Multi-Objective Optimization Based Design of High Efficiency DC-DC Switching Converters

International Journal of Power Electronics and Drive System (IJPEDS)
Vol. 7, No. 2, June 2016, pp. 379~386
ISSN: 2088-8694  379
Journal homepage: http://iaesjournal.com/online/index.php/IJPEDS
Multi-Objective Optimization Based Design of High Efficiency
DC-DC Switching Converters
Angelo Ambrisi*
, Massimiliano de Magistris*
, Raffaele Fresa**
* Dipartimento di Ingegneria Elettrica e delle Tecnologie dell’Informazione, University of Naples Federico II, Italy
** Scuola di Ingegneria, Università della Basilicata, Potenza, Italy
Article Info ABSTRACT
Article history:
Received Jan 4, 2016
Revised Mar 14, 2016
Accepted Mar 29, 2016
In this paper we explore the feasibility of applying multi objective stochastic
optimization algorithms to the optimal design of switching DC-DC
converters, in this way allowing the direct determination of the Pareto
optimal front of the problem. This approach provides the designer, at
affordable computational cost, a complete optimal set of choices, and a more
general insight in the objectives and parameters space, as compared to other
design procedures. As simple but significant study case we consider a low
power DC-DC hybrid control buck converter. Its optimal design is fully
analyzed basing on a Matlab public domain implementations for the
considered algorithms, the GODLIKE package implementing Genetic
Algorithm (GA), Particle Swarm Optimization (PSO) and Simulated
Annealing (SA). In this way, in a unique optimization environment, three
different optimization approaches are easily implemented and compared.
Basic assumptions for the Matlab model of the converter are briefly
discussed, and the optimal design choice is validated “a-posteriori” with
SPICE simulations.
Keyword:
DC-DC converters
Multi objective optimization
Optimal design
Stochastic algorithms
Copyright © 2016 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Massimiliano de Magistris,
Dipartimento di Ingegneria Elettrica e delle Tecnologie dell’Informazione
University of Naples Federico II
Via Claudio 21, 80125 - Napoli, Italy
Email: m.demagistris@unina.it
1. INTRODUCTION
Switching converters are of paramount importance in several applications, ranging from power
delivery, consumer electronics, automotive application etc. There is a well established body of knowledge
(see for example [1], [2]) where “pseudo optimal” design rules are set up, mainly basing on simplified
analytical models, single components optimization, topologies and control strategies choices. This approach,
although simplistic, leads to set the design parameters in an easy but robust way. Nevertheless a more
detailed design requires that several goals have to be met simultaneously within prescribed constraints, and
quite a significant number of design parameters have to be set [3]. This can become a strong need as long as
high efficiency, low dimensions and weight are pursued. A significant example arises in ubiquitous DC-DC
converters in battery operated devices, since the problem of energy efficiency and the reduction of weight
and volume are of primary importance. Some studies for improving the design in those terms are available in
literature [4]-[8], most of which focus their attention on the accurate modeling of some specific effects, in
such a way as to provide the designer with new insight for taking decisions on topology, control strategy and
parameters. In this context some effort has been given to consider the design as a multi-objective
optimization problem [9] [10], but still with a Monte Carlo approach. Some examples of application of multi-
objective optimization techniques to converters in different contexts can be found in [11], [12].

 ISSN: 2088-8694
IJPEDS Vol. 7, No. 2, June 2016 : 379 – 386
380
In this paper we demonstrate how, also for this application field, an appropriate formulation of the
optimization problem together with the state-of-art algorithms and simulation environments can directly be
applied, giving a full satisfactory answer to the design problem and providing the designer with more
accurate and sophisticated tools at an affordable computational burden.
We apply multi-objective optimization (MOO) formulation to the design of a hybrid control low
power buck converter, using a public domain MOO package [10], [13], to obtain a Pareto front in the design
parameters space [14], [15]. This will be done by adopting a simplified model of the design goals as
presented in [5]-[7], with the a-posteriori validation of the designed device by a SPICE simulation. The
availability of a detailed Pareto optimal set, along with giving to the designer more insight and flexibility in
the choices, is greatly helpful when the calculated optimal parameters have to be translated into the discrete
values of commercial components. The possibility of a more accurate modeling approach, by including
SPICE simulations for the evaluation of the design goals will be also discussed, giving an estimation of the
required additional computational effort.
2. STOCHASTIC OPTIMIZATION ALGORITHMS AND PARETO OPTIMAL FRONT
TRACING
Most of the real-world optimization problems involve multiple conflicting objectives that must be
mutually reconciled. These problems are called multi objective optimization problems (MOO), or vector
optimization problems, in contrast to single objective optimization (SOO), or scalar optimization problems
[16]. If we define as  1 2, ,...k Mf x x x the fit function associated to the k-th goal in a proper parameters space,
the optimization problem is often expressed as:
 1 2min ( ), ( ),... ( )N
M
f f f
 
