The Community for Technology Leaders

Existential Architectures: The Metaphysics of Computer Design

Philip G. Emma, IBM T.J. Watson Research Center

Pages: pp. 4-6



The concept for this issue was to highlight work outside the orthodoxy of the usual topics. It was meant to showcase "outside-the-box" topics that tend to not make it into the usual fora for the simple reason that people are not quite sure what the rules are for reviewing unorthodox ideas. Unorthodox ideas are (tautologically) unorthodox, so cannot be compared in a direct way, hence cannot be argued suited for publication by comparison, and so on. The conundrum is clear. (Of course the conundrum is not clear; the fact that it is one is clear.)

As I sometimes find our literature painful and tedious—not to say that it isn't good, but more that it frequently devolves into a belaborment of bit twiddling superimposed over old ideas—I had been hoping to inject some new material into the fray with this issue. I am indebted to the authors and the reviewers for their hard work in making this happen.

I sometimes worry that our literature is becoming stale. I think that culturally, we've become too meticulous and crotchety when it comes to form, and sometimes consider the substance of a work only as an afterthought. It is generally understood that submissions to certain conferences follow a prescribed, precise, and austere format and expository, including the usual bar graphs... I won't bore you with those details.

The point is that while an exposition should be held to certain standards, the exposition should not be the entire point of a work. Exposition is merely the proper plating of what should be Epicurean excellence in the first place; it should not be "Pease Porridge Cold" on a gold charger. Just as jovial conviviality can hardly be construed as wit, flawless expository cannot be a substitute for good and new ideas.

In this special issue, I was interested in new ideas. If an author espoused enough of them, I was willing to forgive him nearly anything—such as having a practical bent and suggesting real applications, or even making an overt display of real intelligence by arguing points crisply and flawlessly. (In fact, I sometimes wish we would have new award categories at conferences, such as "Most Original Work." Or perhaps even "Most Fallible Argument" or "Most Arrogant Delivery.")

The acnestis is that point on the back of an animal (including the human) between the shoulders and lower back that cannot be reached, hence cannot be scratched. This issue is an attempt to scratch the acnestis. Of course, this can never quite be done. As writers and reviewers, we can never quite walk off the beaten path completely. But we can try.

The reviewing process for things off the beaten path is itself a challenge. There was little consensus. But as Oscar Wilde said, "Diversity of opinion about a work of art shows that the work is new, complex, and vital."

While there was an adequate response to the call for papers, there was frequent disagreement among reviewers. This is a good sign. In this issue, we have four diverse works, each in a different area. You might notice that one of the works is by yours, truly. I am well aware that this has the potential of raising eyebrows, so I am compelled to explain.

Since this puts me in a slightly awkward situation, I have been unsure of how to broach the topic. I eventually made the decision to broach it in a way that could at least offer the delight of surprise. So I offer you total candor. (Please retake your seat. As I had said, we are looking for the unorthodox.)

The truth is that after the dust had settled, we were down to three papers that the reviewers felt were both sufficiently interesting and sufficiently solid. This would make for a very slim issue, so it left me with a problem. It so happened that I had written an invited paper with some colleagues in the summer of 2006 for an event that never happened because it became mired in politics, of all things. I had quite forgotten about the paper until this new problem presented itself.

Quite frankly, in my particular job, I am not primarily judged on publications, and have become a little jaded by the vetting process for papers these days, so am generally disinclined to voluntarily submit work. The current situation seemed to be a case in which I could pull out a paper that was already written, and put it through the vetting process without doing any new work. And I was disinterested in whether it was accepted, but felt that it might make interesting fodder for review.

Obviously, I could not be the one to seek reviewers for it, since (among other reasons) I did not wish to know their identities—just in case they offended me by saying things about it that were absolutely true. So I am indebted to Dave Albonesi for selecting the reviewers and managing the review process, and to those selected reviewers for having had the patience to read my paper and the candor to freely offer their many constructive insults. I think I was stabbed in the acnestis.

Our issue opens with a work by Pistol, Dwyer, and Lebeck on nanoscale optical computing. Using properties of optical resonance, the authors posit a rudimentary set of logical operations that can be performed photonically between optical signals at physical densities in the one-nanometer range. Since we are all well aware of the numerous challenges in continuing to push the logic roadmap below 65 and 45 nanometers, and are even more painfully aware of the apparent physical limitations of devices made from real atoms having diameters of several Angstroms, it is clear that at some point, we will need a switching logic not predicated on field-effect transistors. Pistol and his coauthors give us a glimpse of a possible technology.

Next, we turn to system performance evaluation, as Jin and Cheng explain their concept of evolutionary benchmark subsetting. The familiar underlying idea is the characterization of traces by judicious mixes of trace subsets. This reduces simulation time, which allows for a more complete investigation of a design space. Jin and Cheng introduce what they call "survival of the fittest" as a technique to choose and weight trace subsets wisely.

The penultimate article, by Abella et al., is called "Refueling: A Technique to Avoid Wire Degradation." This paper addresses the chip-lifetime issue of shorts and opens caused by metal migration. Abella and coauthors advocate simple devices to keep track of switching on long wires (for example, buses), and to "refresh" wires by running currents in the wires when they are not in use so as to keep the average current (or the integral of the current) in each wire small.

And strictly last, because I felt it too obnoxious to put it ahead of any of the others, is the paper by Emma and his colleagues Reohr and Meterelliyoz, colleagues whom I would like to thank for having had enough good humor to work with me. In this paper, we take a new look at assumptions about retention time in DRAMs, and contrast the requirements of DRAM used in a cache environment with those when using it in memory. We argue that the refresh process can become a limitation as technology evolves to further nodes, and show methods to reduce refresh rates dramatically.

I would like to thank Dave Albonesi for allowing me the honor of having guest edited this issue, and for all the help and guidance he gave me. And I'd like to especially thank Margaret Weatherford for her outstanding knowledge of our language and for bringing that to bear as an editor to greatly improve the quality of our articles.

Finally, I'd like to thank our readers for making IEEE Micro a vibrant source of technical vitality. I sincerely hope that at least one of the ideas in this issue will be new to you, will pique your creative spirit, and will lead you to some more new ideas. I hope that we have done much more than to fill what may have been a much-needed void in the literature.

And I'd like to thank our authors and reviewers once again. I really appreciate your work in making this issue happen. I hope we can always be trop d'audace, without being trop de zele.

About the Authors

Philip G. Emma joined the IBM T.J. Watson Research Center in 1983. He has worked in the areas of microarchitecture, architecture, systems design, circuit design, electronic packaging, materials, interconnection technology, and optics. He is currently the manager of the Systems Technology and Microarchitecture group, which does advanced concept work in architecture and microarchitecture for IBM's p-series and z-series processors, as well as for processors for mobile applications and games. He also works on exploring and leveraging new physical technologies within the context of systems design. He holds over 100 patents and was named an IBM Master Inventor. Emma received his BS, MS, and PhD in electrical engineering from the University of Illinois. He is a member of the IBM Academy of Technology, and is a Fellow of the IEEE.
62 ms
(Ver 3.x)