>> For crying out loud, even AMD is selling about everything they can make,
even with their spotty yield history (bad yield = bad reliability).<<
Yield and reliability are not necessarily correlated.
Yield refers to the chip's ability to pass final test. Particles, the number one yield killer, typically cause shorts or opens in the circuit. It doesn't work now, and it will never work in the future.
Reliability refers to the chip's ability to perform at specification for an extended period (typically 10 years). Reliability is degraded by things like dielectric breakdown, which in turn is due to things like poor film uniformity, poor film composition control, and so on. It strongly depends on the stresses on the chip. For instance, a 300 MHz Pentium chip may have failed during testing at 450 MHz, yet is likely to be completely reliable at 300 MHz.
It's nearly impossible to say exactly what AMD's well publicized yield problems involved, since no chipmaker gives out that kind of information until well after the fact. They could have been faced with particle storms that flat out killed too many of their chips, or they could have been having dimension control problems that kept them from making the high speed chips. Neither of these problems would affect reliability in the field, or give AMD customers any reason to worry.
During production ramp on a new process, it's common to do extensive stress testing for reliability, and the process would be unlikely to handle at-risk production before those tests were passed. That could also account for AMD's delayed production ramp. But no (solvent) chipmaker is going to risk field failures.
Katherine |
