Rupert Goodwins' Diary - Holy Land Special

Thursday 15/12/2005

It's Fab Time. The highlight of the trip awaits. Another hour in the bus and we're at Fab 18 in Kiryat Gat, which is where most of the Pentiums 3 and 4 came from. It's also next door to a large building site, which is going to be Fab 28: Intel announced this earlier in December and is spending billions on getting it ready for production in 2008: it'll be 45nm 12 inch wafers, double the size of Fab 18 and will, by itself, contribute another 2 percent to Israeli GNP.

"In that case," I ask, "isn't it a concern that it's all built on land that probably belongs to the Palestinians?"

This is perhaps not a welcome question, but I think it's worth asking. Kiryat Gat is not part of the occupied territories, being safely in the central southern bit of Israel, but it is rather unusual. The land the plant's on is explicitly mentioned in an annex to an armistice signed between Israel and Egypt in 1949, because the locals and some Egyptian soldiers managed to hold out during the fighting beforehand. That annex guaranteed the property and safety of the inhabitants of the villages there: as you might guess, this didn't pan out. Google 'Faluja pocket' for the details: be prepared for some fierce words and contradictory claims.

All this is unusually well documented. I normally regard Middle Eastern politics with a mixture of dread and resignation — nobody comes out of it well, and if Iran nukes up there's a good chance nobody will come out of it at all. I'm certainly not in the habit of poking my nose into the most sensitive affairs of a country which arms its hippies. But if you're going to have a significant percentage of your national product coming from a small plot of land, why pick one with significant issues? Why not put a farm there, and build your fab plant where the farm was?

The answer is that as far as Intel is concerned, all the deals it does over Kiryat Gat are with the State of Israel. End of story. It can't be a coincidence that the fab is there, but the normal rules of logic do not apply at that end of the Mediterranean. Your guess is as good as mine.

So we get back to fab plants. There's some light relief during another traditional part of any Intel fab briefing: a graph is shown that depicts the speed at which the yield rate is improving and how much better it is now than the last time they changed process. The graph has no labels: we ask what the actual numbers are and we are told that it's top secret.

Fair enough. Yield rate is the most important thing to a plant — it's the number of functional chips that come out at the end of the line. There are a million things that can go wrong and reduce the yield: with each wafer producing hundreds of die that cost around 40 cents each and sell for hundreds of dollars apiece, even a few percent change in yield makes an enormous difference to the bottom line. You certainly don't want your rivals to know, and the last people you'd tell are journalists.

This time, though, Intel had made one tiny mistake...