Thermal Modeling
Modern architectures have high on-chip temperature due to the high power density as well as poor heat dissipation. The high on-chip temperature causes thermal problems such as malfunction and permanent damage. To avoid the thermal problems, we should consider the on-chip temperature of architectures from the early-design stage. For accurate thermal analysis, our research group investigates thermal modeling for modern architectures, with consideration of various cooling solutions.
System-level Thermal Management
With the advance of process technology, the power density of microprocessors is increasing. The high power density leads to the high temperature which causes thermal problems such as temporary malfunction or permanent damage on the hardware. To solve such thermal problems, our research group investigates system-level thermal management techniques based on the DVFS (Dynamic Voltage and Frequency Scaling) and task migration.
System-level Power Management
Recent processing units often consume a considerable amount of power due to temperature-dependent leakage power as well as high dynamic power. For power management of such processing units, systems adopt power management techniques based on software, especially OS (Operating Systems), as well as hardware. Our research group investigates system-level power management techniques for modern processing units.
Reliability
Modern DRAM has become increasingly vulnerable to errors due to high on-chip temperature, process technology scaling, and increasing DRAM density. To mitigate such reliability problems, data centers employ various reliability enhancement techniques such as ECC (Error Correction Code) and rowhammer mitigation schemes. However, such schemes lead to substantial performance and/or storage overhead. Our research group investigates strong and low-overhead reliability enhancement techniques for memory solutions such as DDR5 and HBM (High Bandwidth Memory).