PORTLAND, Ore. — First silicon, the initial prototype of a new chip design, often involves days of tedious, labor-intensive debugging and mask repairs before production-ready chips can be produced. Most problems with initial prototypes can be diagnosed, but repairing problems deep inside a chip becomes increasingly difficult as design rules shrink and mask levels are added.

Some have proposed adding extra tunable structures to affect repairs post-silicon.

FogClear, a software-only approach to post-silicon debugging, was unveiled this week by University of Michigan researchers at the International Conference on Computer-Aided Design. FogClear aims to make obsolete the "first silicon" moniker by repairing errors in prototypes in hours rather than days. Proponents said this can be accomplished by merely changing one of the top interconnection layers.

"The problem with post-silicon debugging today is that even if I know where the problem is and what I have to do to fix it, that information is completely uncorrelated with how the transistors are laid out before the fix--and changing a transistor layer can be very expensive," said University of Michigan engineering professor Valeria Bertacco. "FogClear allows you to repair problems in a less disturbing way. Using puzzle-solving search algorithms it quickly diagnoses problems and automatically fixes them in hours instead of days--usually by changing one of the top interconnection layers.

"If you have added an extra layer to use for repairs, we can use that, but it is not necessary," added Bertacco, who collaborated on the FogClear software with fellow professor Igor Markov and researcher Kai-Hui Chang.

Fundamentally, FogClear works by fixing each bug in a way that minimizes the affects on the rest of the chip layout.

Today, designs are first validated in simulations, then first-silicon prototypes are tested running both numeric and string-manipulation algorithms. If errors are found in the results, they are manually traced back to the transistor layouts. Then, interior masks must be repaired and a new prototype fabricated.

Sometimes the repairs introduce new problems and the process must be repeated--taking weeks or even months to find and repair all bugs.

FogClear programmers claim their algorithms can help by finding the bugs automatically and fixing them noninvasively. As a result, fewer revised prototypes have to be fabricated, the researchers claim.

Besides finding and repairing problems in less invasive ways, the researchers claim FogClear utilizes a wider array of operating conditions than normal. Testing every possible operating condition is an intractable problem, since there are too many combinations. FogClear's automated puzzle-solving algorithms try out a wider variety of those combinations than is typical today, proponents said.