x x x
x X 
(1)
where X is the feasible set of the decision variables vector x. In multi-objective optimization, usually it does
not exist a feasible solution that minimizes all objective functions simultaneously. Therefore, either the
problem is reduced to a scalar one by means of a weighted sum [17] of individual objectives, or a different
notion of optimal has to be defined, that is the well known Pareto optimal. A (feasible) solution is “Pareto
optimal” if it cannot be improved in any of the objectives without degrading at least one of the other
objectives. In such terms, different Pareto optimal solutions can be found, and a their set is defined as “Pareto
optimal front”.
In the last years, Multi-Objective Evolutionary Algorithms (MOEA) have demonstrated to be
extraordinary facilities for solving optimization problems in different areas. Evolutionary algorithms such as
the Genetic Algorithm [18] has become a standard approach, and schemes based on Simulated Annealing
[19] and Particle Swarm Optimization [20], [21] are now familiar. Currently, most evolutionary multi-
objective optimization (EMO) algorithms apply Pareto-based ranking schemes.
Such powerful, systematic and nowadays well assessed approach to optimal design, is still not so
commonly adopted in the electronic circuit design area. Main goal of this paper is show the possibility of
extending the use of MOEA optimization tools to the switching converters design. As simulation
environment we adopted a public domain MATLAB tool, known as the GODLIKE package [22] which
allows, in a unique environment, to use different algorithms and to compare directly their performances by
producing optimal ranking schemes. The test case proposed in this paper confirm the viability and the
effectiveness of this package in the definition of the Pareto optimal front for a typical electronic design
problem. In this way we suggest how, at present, it’s quite realistic to introduce the EMO algorithms in the
standard optimal design of electronic devices.
3. OPTIMAL DESIGN OF AN HYBRID CONVERTER
The effectiveness of the proposed approach is shown by pursuing the optimal design of hybrid
control low power buck DC-DC converter. We already mentioned how some basic important requirements in
their optimal design are to achieve high power efficiency, for best exploitation of batteries, and at the same
time high power density, to reduce volume and weight of the device. The choice of the circuit parameters and
control strategy to fulfill both requirements is not trivial.
In the standard design of fixed frequency converters, it is well known how high switching frequency
is required in order to achieve high power density, since the higher is the frequency, the smaller is the

IJPEDS ISSN: 2088-8694 
Multi-Objective Optimization Based Design of High Efficiency DC-DC … (Massimiliano de Magistris)
381
inductance and capacitance [5]. At the same time there is a trade off problem in the frequency choice, since
losses increase significantly at high frequencies. Converters operating at fixed frequency has the advantage of
simple design, but their efficiency rapidly decays at low loads since the switching losses are constant over the
whole load range (figure 1a); on the other hand, as shown in [6], converters operating at variable frequency
exhibit lower losses at low loads (figure 1b), but the design became more difficult because the range of the
switching frequency can be relatively wide. In Figure 1 is sketched a qualitative comparison between the
efficiency in the two control schemes based on the current literature
A more sophisticated strategy is the so called hybrid control, based on the idea of operating at
variable frequency in Discontinuous Current Mode (DCM) and at fixed frequency (the maximum one
allowed for the converter) in Continuous Current Mode (CCM).The control system select the frequency
according to the load (as it happens in the variable frequency converter) for low loads. As the load increases
and the converter switches to CCM, the control system select a fixed (maximum) frequency, which is
independent from the load.
switching
total
Figure 1. Power losses as function of load for fixed frequency control (a) and
variable frequency control (b) converters
The advantages of the hybrid control are revealed from the (qualitative) diagram reported in Figure
2 [7], where the continuous line (a) indicates the efficiency of a hybrid converter, while dotted line (b) and
dashed (c) the efficiency of converters with variable and fixed frequency. It appears clearly as in the hybrid
control system the efficiency shows an optimal trend, since it is quite independent from the load.
Figure 2. Comparative diagram of hybrid (a), variable (b) and fixed frequency control (c) as reported in [7]
Moreover the frequency range has been reduced since the frequency now has to vary only in DCM
and remains fixed in CCM. This reduction of the frequency range allows a simpler choice of the passive
components. Then it is clear how an hybrid control strategy can strongly improve the converter’s efficiency
helping at the same time the designer to make easier choices.
Once the topology and the control strategy have been fixed, the optimal design of the converter is
reduced to the problem of minimizing the size and maximizing the efficiency for variable loads. It can be

