Dependable Multiprocessor

>> The Problem

Typical design solutions for protecting spacecraft computers from damaging radiation have involved surrounding the computer with radiation shielding and using custom components specially designed and built to be radiation resistant. The difficulty with this approach is that radiation shielding is heavy and bulky, while custom "radiation hardened" electronics are power hungry, slow and large compared to their standard "Commercial Off-The-Shelf" (COTS) counterparts. They are also extremely expensive and generally are several generations behind the state of the art of their commercial equivalents. Since a spacecraft's most precious commodities are mass (weight), volume, and power, it is difficult to build a modern, powerful, space-going computer system within acceptable mass, volume and power constraints.

The problem gets even worse as space science and exploration missions move toward more self-contained and autonomous spacecraft that are also smaller, lighter, and more power constrained. At the same time, space missions are becoming more and more data intensive and ranging farther and farther from home, with ever greater data and command transmission times. Imaging instruments are becoming so sophisticated that they can generate data at rates of over a terabyte per day! (A terabyte is 1000 gigabytes, equivalent to about 300 feature-length movies.) With the current, relatively low-bandwidth, satellite-to-ground-station data links, we cannot get this much data down to Earth for processing by ground based supercomputers. In cases where the results of the data processing are to be used to make mission decisions, such as 'what to do next,' we don't have the time to download, process, decide, and then upload the resulting commands. We must process the data onboard the spacecraft, make required decisions onboard, and send down a much reduced, processed data set. What we need now are small, lightweight, low-power supercomputersthat can operate successfully in space.

If we could get such a powerful computing architecture into space, we could design the spacecraft to process its own science data onboard and make complex decisions immediately about the scene it is observing, without sending the data down to Earth for analysis. Unfortunately, adapting even simple general purpose computing technologies for use in the harsh, error-inducing environment of space has encountered significant difficulties and delays. Transcending these computational limitations is of extremely high priority in the development of the next generation of space missions.