Evaluation of the COTS Google Coral Edge TPU processor for high-performance AI computing in space

CONTENT

• Introduction
• Market Analysis
• COTS-based architecture (proposed): FPGA+TPU
• Overview
• TPU Performance Evaluation
• TPU Radiation Testing
• Conclusion

Speaker: George Lentaris – NTUA Microlab

Dr. Lentaris holds a Ph.D. in Computing from the National & Kapodistrian University of Athens (NKUA), Greece, as well as two MSc degrees in “Logic, Algorithms, and Computation” and in “Electronic Automation”, with a BSc in Physics, all awarded by NKUA. His Ph.D. research (conducted during 2006-2011 at the Electronics lab, DST, Physics DPT) resulted in the thesis entitled “Parallel Architectures and Algorithms for Digital Signal and Image Processing,” which contributes to organizing parallel memories for graphics applications and designing motion estimation architectures for video compression.

The research interests of dr Lentaris are mainly in parallel algorithms and architectures for applications in signal and image/video processing. His work focuses on digital circuit design, and it includes computer vision algorithms, feature extraction, hyperspectral imaging, H.264/AVC encoding, motion estimation techniques, 1D/2D signal interpolation and filtering, OFDM synchronization and baseband processing, FFT implementations, parallel memories organization, as well as process variability and reliability of FPGA devices (incl. radiation testing).

Currently, he is a research associate at Microlab, NTUA, GR, working on high-performance embedded computing. More specifically, his work includes HW/SW co-design of computer vision to enhance autonomous rover and/or spacecraft navigation. He is using single-/multi-/SoC-FPGA and DSP platforms to accelerate stereo vision, visual odometry, and pose estimation of uncooperative targets for future missions of ESA. Additionally, he is working on the FPGA implementation of baseband DSP algorithms for novel 5G network technologies.