promoter mutations, and structural variants. However, the simulation is slow, having research analysts waiting up to 500 hours to sift through just 1,000 mutations for just one of many analysis stages and for just one of several cancer datasets. The amount of cancer mutations to analyze is highly dependent upon the genomic regions where the mutations are identified. It is typical to have more than one million mutations to analyze for some genomic regions, making computer simulation a significant bottleneck in making forward progress in genomic research. We aim to investigate and discover primitive computation operations performed in genomic simulation and build hardware-software co-designed accelerated systems to expedite genomic simulation and advance scientific discoveries in cancer research.  https://apexlab-duke.github.io

 

Students will be investigating ways in computer software and in computer hardware (via accelerator architecture) to implement and deploy an accelerated system in the AWS cloud that will perform genomic simulations order(s) of magnitude faster.

 

Programming languages: C/C++, Python required; Chisel or Verilog desired Coding environment: Unix/Linux; IDEs are okay but basic Unix command line familiarity needed; Git strongly desired Courses: have taken CS/ECE 250 Undergraduate Computer Architecture

 

Faculty Advisor: Prof. Lisa Wu Wills (lisa@cs.duke.edu)


Exploring New Electrical Contact Interfaces to Nanomaterials

 

Nanomaterials are attractive for future electronic device applications owing to their atomic thinness and unique electrical properties. These include 1D carbon nanotubes, 2D graphene, and many other 2D crystals that are semiconducting. One of the foremost challenges for nanomaterial-based devices is the inconsistency and relatively poor performance of the contact interfaces. While there has been some progress in improving the metal-nanomaterial contacts, much work remains and this project will explore new approaches to establishing electrical interfaces with a variety of nanomaterials. In this project, the student will perform work in the cleanroom here at Duke, learning the basics of nanofabrication and electrical characterization. Several new contact structures will be studied by fabricating transistors from the nanomaterials and then characterizing the resultant properties via electron microscopy, atomic force microscopy, and spectroscopy techniques. These material characterization results will then be correlated with the electrical characterization of the devices. The REU student will be an active contributor to this research and also be expected to take part in discussions where results will be analyzed and new ideas potentially formulated for inclusion in the project.
 
The student will be trained on all needed equipment and characterization processes by a PhD student mentor. This is hands-on, laboratory-based, experimental work, so the student should be comfortable with working in a cleanroom, using fume hoods, fabricating & testing electronic devices, etc.
 
The student should have some previous knowledge and experience in solid-state physics including carrier transport in semiconductors, previous knowledge and/or interest in electronics and nanomaterials, and be competent in operating complex tools. A student without this background will still be considered so long as there is substantial interest in learning about this research area. They should also be self-motivated and maintain a strong work ethic in terms of commitment and follow-through. A collaborative, team player is a must.

 

Faculty Advisor: Prof. Aaron Franklin (aaron.franklin@duke.edu)


Paper-Based Printed Electronics and Biosensors

 

Printing is a potentially low-cost, highly customizable approach for producing electronics for applications such as biosensors. However, most electronic inks require processing external from the printer that often includes high-temperature conditions. These post-printing treatments limit the type of substrates that can be used for printed electronics. In this project, the REU student will work with a PhD student on developing inks and printing processes that allow for the direct printing of electronic devices without external processing. Inks will be formulated from various nanomaterials, including carbon nanotubes (CNTs), graphene and other 2D materials, and printing will be carried out using an aerosol jet printing technology. The goal will be to realize fully printed devices on paper substrates, with full electrical and materials characterization of the devices. Characterization will include electrical testing & analysis, atomic force & electron microscopies, and other material analysis techniques. Time permitting, this project may also involve the printing of biosensing devices onto the paper substrates to demonstrate paper-based electronic biosensors. Ultimately, the REU student involved in this project will gain experience with nanomaterials, printed electronics, device characterization, and biosensing.  franklin.pratt.duke.edu
 
The student will be trained on all needed equipment and characterization processes by a PhD student mentor. This is hands-on, laboratory-based, experimental work, so the student should be comfortable with working in a chemical lab, using fume hoods, fabricating & testing electronic devices, etc.
 
 
The student should have some previous knowledge and experience in solid-state physics including carrier transport in semiconductors, previous knowledge and/or interest in electronics and nanomaterials, and be competent in operating complex tools. A student without this background will still be considered so long as there is substantial interest in learning about this research area. They should also be self-motivated and maintain a strong work ethic in terms of commitment and follow-through. A collaborative, team player is a must.

