In the early days of computing logic was expensive and wires were nearly free. Now, however, the reverse is true: logic is nearly free and wires are expensive in area, delay, and energy. Nevertheless today's computers, though faster and smaller, are otherwise very like those of 50 years ago. Can we discover how to use the new properties of integrated circuits to make better computing structures? Or like bridge builders, will we long persist in copying stone arches in the same steel that makes suspension bridges possible? We seek a new paradigm for computing appropriate to modern integrated circuits.
FLEET explores an alternate computing structure with a focus on communication and concurrency. Where yesterday's computers put programmers and compilers in charge of logical operations which were then expensive, FLEET puts programmers and compilers in charge of communication because that's the expensive part of today's technology.
FLEET offers low level concurrency on a grand scale. Instead of trying to program concurrency on multiple cores that are otherwise sequential, FLEET treats concurrency as fundamental. The program must impose sequence, if sequence is necessary.
We have finished testing a simple "warm up" chip in 90 Nanometer CMOS. It demonstrates that the communication structure planned for FLEET will run at nearly 4 GHz. Simulation of simple programs demonstrates that communication-centric programs are possible, though quite unlike standard codes. This talk will describe our progress towards the design and programming of a first simple FLEET computer.
Ivan Sutherland is a Vice President and Fellow at Sun Microsystems. He has had research adventures in graphics, robotics, and integrated circuit design. Ivan is a member of the National Academy of Engineering and the National Academy of Sciences, and the winner of the 1988 Turning Award. He spends most of his time working on FLEET at the University of California in Berkeley.
Adam Megacz is a graduate student at UC Berkeley who came into the FLEET project through Ivan's research course about FLEET. Adam has become the main software architect in the project. The design work now proceeds as a collaboration between Adam advocating what is programmable and Ivan advocating structures appropriate to modern silicon technology.
James Hook