Tom Anderson
January 10th, 2004, 08:40 PM
Typical in many electronics companies is to test an integrated circuit before and after stress - burn in. The test before burn in reduces many of the potential failures that burn in alone might induce. Consequently, a single sampling plan operating characteristics is highly likely to truncate the "incoming" defective that is represented on an Operating Characteristic curve. This should imply that from a simple minded view the "beta" - type II error of the OC-curve is potentially quite pessimistic and the defective proportion escaping with some probability is actually much lower than that estimated from an OC-curve.
How might one adjust an OC-curve to account for this, when, for example, a sample of 350 units, would normally predict for an average 2000 unit lot, an AOQL=682 ppm, and with beta = 0.1 a LTPD ~ 6500 ppm.
In a case similar to the plan mentioned, the pre-testing before burn in has been demonstrated to typically reduce most lots to < 2500 ppm, So I feel that the type II error is much less, but I have not found a way to re-estimate the operating characteristics.
How might one adjust an OC-curve to account for this, when, for example, a sample of 350 units, would normally predict for an average 2000 unit lot, an AOQL=682 ppm, and with beta = 0.1 a LTPD ~ 6500 ppm.
In a case similar to the plan mentioned, the pre-testing before burn in has been demonstrated to typically reduce most lots to < 2500 ppm, So I feel that the type II error is much less, but I have not found a way to re-estimate the operating characteristics.