Research Topics

Carbon-efficient & Sustainable Computing

Computers, data centers, and devices use a lot of energy, often from fossil fuels, which release carbon and cause climate change. By making computing carbon neutral/efficient, we can protect the environment by cutting carbon emissions.

Our carbon-efficient computing aims to designing and operating computing systems to minimize their carbon footprint, aiming for greater energy efficiency and reduced environmental impact. Our dream is not just carbon-effiency but the true carbon neutrality in computing. The following are our directions:

Undervolted/Near-Threshold Computing

Reducing the operating voltage to near-threshold levels greatly improves carbon efficiency by lowering power consumption, which directly reduces carbon emissions. However, this also increases the risk of noise and errors, affecting system reliability. Our research focuses on exploring the impact of undervolting on chip performance and developing robust strategies to manage noise and errors, ensuring reliable operation while maximizing energy savings and reducing the carbon footprint of computing systems.

Approximate Computing

To minimize power consumption and area costs, we utilize approximate circuits that prioritize efficiency by delivering "good enough" results instead of precise calculations. This trade-off reduces energy use and chip size, which in turn lowers the carbon emissions associated with manufacturing and operating these devices. By focusing on efficient computation, we aim to contribute to a more sustainable and environmentally friendly technology landscape.

Sustainable Computing

Devices can experience permanent defects over time, affecting performance accuracy. We research methods to predict and tolerate these potential faults. By enhancing fault tolerance, we aim to extend the operational lifetime of these devices, reducing the frequency of replacements and the need for new semiconductor production. This reduction directly contributes to lowering carbon emissions associated with manufacturing, supporting more sustainable and carbon-efficient computing practices.

Edge Intelligence for Disasters

This topic leverages edge computing and AI to enable real-time disaster monitoring, prediction, and response. By processing data locally on IoT devices and low-power edge hardware, this approach reduces latency and ensures rapid decision-making even in areas with limited connectivity. Applications include landslide detection, flood forecasting, and human monitoring, where AI-driven edge devices analyze sensor data to provide early warnings and guidance for rescue. This decentralized system enhances resilience by reducing dependence on cloud infrastructure, improving energy efficiency, and enabling autonomous disaster response. Integrating edge intelligence into disaster management can significantly enhance preparedness and mitigate risks in vulnerable regions.

Three-Dimensional (3D) Integrated Circuits

With Moore's Law reaching limitations in transistor scaling, one solution is to expand vertically by stacking multiple silicon layers interconnected by vias. Our focus is on designing 3D architectures, such as memory stacks (SRAM, eDRAM, etc.), and converting conventional 2D architectures to 3D.

Neuromorphic Computing

Inspired by the human brain, we design neuromorphic chips featuring physical artificial neurons made of silicon. These chips perform computations that mimic biological brain functions. Our approach combines digital methods with 3D Integrated Circuits and Network-on-Chip (NoC) structures, developed in both hardware (Verilog HDL) and software (Python) models.

Collaborators

I am open to research collaborations. Please feel free to contact me if you are interested.