1 Summary Fault Tolerant Computing 6.911 Architectures Anonymous

As the limits of computer technology are pushed into the domains of the very small, the very cheap, and the very robust, the near-perfect reliability which had been granted by the digital abstraction has eroded. Small components may be cheap and fast, but it is an ever greater challenge to produce the correct answer every time, an immovable constraint of the design. (gcc--wrong-answer-very-quickly isn’t in terribly high demand.) To compound the problem, the environment is anything but friendly to computer systems: a typical system is under mechanical, thermal, radiative, and electromagnetic stresses which may interfere with the operation of the machine. The set of errors that can occur are divided into two broad categories, defects and faults. Defects, which are errors made in the hardware during manufacture, become far more common as the complexity and density of the circuits increases, resulting in ever lower perfect-chip yields as time goes on. The most common results of chip defects are faulty logic and flak