| The complexity of engineering problems nowadays are getting higher and higher. One of the most common scenario is that engineers crave for useful knowledge from the immense amount of data. While many applications in the real world work on numerical data, the electronic design automation applications often need to deal with data in binary bitstream. Therefore, this dissertation presents a data learning framework called ordered binary decision forest (OBDF) dedicated to Boolean domain. OBDF learning mainly incorporates the ensemble learning concept with many individual techniques including bootstrap sampling, Gini impurity estimation, Boolean data mining and out-of-bag weighting. The proposed OBDF learning framework is also applied to three different applications, and for each application, the collaborative methodology is also presented: (1) Signal controllability estimation is achieved by an incremental learning methodology and exercised on a commercial microprocessor design. (2) A simulation-based functional test generation methodology called TTPG is proposed for embedded processors and demonstrate its effectiveness on both datapath and control logic. (3) A software-based self-test (SBST) methodology is modified for speed binning and functional paths with long delays can be exercised through the proposed SBST methodology. |
