Fuzzy logic technique for accurate analog circuits macromodel sizing

Abstract

SUMMARYDesign of analog integrated circuits requires extensive simulations at low levels of the analog design hierarchy. The simulations increase the overall design time and, thus, the analog part of a mixed ASIC requires most of the design time while using only a small part of the silicon die. To improve the efficiency of these simulations, it is necessary to use methodologies and tools based on hierarchical descriptions, where macromodels play an essential role. In this paper, an approach for applying fuzzy logic for accurate analog circuit macromodel sizing is presented. In our proposed method, multiple adaptive neuro-fuzzy inference systems are trained to predict the performance characteristics (gain, bandwidth) of CMOS analog circuits. Moreover, the presented methodology provides reusable macromodels, since it is applicable for large number and large range of design parameters. This technique is applied to the accurate sizing of fully differential telescopic operational transconductance amplifier macromodel, a circuit widely used in the design of high-performance analog-to-digital converters. The neuro-fuzzy computed characteristic values are in excellent agreement and one order of magnitude faster than those obtained from device level SPICE simulations. Moreover, this method offers the best accuracy in comparison with other classical techniques such as polynomial regression, spline interpolation or artificial neural networks.

Synthesis of CMOS analog circuit VHDL‐AMS descriptions using parameterizable macromodels

“…This approximation simplifies the model complexity. Figure 7 shows this macromodel, the behavior of which has already been proven by modeling a gain-boosted telescopic amplifier [22].…”

Alternate characterization technique for static random‐access memory static noise margin determination

“…It applies a combination of the least-squares method and the backpropagation gradient descent method for training the fuzzy inference system membership function parameters to emulate a given training data set. ANFIS architecture has been successfully used to model nonlinear functions or for chaotic time series prediction [18].…”

Behavioral Modeling of Mixed-Mode Integrated Circuits

“…Some recent developments related to symbolic behavioral modeling can be found in (Fakhfakh et al, 2010). The generation of behavioral models is very useful to perform different design tasks, such as synthesis (Saad & Soliman, 2008) and sizing (Diaz-Madrid et al, 2008). The applications to analog design also include behavioral modeling of power (Suissa et al, 2010), carbon nanotube field-effect-transistors (Chek et al, 2010), statistical modeling (Li et al, 2010), efficient RF/microwave transistor modeling (Gaoua et al, 2010), etc.…”