 ISSN: 2088-8694
382
explored with reference to the following case, where we consider a synchronous low power buck converter
with hybrid control and design constraints defined as follows:
 input voltage = 5 V;
 output voltage = 3.3 V;
 maximum current load = 500 mA;
 maximum ripple =2.5%.
The circuit topology is sketched in Figure 3
Figure 3. Schematics of the considered synchronous rectifier low power buck converter
For the optimal design we use the optimization criteria defined in [7] for hybrid converters, by
formulating a multi objective optimization problem. In particular we assume the inductance L and the
maximum switching frequency sf as decision variables, whereas the considered objectives to minimize are:
1. Power loss ratio (1 )E , where /( )loss loss loadE P P P  is the converter efficiency;
2. Capacitance 2
[ (1 )]/(8 )out s cC V L f V     , cV is the maximum output ripple and  is the duty-cycle;
3. Inductor diameter D (with other geometrical parameters kept fixed).
Note that objective 2 and 3 are directly related with size and weight of the converter, as well as to
the handiness in choosing or realizing the passive components. For the inductance L a simplified model of a
cylindrical air inductor with 0.30 mm copper wire diameter has been considered. Allowed ranges for decision
variables are 1-50 μH for L and 0.1-20 MHz for sf .
The model for the circuit operation and for losses is the same as described in [6], [7]. Solving the
optimization problem (1), we get a set of optimal results (not dominated in the Pareto sense), each of them
representing the best reachable goal for a selected direction of search. In our case, as said, this has been
achieved within the MATLAB®
environment and GODLIKE package, by using and comparing Genetic
Algorithm, Simulated Annealing and Particle Swarm Optimization algorithms. Among the possible settings
for those algorithms, few parameters are most significant, namely initial population, number of generations
and Pareto fraction (which is the number of points on the final front as compared to the initial population).
High values of the initial population makes anyway the Pareto front more detailed, as well as an high value of
the number of generations improve the accuracy of the results. These values have to be considered in a trade
off with a reasonable computational cost once the interest problem is specified. Moreover some limitation to
the population size arises intrinsically in the algorithm implementations, as it will be evidenced hereinafter.
With reference to the considered (simplified) model for the direct multi-objective optimization
problem, we have set the initial population parameter to range from 210 to 2400 points (higher values lead to
non convergence problems of the code when compared with different algorithms). After some trials we
estimated that 20 is good choice for the number of generations. Finally, for each simulation we assumed an
unitary value for the Pareto fraction. In Table 1, we compared the performances of the different algorithms,
by using a single-core Intel i5-5200U at 2.2 GHz for the computations.
The results in term of Pareto front evaluation are reported in Figure 4, for the case of GA, 2400
points. The availability of the complete Pareto optimal front, allows making an easier choice of the final
design. Furthermore, a reduction of the possible choices is easily obtained by restricting a-posteriori some
design variable intervals that correspond to select a subset of the entire Pareto Front to offer to the designer.

383
For example, if we ask a maximum power efficiency of at least 88%, using a capacitance value less than 1
μF, we get the subset of points reported in Table II (with reference to the SA algorithm results). Among
them, the possible final design choice is evidenced in Figure 4, being characterized by closest commercial
values for inductance and capacitance, respectively of 2.2 μH and 560 nF.
Table 1. Computation times for different algorithms
algorithm
initial
population
Pareto
points
time [s]
GA 210 210 4.15
SA 210 210 4.10
PSO 210 210 4.37
GA 600 600 7.21
SA 600 600 7.16
PSO 600 600 8.02
GA 2400 2400 33.7
SA 2400 2400 32.8
PSO 2400 2400 37.4
With the proposed choice we get a maximum efficiency of 90% in CCM with maximum load (500 mA) at a
maximum switching frequency 1.16sf  MHz. As the load decreases, the efficiency also becomes lower,
reaching the value of 85% at 218 mA of current load (this is the boundary between the CCM mode and the
DCM mode). For the adopted control scheme the efficiency remains almost constant for lower loads.
Figure 4. Pareto front of the problem (2400 points) as obtained by SA in GODLIKE.
The final design choice is evidenced by a large dot
Table 2. A subset of optimal design choices (Pareto front with 2400 points)
L [μH] f [MHz] LOSS % C [nF] D [mm]
14,8 0,41 7,68 681 11,6
4,55 0,70 7,86 772 6,44
6,38 0,74 8,42 482 7,63
6,56 0,80 8,76 406 7,73
5,91 0,86 9,02 386 7,34
28,5 0,57 9,97 187 16,1
22,4 0,68 10,0 165 14,2
2,19 1,16 10,1 558 4,96
12,5 0,93 10,4 156 10,7
16,7 0,91 10,7 122 12,3
10,3 1,06 10,8 147 9,71
7,98 1,21 11,3 145 8,53
31,6 0,76 11,3 092 17,0
36,3 0,78 11,7 077 18,2
39,4 0,78 11,9 071 18,9
00.10.20.30.40.50.60.7
0
2
4
6
x 10
-6
0
0.005
0.01
0.015
0.02
0.025
Power Loss Ratio (1-E)
Simulated Annealing
Capacitance [F]
InductorDiameter[m]

