You are on page 1of 3

SYNOPSIS

The paper addresses the use of multi-objective evolutionary algorithm for the synthesis of hybrid
CBL/CMOS cell. The evolutionary algorithm named as ‘Strength Pareto Evolutionary Algorithm
(SPEA) is used to find sizes of transistor ratio W/L of hybrid CBL/CMOS cell. The transistor
sizes so obtained are called ‘Pareto Optimal Solutions’, which are intended to optimize some
circuit specifications. These circuit specifications are formulated as objective functions to be
minimized. The paper has focused on following objective functions (also called ‘cost factor’).

1. Peak supply current ( ‘Idd’ )

2. Peak supply voltage ( ‘Vdd’ )

3. Average power consumption ( ‘Pow’ )

4. Rms supply current (Idd(rms))

5. Propagation delays (Tdelay , Tfalll , Trise)

6. Power-delay product (PDP).

ELDO simulator has been used to carry out simulation and measurement of each coded
solution. The non-dominated W/L solutions of each transistor are plotted w.r.t given
objective function and thus optimum value of W/L ratio is extracted for each transistor. In
conclusion, dependency of various objective functions (Idd(peak), Vdd(peak), Idd(rms), Pow(avg.),
propagation delays, PDP) is shown and final solutions are discussed in tabular form.

Methodology: The work has been divided into three sub-sections to achieve the final
solutions.

1. Design and simulation of hybrid CBL/CMOS cell.

2. Using SPEA (Strength Pareto Evolutionary Algorithm) as multi-objective optimizer.

3. Applying “clustering algorithm”.


The hybrid CBL/CMOS cell is designed using ELDO simulator (0.35 CMOS
technology). The simulation is performed using low level signal (voltage = 0.162V) and a
supply current of 0.47mA.

The deviation (relative error) between calculation and simulated sizes is observed which
comes out to be 21% of PMOS tree and 0.4% of the PMOS CBL transistor.

Next, SPEA (a stochastic optimization procedure) optimizes the vector of objective functions
as described above. The solutions obtained from SPEA are called non-dominated Pareto-
optimal solution. Following values of various parameters is used for SPEA in all the runs of
algorithm.

- I/P pulse train of 100 KHz, 3.3V.

- Maximum size of external archive = 80

- Crossover probability = 0.8

- Mutation probability = 0.01

The solutions are stored in external archive E where clustering algorithm is applied. Main
goals of clustering algorithm are:-

1. To avoid that the external archive size does not exceed the user defined maximum size.

2. Encourage dissimilarities between stored solutions.

Following restrictions are imposed to reduce input search space:

3 < Wi < 0.3 transistors

2 < Li < 0.3 transistors

Advantages of Pareto optimal approach are:-


- It provides a set of Pareto-optimal solutions rather than a single solution to get a deeper
insight into the problem and to provide alternate solutions.

- It converges faster than any other multi-objective optimization method used for example-
as Energy minimization evolutionary algorithm (EMEA).

References:

• Multi-objective optimization using evolutionary algorithms by Kalyanmoy Deb from


http://books.google.co.in/

• Proceedings of Summer Program 2000 from “Center for Turbulence Research”

You might also like