Circuit Optimisation using Device Layout Motifs

Circuit designers face great challenges as CMOS devices continue to scale to nano dimensions, in particular, stochastic variability caused by the physical properties of transistors. Stochastic variability is an undesired and uncertain component caused by fundamental phenomena associated with device structure evolution, which cannot be avoided during the manufacturing process. In order to examine the problem of variability at atomic levels, the 'Motif' concept, defined as a set of repeating patterns of fundamental geometrical forms used as design units, is proposed to capture the presence of statistical variability and improve the device/circuit layout regularity. A set of 3D motifs with stochastic variability are investigated and performed by technology computer aided design simulations.

The statistical motifs compact model is used to bridge between device technology and circuit design. The statistical variability information is transferred into motifs' compact model in order to facilitate variation-aware circuit designs. The uniform motif compact model extraction is performed by a novel two-step evolutionary algorithm. The proposed extraction method overcomes the drawbacks of conventional extraction methods of poor convergence without good initial conditions and the difficulty of simulating multi-objective optimisations. After uniform motif compact models are obtained, the statistical variability information is injected into these compact models to generate the final motif statistical variability model.

The thesis also considers the influence of different choices of motif for each device on circuit performance and its statistical variability characteristics. A set of basic logic gates is constructed using different motif choices. Results show that circuit performance and variability mitigation can benefit from specific motif permutations. A multi-stage optimisation methodology is introduced, in which the processes of optimisation are divided into several stages. Benchmark circuits show the efficacy of the proposed methods. The results presented in this thesis indicate that the proposed methods are able to provide circuit performance improvements and are able to create circuits that are more robust against variability.