ICCD 2017


    Important Dates:

    Abstract Submission
    Closed on 19-Jun-2017

    Paper Submission
    Closed on 28-Jun-2017

    Notification
    Closed on 31-Aug-2017

    Camera-Ready
    30-Sep-2017

    Contact Us

    Web Chair

2017 IEEE International Conference on Computer Design (ICCD)

Philip Emma

Chief Scientist (retired), IBM T.J. Watson Research Center, IEEE Fellow, CEO EmmaSys

Title: How we need to Evolve Computer Architecture: A Long Overdue Evolution

Abstract: The first electronic stored-program computer was made by von Neumann in WW2. If we look at the architecture of that machine, it’s not much different from architecture today. I’ll discuss today’s architecture, how it’s failed to evolve, and where it needs to evolve. The irony of Moore’s Law is that it provided (nearly) free performance gains for several decades, which obviated the need for real evolutions in computing and computing systems. I’ll discuss some completely new – and long overdue – extensions that need to be made in modern computers, and will also discuss some new applications that can use a new dimension in systems – the third dimension. 3D will naturally facilitate some new kinds of architecture and new applications. Our notions of “what a computer is” are antiquated, and need to extend to new applications. When new technology – and new approaches - facilitate orders odd magnitude performance improvements, unforeseen applications emerge. I’ll portend several, and will discuss how new technology can drive new applications.

 

 

Bio

Philip Emma has a PhD in Electrical Engineering, and is a Fellow of the Institute of Electrical and Electronics Engineers. He recently retired as Chief Scientist from IBM TJ Watson Research Center in Yorktown Heights, NY. He went to IBM Research in 1983 to work for von Neumann’s chief engineer, Jim Pomerene, who had come to IBM to harvest after completion of the IAS machine. He formalized the first rigorous CPI decomposition to enable the first analytical performance models to be used to compare microarchitectures, predating the Hennesey & Patterson formulation by about a decade – although was not allowed to publish the work. He did the first foundational work in the field on branch prediction and evolved this whole area into the complicated set of systems that it has become. He did similar work on cache management and prefetching, and formalized the (rather abstruse) coherency rules used in systems today. He led the design on a completely new approach to all of the checking and recovery on the first IBM CMOS mainframes, facilitating 100% checking and recovery via unit replication – done to minimize the cycle times on our first CMOS machine. Then he headed a completely exploratory group to expand the application of technology to architecture, including many innovations in packaging, cooling, and optics. As Chief Scientist, he did lots of work exploring the opportunities afforded by 3D systems, and has recently formulated some theory for dis-integrating large frames using bandwidth. He will discuss both of these things today. He holds over 200 patents, and has written parts of four books on technology, as well as over 300 articles.