Energy-Efficient Design

We view energy-efficiency as an integrated problem that connects power delivery, low-power circuit/system design, and thermal management; and modulated by advanced 2.5D/3D packaging technologies. Our holistic approach to energy-efficient design consists of following thrusts:

  1. Computationally Efficient Artificial Intelligence: We are developing computationally efficient AI/ML methods coupling architecture, circuit, and device innovations .
    • Processing-in-memory:  We are developing a unique in-memory and programmable computing platform to accelerate training and inference of deep learning networks.
    • Energy-efficient Training and Inference: Algorithm-architecture co-design techniques to reduce memory and computation demands of learning algorithms during training and inference.
    • Exploring post-CMOS devices for Learning: Application of emerging devices such as Resistive devices (RRAM), Ferroelectric FETs (FeFETs), and Tunneling transistors for machine learning platforms.
  2. Domain-Specific Acceleration beyond Machine Learning .  Our goal is to explore architecture and circuit design techniques for domain specific acceleration. Our interests include:
    • Acceleration of machine learning and artificial intelligence methods.
    • Acceleration of high-bandwidth signal processing algorithms.
    • Acceleration of  scientific computation including solution of partial differential equations.
  3. Integrated Voltage Regulation. Our objective is to design efficient voltage regulators (VR) integrated in the same chip/package with a System-on-Chip (SoC). The research focusses on (i) high-frequency inductive regulators and digital low-dropout (DLDO) regulators, and (ii) Inductive boost/buck converters for energy harvesting.
  4. Advanced 2.5D/3D Packaging: Our research is focussed on designing energy-efficient system-in-package platforms with Monolithic 3D, TSV-based 3D, and 2.5D (interposer) integration. Our interests include (i) innovative system architecture, (ii) circuit techniques and tools, specifically, for memory design, and (iii) power delivery, testability, and thermal management solutions.
  5. Low-power circuits and systems: The driving principle behind our low-power research is to minimize built-in safety margin in the design of integrated circuits and optimally (dynamically) balance the energy dissipation and  quality-of-service of a system under all environments.
  6. Electro-thermal co-design for mobile platforms: GREEN lab has research explores electro-thermal co-design methods for systems with limited cooling capacity. Our contributions include innovative approaches and hardware demonstrations for (i) Post-silicon thermal characterization and management, and (ii) Integration of advanced active cooling, such as, thermoelectric  devices and in-package microfluidics for mobile processors.