Book Description
The automated synthesis of mask geometry for VLSI leaf cells, referred to as the cell synthesis problem, is an important component of any structured custom integrated circuit design environment. Traditional approaches based on the classic functional cell style of Uehara & VanCleemput pose this problem as a straightforward one-dimensional graph optimization problem for which optimal solution methods are known. However, these approaches are only directly applicable to static CMOS circuits and they break down when faced with more exotic logic styles. Our methodology is centered around techniques for the efficient modeling and optimization of geometry sharing. Chains of diffusion-merged transistors are formed explicitly and their ordering optimized for area and global routing. In addition, more arbitrary merged structures are supported by allowing electrically compatible adjacent transistors to overlap during placement. The synthesis flow in TEMPO begins with a static transistor chain formation step. These chains are broken at the diffusion breaks and the resulting sub-chains passed to the placement step. During placement, an ordering is found for each chain and a location and orientation is assigned to each sub-chain. Different chain orderings affect the placement by changing the relative sizes of the sub-chains and their routing contribution. We conclude with a detailed routing step and an optional compaction step.
