About
My research interest is high performance and energy efficient computer architecture design for accelerating deep neural network and autonomous driving system.
Research Experience
Efficient computer architecture design for deep neural networks.
Compared with inference, the training of neural network is more compute and memory intensive, especially for the memory capacity and data movement overhead. Our goal is to improve the performance and energy efficiency of neural network training. Our approach includes processing in memory with efficient programming model, emerging non-volatile memory technology, scheduling and data remapping schemes. (MICRO'18, MEMSYS'18, AIM'17)
Energy efficient hardware design for image and video processing.
Image and video processing applications inherently have error-tolerant ability, thus our approach can fully use this benefit. The goal of our research is to reduce the energy consumption and latency as much as possible under the premise that the quality of image and video will be acceptable. Our approach includes fully exploring the characteristics of emerging non-volatile memory technology, state of art power reduction methods and comprehensive quality evaluation standard. (ICCAD'17, ISQED'16, ISCAS'15)
Research Experience
Efficient computer architecture design for deep neural networks.
Compared with inference, the training of neural network is more compute and memory intensive, especially for the memory capacity and data movement overhead. Our goal is to improve the performance and energy efficiency of neural network training. Our approach includes processing in memory with efficient programming model, emerging non-volatile memory technology, scheduling and data remapping schemes. (MICRO'18, MEMSYS'18, AIM'17)
Energy efficient hardware design for image and video processing.
Image and video processing applications inherently have error-tolerant ability, thus our approach can fully use this benefit. The goal of our research is to reduce the energy consumption and latency as much as possible under the premise that the quality of image and video will be acceptable. Our approach includes fully exploring the characteristics of emerging non-volatile memory technology, state of art power reduction methods and comprehensive quality evaluation standard. (ICCAD'17, ISQED'16, ISCAS'15)