Optimization processing unit (OPU) applied to integrated circuit design and manufacturing

摘要

The authors proposed a general purpose parameter optimizer, Optimization Processing Unit (OPU), which can handle multi-objective parameter optimization in a multi-input multi-output (MIMO) system (applying parameter optimization to a system implies that it is a functional system). Practical computer-aided methods are used to avoid the difficulty of deriving equations that are usually required for system modeling. For multi-objective optimization, OPU only needs the known data of the system input, the corresponding output data, the objectives, the tolerances of objectives, and required fitting accuracy of the system approximation. In current practice, singular points are avoided or limited. Hence, we demonstrate OPU can handle all the field parameter optimization by its very nature. OPU includes four sub-units: Data-adjusted Automatic Taylor Expansion Unit, Tolerance and Constraint Adjustment Unit, Accuracy and Result Self-confirm Unit, and Automatic Optimizer. The fully automatic flow of optimization by OPU has been proposed in this paper. The optimizer applied simulated annealing algorithm [1]–[2] together with dynamic energy threshold and weighted sum method [3]. An Algorithm of Data-adjusted Automatic Taylor Expansion Unit is proposed to translate the system input data and corresponding output to system equations in a Taylor Series format. In addition, the tolerance and constraint adjustment method in OPU was presented. This OPU has been verified and reduced to practice through performance estimation of circuits for various transistor device indexes by using the BSIM4 [4] and BSIM-CMG [5] modeling optimization system. A Vdd-Vth (Vdd: operation voltage, Vth: threshold voltage) driven relation was used to co-optimizes both design and process for logic circuit power minimization.