Codesign And Parallel Processing Lab
In Codesign and Parallel Processing laboratory, we are developing an embedded software design framework for parallel embedded systems. Also, we apply embedded software design methodologies to real-world applications such as embedded deep learning and the Internet of Things for performance analysis and optimization, simulation, and high-level specification.
Parallel Embedded Systems and Embedded SW Design Methodology
Our design methodology is based on formal-models such as SDF (Synchronous Dataflow) and FSM. The main benefit of using formal models is that we can estimate the system performance and resource requirement statically. We can also perform syntax analysis to detect some errors such deadlock and buffer overflow. By generating the target SW automatically from the high-level model specification, we can reduce the time and efforts for software development. This research includes architecture optimization, SW parallelization, and static performance estimation.
Embedded Deep Learning
Deep learning is computation-intensive and uses a lot of memory space, so deep learning algorithms need many optimizations and memory usage reduction to adopt a deep learning into real life embedded systems. We research state-of-the-art optimization techniques for deep learning and methodologies of applying existing embedded optimization methods to embedded devices. Also, for performance analysis on new embedded systems, we study simulation techniques for deep learning applications.
Internet of Things Platform
We research the Internet of Things platform which can integrate various computing resources connected to network or Internet. The computing systems of devices are abstracted and integrated into the platform as services. Based on this platform, a user can describe a composite service using voice or programming language with multiple services. Also, our research includes simulating the behavior of the Internet of Things and service scheduling among various devices.