Events Calendar

CENG Seminar

Speaker: Peng Li, Texas A&M University

Talk Title: From Integrated Circuit Design to Brain Modeling: Coping with System Complexity by Leveraging Application-Specific Parallel Computing

Abstract: We are confronted with system complexity while dealing with large natural and engineered systems such as a multi-billion transistor microprocessor, or the 100 billion neuron human brain. In the meantime, the recent change in the computing landscape has rendered the use of parallel compute power critical for dealing with the complexity in a broader spectrum of engineering and healthcare applications.

The drive for higher performance has resulted in an explosion of IC design complexity. To push the envelope of design algorithms and tools, “going parallel” is both a natural choice and a necessity. Nevertheless, one must rethink how compute-intensive algorithms shall be designed on disparate hardware platforms to make the best use of parallel compute power. I will present our work on highly parallel circuit simulation, where a rich set of application-specific intra- and inter-algorithm parallelisms are explored to help remove the simulation bottleneck from the IC design flow. I will describe how “smart” numerical algorithms can be designed to expose the “hidden” data parallelism to allow for efficient hybrid GPU-CPU based analysis of large power delivery networks, thereby facilitating the design of high-performance & low-power chips.

Computing also plays an increasingly vital in understanding mammalian brains. With a vast amount of data made available by neuron recording and imaging, one holy grail is to map out the circuitry of the human brain and unravel the mysteries of brain functions by computer simulation. If successful, this would have profound implications: it will enable the test of hypotheses of neurological disorders and the development of treatments; it will stimulate new bio-inspired computing and biomimetics. To examine some of the grand challenges and opportunities, I will describe a large-scale thalamocortical model that includes multicompartmental Hodgkin-Huxley neuron models capturing dynamics of ion channels and dendrites, detailed cortical microcircuitry, local/global connectivity, all modeled on a biophysical basis. While exploring the dynamic properties of the network, advanced numerical and parallel computing techniques have been developed to alleviate the significant simulation challenge. The biological realism of the model allows us to attribute network-level spike-and-wave oscillations, a characteristic of generalized absence epilepsy, to cell-level biophysical interactions and shed light on the therapeutic treatments of this brain disorder.


Biography: Peng Li received the Ph.D. degree in ECE from CMU in 2003. He is an associate professor of ECE at Texas A&M University, where he is also a member of the Faculty of Neuroscience. His research interests include integrated circuits and systems, CAD, parallel computing, biophysical modeling of nervous systems, computer-aided diagnosis and therapy of brain disorders. He has edited two books, published over 100 papers, and six book chapters. He is a recipient of two IEEE/ACM DAC Best Paper Awards and one DAC Best Paper Award nomination, four nominations for the IEEE/ACM William J. McCalla ICCAD Best Paper Award, an NSF CAREER Award, four Inventor Recognition Awards from SRC and MARCO, and an ECE Outstanding Professor Award from Texas A&M. He is an Associate Editor of IEEE Trans. on CAD and IEEE Trans. on Circuits and Systems II. He has served on the committees of DAC, ICCAD, ISQED, ISCAS, TAU and VLSI-DAT, the selection committees of ICCAD Best Paper Award and ACM Outstanding Ph.D. Dissertation Award in EDA. He served as the program chair and general chair of the ACM TAU Workshop.

Host: Alice Parker

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Estela Lopez