ECE IBM Distinguished Lectures 2010-11

Silicon Integrated Nanophotonics: a journey from fundamental science to disruptive technology

Dr. Yurii Vlasov, Manager of Silicon Nanophotonics Department at the IBM TJ Watson Research Center

Silicon Integrated Nanophotonics allows ultra-dense monolithic single-chip integration of optical and electrical functions. This technology can enable future supercomputers capable of delivering Exaflops (10^18 floating point operations per second) by connecting racks, modules, and chips together with ultra-low power massively parallel optical interconnects. Such a disruptive technology is a result of a decade of multidisciplinary exploration of various scientific fields in material science, fundamental physics and optics, advanced device engineering and technology qualification.

Solar Energy Conversion Research at IBM

Supratik Guha, Director of Physical Sciences at IBM Research

Following an overview of the research activity in Physical Sciences at IBM, related to the physics and technology of information processing and topics related to IBM’s Smarter Planet activities, Supratik Guha will describe in some detail research in three different areas in photovoltaics that are ongoing. The three areas are: (i) thin film solar cells from earth abundant materials, (ii) nanowire based solar cells, and (iii) solar power forecasting. Earth abundant thin films, that can support manufacturing rates of >10-20 GW/yr with readily available materials is an important topic of research. He will describe results that they have in this area within the copper-zinc-tin-sulfide (CZTS) system, focusing on vacuum deposited material. He will describe some results of their work on silicon nanowire photovoltaics, where the benefits are one of enhanced light trapping and, possibly, the ability to have devices that are small enough so that generated carriers can be collected without relying upon minority carrier diffusion. The drawback here has been the detrimental role of surfaces and interfaces that has prevented nanowire solar cells (and other nanostructured solar cells for that matter) from demonstrating high efficiencies. Finally, Supratik will briefly describe some results from a relatively new experimental project where they are trying to predict the solar power production from a solar array on the scale of minutes to an hour by predicting detailed cloud coverage.

Drilling Through the Layers of Liars

Kevin Stoodley, Chief Technical Officer (CTO) for Compilers, IBM

In the beginning we had “The Policy of Truth”, but the temptations of that innocent world led us to tie software and hardware systems together too tightly and the resulting brittleness and widespread complexity proved ultimately to be unsustainable and caused us to eschew truth as a virtue when it came to systems design and implementation. And so we have taught ourselves to separate concerns, create (hopefully) simplifying abstractions and to layer them all together to compose a backwards compatible, extensible and maintainable software and hardware stack. However, as in real life, duplicity, while often expediently achieving some short term goal, is rarely without downstream consequences and similarly in the systems design space, we now find ourselves mesmerized by the glamour of our own abstractions. We’ve convinced ourselves that the benefits we’ve gotten are worth the post-dated cheques we have written against the perpetuity of single-thread performance improvement. Sadly our systems are beginning to suffocate under the burden of impenetrable “layers of liars” and we find our isolationist, information-hiding and even actively duplicitous abstractions have come back haunt us.

The gigahertz perpetuity has failed and the new kid on the block, “core count” is marking our post-dated cheques “NSF.”

This talk takes a brief tour of some common abstractions and how they are starting to cause problems, mostly performance problems, and presents a few ideas on how to maintain the best part of both worlds including some early motivational work on virtualization-aware JVMs which shows the kind of value we can get by relaxing our abstractions a bit. In a very real sense, our systems will have to become smarter in order to continue delivering on the value propositions that users have come to expect.