Sun gets government grant to replace chip interconnects with lasers

By Michael Hatamoto | Published March 24, 2008, 4:28 PM

Sun Microsystems has been granted a $44 million Defense Advanced Research Projects Agency (DARPA) contract from the US Department of Defense aimed at creating the next generation of supercomputers that could break Moore's Law.

Recently, researchers at Sun have shown a higher level of interest in wafer-scale integration, which is a process for creating faster computers at a lower cost by implementing as many functions as possible onto each die. The downside of this approach, typically, is that there are fewer, though larger, chips on each single wafer of silicon.

According to Sun, if it is successful with its interconnect design, new computers will have the potential to be more than one thousand times faster than its current generation. The end result could be a new virtual supercomputer that is more efficient, faster and cheaper than anything available on the market now.

The five-and-one-half-year research project will be part of the DARPA High Productivity Computing Systems program, and will offer $8.1 million contributions to the Sun Microelectronics and Laboratories divisions.

Sun's research efforts will mainly focus on microchip interconnectivity via on-chip optical networks, which essentially means the wires between computer chips will be replaced with lasers. The Santa Clara-based company became interested in this line of research after Sun engineers discovered several years ago that they could transfer data by transmitting light across the chips in tiny channels called waveguides. The new technology, described as a "macrochip," allows researchers to connect chip-to-chip I/O technology while also creating low-cost chips that can be used in faster and smaller computer systems.

Sun will work alongside two silicon photonics firms, Kotura and Luxtera, and the University of California, San Diego, and Stanford University for the next few years. Sun was chosen for the project over Intel, IBM, Hewlett-Packard, and MIT.

The company has a realistic yet bleak view on the project, projecting a 50% chance of failure.