Sigma-Delta Modulation Fault Diagnosis Using an Unsupervised Expert Network
Campin, Michael James
MetadataShow full item record
This thesis introduces an alternative method of testing mixed signal integrated circuits (ICs). Unsupervised competitive learning and supervised growth is combined to create a expert neural network algorithm that is capable of operating in a real- time IC test environment. This expert network algorithm has an advantage over neural network algorithms using supervised training in its ability to quickly adapt to novel input vectors and to follow the fabrication process drift. Because unsupervised learning is used, a large and accurate training database is not required. Such a database is in general difficult to generate but required for supervised training. To test this approach, a Monte-Carlo simulation database of a single loop sigma-delta (epsilon-delta) modulator is used.