Embedded Machine Learning

We study the optimization of machine learning (ML) workloads on top of embedded or cyber-physical systems, which are typically associated with a number of non-functional design concerns such as limited power/memory budgets or real-time constraints. Our research focuses on the co-optimization of neural network (NN) model and hardware architecture. We also study the distributed NN inference and online learning for distributed embedded systems such as wearable devices.

<multi-phase SIMD-aware CNN pruning technique for embedded microprocessor (CODES-WiP, 2019)>

Related projects

  • NEuromorphic computing Software plaTform for artificial intelligence systems (NEST, Sponsored by IITP)
  • Research of human body communication technique and its application for  future soldiers (FNT-31, Sponsored by ADD/DAPA)

Satellite Embedded Systems

Miniaturized satellites such as CubeSats are becoming increasingly popular both in the private and public sectors. While the harsh operating conditions, e.g. extreme environmental temperatures, of satellite systems hugely compromise their life-time and reliability, it is not common to have thermal control or insulation as a protective solution due to the cost problem. We study the cost effective solution to the enhancement of the reliability and life-time of the satellite embedded systems. Further, we seek to find a SW solution to maintain the battery healthiness, which is known to be the bottleneck of life-time of satellite systems. 

<Temperature-aware lifetime enhancement of satellite embedded systems (Sensors 2019)>

Related projects

  • Satellite Information Processing and Convergence Application Service (SICAS, Sponsored by IITP)

Transiently-Powered Real-Time Systems

Modern embedded systems are often operated remotely for a long time without constant power sources or periodic maintenance. One solution to this unreliable power supply is a transiently powered system which is equipped with an energy harvester. In such transiently powered systems, it is common to utilize non-volatile memory (NVM) or non-volatile processor (NVP) to preserve the system status even in case of sudden power shortages. We study the analysis and optimization of the transiently powered systems with NVM/NVP. In particular, we are interested in guaranteeing the real-time schedulability even with the existence of uncertainty in power supply.


<RTC-based real-time schedulability analysis of transiently powered processors with NVMs (IEEE TC 2020)>

Related projects: 

  • Ultra Low-Power Real-Time Image Processing (Sponsored by NRF Korea)

Design, Optimization, and Analysis of Multi-Processor Systems

Modern computer system design practice is moving towards (homogeneous and heterogeneous) multiprocessor system-on-a-chip (MPSoC) architecture. How to program such MPSoCs in efficient and predictable way is a huge challenge. We are particularly interested in temperature-aware and fault-tolerant design of such multi-core systems.

<A case-study of the power-optimization of multi-core embedded system: stereo-vision for drone obstacle avoidance  (DAC 2018, IEEE Access 2019)>

Related projects:

  • EURETILE: EUropean REference TILed architecture Experiment
  • Pro3D: Programming for Future 3D Architecture with Many-Cores
  • HOPES: Hope Of Parallel Embedded Software development

Hardware/Software Co-design of Embedded Systems

We investigate a large area in the hardware/software co-design methodology for mobile embedded systems: specification, simulation, optimization, verification, and prototyping. In particular, we are interested in the model-based design of mobile embedded systems.

<Power-performance co-optimization of OpenCV applications on multi-core embedded microprocessor (Symmetry 2019)>

Related projects:

  • Research of Non-Volatile Memory for IoTs (Completed, Sponsored by NRF Korea)
  • PeaCE (Ptolemy extension as Codesign Environment, Completed): Co-design environment for rapid development of heterogeneous digital systems 

Embedded Software testing

As the importance and complexity of embedded software have increased in recent years, we conduct research on software testing and automation in a variety of areas (e.g. digital health care and weapons systems).

<A Test-case generation framework from Software Requirement Specification (SRS) written in Simplified Technical Korean (IEMEK Journal (Korean domestic) 2019)>

Related projects: 

  • Industrial Infrastructure Program for Digital Health-Care (DHC, Sponsored by MOTIE)
  • Research and Development of the Testing Methodology of Military Weapon Systems – Sponsor (Completed, Sponsored by ADD)

Mobile and Swarm Robots

We are interested in how multiple mobile agents or robots collaborate to accomplish a cooperative mission. We mainly focus on how our experiences on the single computer system design methodology can be extended to swarm robotics. Specifically, we are investigating the mission specification model for swarm computing.