Automatic design of analog integrated circuits under process parameter variability

• Autor(a) principal: Quirino, Felipe Antunes
• Data de Publicação: 2020
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: eng
• Título da fonte: Repositório Institucional da UNIPAMPA
• Texto Completo: http://dspace.unipampa.edu.br:8080/jspui/handle/riu/5381
• topic: CNPQ::CIENCIAS EXATAS E DA TERRA; Ciência da computação; Algoritmos; Circuitos integrados lineares; Computer science; Algorithms; Linear integrated circuits

Analog integrated circuits have a high range of applications, from interface circuits to signal processing. These systems need to be carefully designed in order to achieve a suitable trade-off between performance and power consumption. Traditional method of designing an analog circuit is based on trial-and-error. The designer uses his own experience for testing and modifying circuit parameters with the aid of an electrical simulator, until achieving the desired solution. However, the process of sizing the circuit requires several hours of design. In addition, a SPICE simulation can take a long time. The designer needs to wait for the end of the simulation, verify the results, and from there to do another simulation. An alternative for design automation is to abstract the circuit as an optimization problem. Using an optimization algorithm it is possible to explore the design space in the search for an optimum solution. This work demonstrates an optimization analysis performed with a low-voltage bulk-driven operational transconductance amplifier. Previous works demonstrated the modeling of this circuit as an optimization problem, but only for nominal values. However, a nominal analysis presents the performance disregarding process parameter variability that affect the performance. In the present work, we propose a design automation tool of analog integrated circuits using yield analysis, i.e, estimating performance with Monte Carlo electrical in order to evaluate the impact of process variability on circuit performance. In general, optimization algorithms present random variables (i.e., even with the same parameters, different seeds for the random number generator function may converge to different results). In order to further understanding this behavior, this work analyses the behavior of many executions of the algorithm. The adjustable parameter on the present algorithm (Cuckoo Search) is the number of nests. We performed the simulation 30 times for the same number of nests, varying the number of nests from 10 to 490 with a total of 1470 executions. As result, this work demonstrates a statistic analysis of all designs intending to find the best parameters for the tool. After getting the parameters, we analysed the circuit behavior for the worst, median and best cases, with the fixed parameter. The tool is able to design analog integrated circuits automatically, however without a guarantee of a feasible solution. For instance, in the worst case example, the algorithm converges to an unfeasible solution for practical terms. However, this could be mitigated with a large member of iterations or with more than one execution. In the best case, the amplifier designed with the optimization tool presents improvements in terms of (8.86nW of mean ± 0.22 of standard deviation in comparison to 18nW designed manually), PM (59.99 ∘ ± 10.15 in comparison to 52.50 ∘ ) and GBW (3.81kHz ± 0.65 in comparison to 1.88kHz).