variate, muti-objective optimization algorithm for arbitrary cost

functions having multiple local optima. It can be used by circuit

designers to quickly explore different circuit schematics to iden-

tify the right choice. It also helps in finding a robust design solu-

tion against variations in the manufacturing process, tempera-

ture, and supply voltage. It takes as input an unsized SPICE net-

list along with design specifications and constraints. It calls a

SPICE simulator in a loop to evaluate the circuit at different test

points. Finally, it produces an optimized and centered netlist.

Inputs

• SPICE netlist of circuit to be optimized, and/or a set of design

equations the user wants to optimize.

• Set of design objectives which the circuit should meet and a

set of constraints which must be honored. Any quantity which

can be simulated and measured can be a design objective.

• Set of design variables which the tool can tune, along with

their allowed range of values. Typically, these design variables

represent transistor dimensions, bias voltages, etc.

• Set of “corners”, including manufacturing process corners (e.g.

fast, typical, slow), different temperatures and different supply

voltages. AnXplorer optimizes the circuit to ensure that design

objectives are met across these corners.

Outputs

• An optimized and centered circuit netlist, which meets or

exceeds the design objectives across all the user specified cor-

ners.

• An exploration database, which is a persistent storage for all

design points explored by the tool. This database can be que-

ried and used for effective trade-off analysis between conflict-

ing design objectives.

Pictorially, the inputs and outputs are shown in figure 1. Figure 2

explains how AnXplorer automates the existing design flow.

Figure 1: Inputs and outputs of AnXplorer

Figure 2: Analog design flow with AnXplorer

Currently, AnXplorer works seamlessly with several industry

standard SPICE simulators.

DIFFERENTIATING FEATURES

Design equation based optimization: In addition to simulation

based design optimization, AnXplorer also supports design equa-

tion based optimization, or a combination of both. Unlike

gradient-search based tools, AnXplorer supports arbitrary func-

tions with multiple local and global optima.

Hierarchical design objectives: Unlike a weight based prioritiza-

tion of multiple objectives, commonly used in other optimization

tools, AnXplorer employs a hierarchical arrangement for design

objectives. The user specifies the relative order of importance of

each objective, instead of a weight. This is specified in the form of

a directed graph. The tool tries to achieve the objective with

higher priority before optimizing the other objectives.

Trade-off analysis with exploration database: AnXplorer records

all design points explored during the optimization process and

these are stored in a persistent database. This database can be

queried. This creates an effective tool for “what-if” analysis (how

much can one objective be achieved, if another is constrained

below a certain value?). It also enables trade-off analysis for con-

flicting objectives (e.g. circuit area and noise, speed and power) if

all objectives cannot be realistically met.

Implicit objectives: In addition to explicit objectives specified by

the user, AnXplorer imposes certain implicit objectives on the

operating condition of devices. Transistors which are supposed to

be in saturation are pushed deep into saturation. This ensures an

inherently robust design.

VALUE ADDITION

Productivity improvement: AnXplorer improves engineering

efficiency by at least 5X. This is achieved by automating the

manual and routine job of sizing circuit elements to meet design

objectives. This leads to faster time-to-market. The productivity

improvement enabled by AnXplorer has other benefits as well. It

helps engineers to quickly judge the suitability of any given

schematic for a particular purpose, and thus helps in evaluating

multiple options before selecting a topology. It also helps less

experienced engineers to take up challenging design tasks.

Yield improvement: AnXplorer helps achieve a robust design

point for any circuit, so that it meets its design objectives in the

face of variation in the manufacturing process, temperature, and

supply voltage. This is a major challenge in the current manual

design flow, and is responsible for most silicon re-spins. AnX-

plorer, by rigorously exploring the design space for robust solu-

tions, improves yield and increases the chances of first-time sili-

con success, thereby saving millions.

Porting across process nodes: As manufacturing processes

change every day, it becomes necessary to port designs from one

process node to another. This involves re-sizing the circuit ele-

ments so that the circuit meets its objectives in the new process.

AnXplorer can automate this task. This is especially helpful for

analog IP companies who often have to re-target the same design

for different customer foundries.

EXAMPLES OF SIMULATION BASED OPTIMIZATION

Two-stage OpAmp

Source: Krasnicki, M. J., Phelps, R., Hellums, J. R., McClang, M.,

Rutenbar, R. A., Carley, L. R., “ASF: A practical simulation-based

AnXplorer: EDA Tool for Schematic Level Optimization of Analog & RF Circuits