If one circuit on a chip breaks, will the chip fail?

[Schmave]

I was thinking about this and was really wondering what would happen. Let's say you have a microprocessor with 50 million transistors on it. Then let's say an interconnect between two of the transisters gets blown apart or something, or was never connected in the first place. Will this cause the chip to fail? Or will the chip only fail or generate an error of some kind when it needs to use that part of the circuit? It seems like since there are so many transistors and interconnects on a chip that there are bound to be some broken circuits on the chip. It also seems like one interconnect on a chip would be nearly insignificant, but something tells me that a broken circuit on the chip would cause problems.

 

[CTho9305]

IIRC, there is NO redundancy in a modern processor. any one defect, and you will notice it. of course, you could theoretically work around some. for example, if the defective part of the chip was in the floating point unit, as long as you didn't ever use the FPU, the chip might work. but if you do try a floating point operation, you might get a crash or an incorrect result.

(exception: I think cache has some redundacy of some sort, but I'm not sure)

 

[zephyrprime]

<< It seems like since there are so many transistors and interconnects on a chip that there are bound to be some broken circuits on the chip >>


Nope. If a chip is to work, all of the circuits have to work. I suppose that if a really arcane useless instruction were toasted, it may never be noticed. But even if a circuit were used only once in a while, computer would generate erroneous results and crash soon after. If one considers a single transistor to be a "part", then modern ICs are by far the most reliable machines ever made by man on a # failures per part basis.

 

[rimshaker]

<< It seems like since there are so many transistors and interconnects on a chip that there are bound to be some broken circuits on the chip >>



No, there actually isn't surprisingly. Why do you think chip companies like intel and amd spend billions upon billions of dollars for products ultimately smaller than your thumbnail?

 

[JeremiahTheGreat]

is it even possible to put any electronic redudancy into todays CPU's?

 

[FishTankX]

Jeremiah the great:The Itanium has some redundant execution units. I forget if it was ALU or FPU. Would peter care to comment?

Redundancy is only found in enterprise class CPU's because they only make a few hundred thousands of these CPU's each year and these CPU's are monstrously huge (As in double the size of the P4) so redundancy in execution units can help *drastically* with yields at such monstrous sizes. I assume the yield rate would be something like 50%, meaning that without redundant execution units, half of all of the CPU's would be DOA.

A silicon wafer has a number of defects on it, if one defect is inside a processor, that processor's toast. Thus smaller processors have better yield rates because they having a certian % of their brethren having a defect.

 

[SuperTool]

There is redundancy in some areas of the chip like the cache. Because they occupy such a large area, the likelyhood of bugs is high, and there are usually spare columns that can be used (with a fuse) to replace columns with a bad transistor)
There is no redundancy usually on functional units.