Advanced Micro Devices Inc (AMD)

[sgolds]

Ixse, hopefully some folks here who can better answer your question will contribute. However, I'll call attention to your last line -

BTW Intel are unlikely to use copy exact everywhere either; they are likely to use it only for identical products (and they have many different products).

Different products often use different manufacturing processes, so 'copy exact' is used between lines which use the same process. Generally, it is a related family of products (such as processors on a given geometry). For instance, running 90nm and 130nm concurrently cannot be done on a single production line.

Even within 'copy exact', there are subtle differences. Although two processor models may use the same basic process and identical equipment, there is fine tuning for characteristics of each product to maximize yield and bin splits. You can't simply take your last wafer out on one product, change masks, and start the other product without some tuning. Think of AMD's K7 line, which uses bulk silicon, and K8 line, which uses SOI - radically different stuff, requiring much different process!

The other thing you should consider is volume, which requires many things to happen concurrently. In other words, if your factory simply takes a wafer boat (cassette of 25) and processes it - then why have a megafab? No, you need to start many boats concurrently, and each concurrent boat is running on a different set of equipment, defining a production line.

To get a realistic idea of how things work at a fab like Fab 30, imagine different parts of the factory working on different products, and some developing future processes, each section with its set of production lines. This imagery will serve you better than the single line image.