Data mining for manufacturing control: an application in optimizing IC tests

摘要: We describe an application of machine learning and decision analysis to the problem die-level functional tests in integrated circuit manufacturing. Integrated circuits (ICs) are fabricated on large wafers that can hold hundreds individual chips (die). In current practice, expensive machines test each these die check they functioning properly (die-level or DLFT), then cut up, good assembled into packages connected package pins. Finally, resulting tested ensure final product is correctly. The purpose avoid expense packaging bad provide rapid feedback fabrication process by detecting failures. challenge for a decision-theoretic approach reduce amount DLFT (and associated costs) while still providing feedback. which historical data mined create probabilistic model patterns failure. This combined with greedy value-of-information computations decide real time next when stop testing. report results several experiments demonstrate ability this procedure make testing decisions, stopping detect anomalous die. Based from Hewlett-Packard, system has potential improve profits mature IC products.