Advancing the customization and unlocking the potentials of future computing systems
Prof. Iraklis Anagnostopoulos
Electrical and Computer Engineering Department, Southern Illinois University Carbondale
- Date: Jan 22nd, 2021, 4:30 pm
In the past few years, we witnessed embedded systems to widely adopt the many-core computing paradigm for satisfying thermal/power constraints and reducing design complexity. This technology push was so strong that currently 99.9% of the Android devices (2.5B devices) have multiple cores. Even though this increase in the number of processing units was guided by the Dennard's scaling, leakage power becomes dominant in the nano era, further increasing the on-chip power. Platform heterogeneity prevails as a solution to the power/performance dilemma of modern devices imposed by parallel applications and technology scaling. Particularly, heterogeneity can be classified into per-formance and functional. Performance heterogeneity refers to the integration of processing elements with different performance and power properties but the same Instruction Set Architecture (iso-ISA cores). Functional heterogeneity refers to the integration of specialized processing elements with dif-ferent ISA. This includes GPU cores, Digital Signal Processing (DSP) blocks, and recently Neural Proces-sing Units (NPUs). In this talk, we are presenting a method for predicting the power consumption and performance of concurrently executing applications on modern Chip Multi-Processor systems along with a unified framework for enabling formal property exploration and integrating robustness analy-sis. Finally, we will also present our research interests and activities in the area of heterogeneity-aware optimizations.
Iraklis Anagnostopoulos received the Ph.D. degree from the Microprocessors and Digital Systems La-boratory, National Technical University of Athens. He is currently an Assistant Professor with the Electrical and Computer Engineering Department, Southern Illinois University Carbondale. He is the Director of the Embedded Systems Software Laboratory, which works on run-time resource manage-ment of modern and heterogeneous embedded many-core architectures, and he is also affiliated with the Center for Embedded Systems. His research interests lie in the area of constrained application mapping for many-core systems, design and exploration of heterogeneous platforms, resource con-tention minimization, and power-aware design of embedded systems.