- Jul 11, 2001
- 1,127
- 0
- 19,280
"THG: Dual-core (two CPUs on a single chip) nearly doubles a processor's surface. What happens to the yield rate in this case?
Wallers: Dual-core in 90 nm takes up roughly the same space as a single-core manufactured with a 130 nm. But that doesn't impact the yield rate or error rate on the wafer. "
While this is true, the fact that you're relying on two die instead of one for a single chip definitely should impact yields.
Example:
Say AMD's yields are 50%. This means for any given die, it has a 50% chance of going to market as a great finished product. For dual core, you have two die next to each other that each have a 50% chance of being good. If both are good, the chip is good. If die 1 is bad and 2 is good, the chip is bad. If die 1 is good and 2 is bad, the chip is bad. If die 1 is bad and 2 is bad, the chip is bad. So really, what we have is a chip that has about a 25% chance of being good (since with a 50% yield, you're just as likely to have each situation).
What he said is sort of true - having more transistors doesn't necessarily mean lower yields. But that doesn't take into account the fact that you're now depending on two die to be good for a single product as opposed to 1 die. Combine that with the fact that even if yields were the exact same with Dual-core, you fit at best half the number of product on the same sized wafer (when comparing the same sized process...i.e. 90nm DC compared to 90nm Single Core).
The quote's not saying out-right that yields will stay the same when going to Dual Core, but if you read quickly, you might misinterpret it as such.
I'm just your average habitual smiler =D
Wallers: Dual-core in 90 nm takes up roughly the same space as a single-core manufactured with a 130 nm. But that doesn't impact the yield rate or error rate on the wafer. "
While this is true, the fact that you're relying on two die instead of one for a single chip definitely should impact yields.
Example:
Say AMD's yields are 50%. This means for any given die, it has a 50% chance of going to market as a great finished product. For dual core, you have two die next to each other that each have a 50% chance of being good. If both are good, the chip is good. If die 1 is bad and 2 is good, the chip is bad. If die 1 is good and 2 is bad, the chip is bad. If die 1 is bad and 2 is bad, the chip is bad. So really, what we have is a chip that has about a 25% chance of being good (since with a 50% yield, you're just as likely to have each situation).
What he said is sort of true - having more transistors doesn't necessarily mean lower yields. But that doesn't take into account the fact that you're now depending on two die to be good for a single product as opposed to 1 die. Combine that with the fact that even if yields were the exact same with Dual-core, you fit at best half the number of product on the same sized wafer (when comparing the same sized process...i.e. 90nm DC compared to 90nm Single Core).
The quote's not saying out-right that yields will stay the same when going to Dual Core, but if you read quickly, you might misinterpret it as such.
I'm just your average habitual smiler =D
Facebook
Twitter
Instagram