 

Faculty Advisor: Prof. Aaron Franklin (aaron.franklin@duke.edu)


Automatic Exploration of Machine Learning Models

 

Machine learning has enabled remarkable progress over recent years on both vision and language tasks. Traditionally, machine learning models are carefully designed by human experts to achieve state-of-the-art performance. However, this manual design process is usually time-consuming and error-prone. With the rise of AutoML and Meta-learning, novel ML models are automatically designed by computer programs in an end-to-end fashion. These models usually outperform hand-crafted models by a large margin. Therefore, it is necessary to spend some efforts on architecture engineering of ML models. The primary interest of this research is to devise algorithms to explore novel but powerful machine learning models for a wide range of applications, such as image recognition, object segmentation, and speech recognition. This includes the devising of a search algorithm (how should we search for the architecture), the search space (which architecture can be represented in principle), and the performance estimation strategy (how should we evaluate our candidate architectures on a dataset). Furthermore, this research may also address some common problems in the current AutoML community, such as the explosion of model size and search cost. A submission to the top ML conference will be made if remarkable progress has been made.
 
  • Familiar with object-oriented programming languages (e.g. C++, Python). Experience with any deep learning framework (e.g. TensorFlow, PyTorch, MXNet) is preferred
  • Strong interest in deep learning. Experience with Computer Vision (CV) or Natural Language Processing (NLP) is preferred
  • A strong motivation and a good team spirit in research

 

Faculty Advisor: Prof. Hai Li (hai.li@duke.edu)


Machine Learning for Medical Imaging

 

This is a collaboration between radiology (Joseph Lo), ECE (Larry Carin), and Duke Forge (Ricardo Henao). Our goal is to develop a framework for classifying multiple organs and multiple diseases from computed tomography (CT) scans. This project will explore many aspects of deep learning including natural language processing of radiology report text, segmentation of multiple organs, and classification of multiple diseases. We are drawing from a database of up to 400,000 CT scans from Duke Health.
 
Students will develop machine learning models for medical image analysis.
 

Prior experience with computer vision and machine learning with image data is necessary because of the very advanced nature of this project. Specific expertise with Linux, python, and Tensorflow is desirable.

 

Faculty Advisor: Prof. Joseph Lo (joseph.lo@duke.edu)


Enable Efficient Deep Learning on Mobile Devices

 

We are investigating how to support efficient training and inference for deep learning models on mobile devices. Enabling on-device learning can facilitate in offering personalized service, preventing privacy information leakage, etc. Although existing deep learning frameworks, such as TensorFlow and PyTorch, support performing inferences using pre-trained deep learning models on mobile devices, the capability of on-device training is not supported yet. It is necessary to extend existing deep learning frameworks to support operations that are related to training on mobile devices. In addition, due to the limited computation resources and data resources, how to train a deep learning model with less data is another challenge to enable on-device learning. Existing knowledge transfer techniques and few-shot learning algorithms may provide potential solutions. With the rapid development of accelerators on mobile devices, such as GPUs, we need to explore how to leverage such available resources to accelerate various deep learning operations on mobile devices. For example, how to accelerate the backpropagations during the training process by utilizing existing parallel computing and mathematic computation libraries needs to be further explored.
 
The project requires general software-development skills. Experience with C++, Python, TensorFlow, PyTorch, and/or Android-based app development would be potentially advantageous but is not required.
 
 
Faculty Advisor: Prof. Hai Li (hai.li@duke.edu)

Certification Testing for Driver Monitoring Systems

 

This project entails developing a testing protocol and gathering data to determine the limits of commercial driver monitoring systems.
 
Students will assist in the gathering of driver monitoring information in the operation of cars and help develop automated analytical tools.
 
Python and/or C, driver license, and the ideal candidate would have computer vision experience although this is not a hard requirement
 
 

Faculty Advisor: Prof. Mary Cummings (m.cummings@duke.edu)


Human-Guided Acoustic Drone Detection

 

This project will focus on developing and testing an automated machine learning modeling system that works with users to update the acoustic profile of a region prone to illegal drone activity such as prisons.
 
Develop and refine neural nets that take input from users and test the fidelity of the algorithms for drone detection in real environments.
 
Background in python and some machine learning is preferable, as well as work with android applications
 
 
Faculty Advisor: Prof. Mary Cummings (m.cummings@duke.edu

 

Back to top