An evolutionary approach for Nominal design and yield enhancement of analog amplifiers

The microelectronic industry is driven by the continuous demand for processing speed and capacity. To answer such demands, novel design paradigms target design automation. While digital design is mostly automatic, design automation in the analog domain is limited and mainly used for high-level synthesis. A promising solution to overcome limitations and constrains of automatic analog design is to involve computational intelligence techniques. This work proposes an evolutionary design methodology to bring contributions to transistor-level design automation. The proposed framework is built around a typical design automation workflow: Feasibility design, Nominal design and Design centering. The Feasibility design layer employs an artificial neural network to generate an initial solution. The following two layers employ genetic algorithms to optimize the amplifier for performance and yield respectively. The proposed design methodology is illustrated on the design of a folded-cascode operational transconductance amplifier and is validated by the results of extensive simulation.