Date of Award

Spring 2011

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical/Computer Engineering

Committee Director

Shirshak K. Dhali

Committee Member

Helmut Baumbart

Committee Member

Julie Hao

Committee Member

Gon Namkoong

Abstract

Analog circuit design does not enjoy as much automation as its digital counterpart. Analog sizing is inherently knowledge intensive and requires accurate modeling of the different parametric effects of the devices. Besides, the set of constraints in a typical analog design problem is large, involving complex tradeoffs. For these reasons, the task of modeling an analog design problem in a form viable for automation is much more tedious than the digital design. Consequently, analog blocks are still handcrafted intuitively and often become a bottleneck in the integrated circuit design, thereby increasing the time to market.

In this work, we address the problem of automatically solving an analog circuit design problem. Specifically, we propose methods to automate the transistor-level sizing of OpAmps. Given the specifications and the netlist of the OpAmp, our methodology produces a design that has the accuracy of the BSIM models used for simulation and the advantage of a quick design time. The approach is based on generating an initial first-order design and then refining it. In principle, the refining approach is a simulated-annealing scheme that uses (i) localized simulations and (ii) convex optimization scheme (COS). The optimal set of input variables for localized simulations has been selected by using techniques from Design of Experiments (DOE). To formulate the design problem as a COS problem, we have used monomial circuit models that are fitted from simulation data. These models accurately predict the performance of the circuit in the proximity of the initial guess. The models can also be used to gain valuable insight into the behavior of the circuit and understand the interrelations between the different performance constraints.

A software framework that implements this methodology has been coded in SKILL language of Cadence. The methodology can be applied to design different OpAmp topologies across different technologies. In other words, the framework is both technology independent and topology independent.

In addition, we develop a scheme to empirically model the small signal parameters like 'gm' and 'gds' of CMOS transistors. The monomial device models are reusable for a given technology and can be used to formulate the OpAmp design problem as a COS problem.

The efficacy of the framework has been demonstrated by automatically designing different OpAmp topologies across different technologies. We designed a two-stage OpAmp and a telescopic OpAmp in TSMC025 and AMI016 technologies. Our results show significant (10–15%) improvement in the performance of both the OpAmps in both the technologies. While the methodology has shown encouraging results in the sub-micrometer regime, the effectiveness of the tool has to be investigated in the deep-sub-micron technologies.

DOI

10.25777/x0sw-pg67

ISBN

9781124696676

Share

COinS