In this thesis, we examine the application of reverse engineering and control logic extraction to assist in the analysis and verification of clock-gated circuits. The proposed methodology also enables sequential clock-gating synthesis to further reduce dynamic power. A secondary focus is on recognizing circuit functionalities with deep learning techniques.

The first part of the work deals with the use of transparent logic to recognize control and data paths of gated-level circuits. We invent abstraction models (dependencies graphs, DGs) of sequential circuits and then explain how they can be used to formulate sufficient conditions for legal clock-gating. It is then demonstrated how to perform efficient sequential equivalence checking (SEC) between a circuit before and after clock-gating synthesis based on DGs. The proposed formulation is extended to allow sequential clock-gating synthesis to be done systematically and automatically.

The second part of the thesis introduces the use of neural networks to recognize circuit properties, which can be used to benefit and improve reverse engineering methods. We invent a representation of gate-level circuits to work with neural networks and build a framework for circuit recognition, including function classification and detection. The proposed framework can also be used to locate high-level constructs in the sea of logic gates.