 ISSN: 2088-8694
384
4. SPICE VALIDATION OF THE OPTIMAL DESIGN
A validation of the described design procedure can be straightforwardly obtained by using a full
SPICE simulation of the designed circuit with the parameters value returned by the MATLAB optimized
design and realistic models for the PMOS and NMOS. Figure 5a, b show the results of a test case of this
validation step.
In these diagrams, V(10) is the control voltage: when this signal is high, the upper MOS (PMOS) is
on; I(L1) shows the classical trend of the inductor current; V(3) and I(Rload) are the load voltage and current.
The values fully agree with the results of the MATLAB optimized design, and the simplifying assumptions
used for the optimization problem are validated.
(a)
(b)
Figure 5. SPICE validation: (a) transient; (b) regime
Although in the considered test case the simplified model shows to be fully satisfactory, an
important extension to the proposed scheme could be the inclusion of the full SPICE simulation in the
optimization loop. This can be in principle easily realized by using a SPICE model for the forward problem,
including in the MATLAB optimization loop a proper call to SPICE and using the evaluations of SPICE for
the computation of the objective functions. This scheme is in principle quite interesting since avoids any
simplifying assumptions in the device model, in such a way to enlarge the scope of the optimal design to
those circuits where getting reliable simplified models is more difficult. On the other hand in such
implementation, the evaluation of the direct problem (a SPICE call) easily became dominating in the
computational burden, resulting in the total simulation time roughly proportional to the number of SPICE
calls multiplied by the single SPICE simulation time.
5. CONCLUSION
We have shown how the application of multi-objective optimization gives nowadays useful and
realistic opportunities for the optimal design of switching converters. The possibility of calculating a
complete Pareto optimal front, as well the corresponding decision parameter set, gives to the designer a more
complete picture of the problem parameters space at affordable computational time and burden. Moreover,
the complete knowledge of not dominated optimal set allows more easily the choice of commercial discrete
values for components, keeping the achieved optimal goals. The relative easiness of using public domain
standard optimization environments make us confident of the straightforward possibility in extending such
tools to a broader class of converters, and more generally design problems in electronic circuits. More
accurate modeling of the device, as for example with a full SPICE numerical model, can be in principle
integrated in the optimization environment via Matlab, at the price of increasing significantly the
computational burden.
ACKNOWLEDGEMENTS
The authors wish to recognize the contribution of Marco Sorrentino, who performed significant
preliminary analysis for this work during his Master Thesis in Electronic Engineering, at University of
Naples FEDERICO II, Italy.

385
REFERENCES
[1] Hart D.W., Introduction to Power Electronics, Prentice Hall, Upper Saddle River, New Jersey USA, 1996.
[2] Shepard, W., Hull L.N., Liang D.T.W., Power Electronics and Motor Control”, 2nd ed., Cambridge; New York,
NY, USA: Cambridge University Press, 1995.
[3] Tourkhani F, Viarouge P., Meynard T.A., A Simulation-Optimization System for the Optimal Design of a
Multilevel Inverter, IEEE Transactions on Power Electronics, vol. 14, no. 6, 1999.
[4] Katayama Y., Sugahara S., Nakazawa H., Edo M., High-Power-Density MHz-Switching Monolithic DC-DC
Converter with Thin-Film Inductor, Power Electronics Specialists Conference, 2000. PESC 00. 2000 IEEE 31st
Annual Publication Volume: 3, page(s): 1485-1490 Digital Object Identifier: 10.1109/PESC.2000.880526
[5] Kollman R., Collins G., Plumton D., 10 MHz PWM converters with GaAs VFETs Applied Power Electronics
Conference and Exposition, 1996. APEC '96. Conference Proceedings 1996. Eleventh Annual Volume 1, 3-7
March 1996 Page(s): 264-269 vol.1 Digital Object Identifier 10.1109/APEC.1996.500453
[6] Arbetter B., Erickson R., Maksimovic D., DC-DC converter design for battery-operated systems, Power Electronics
Specialists Conference, 1995. PESC '95 Record. 26th
Annual IEEE Volume 1 , 18-22 June 1995 Page(s):103-109
vol.1 Digital Object Identifier 10.1109/PESC.1995.474799
[7] Zhou X., Wang T.G, Lee F.C., Optimizing design for low voltage DC-DC converters, Applied Power Electronics
Conference and Exposition, 1997. APEC '97 Conference Proceedings 1997, Twelfth Annual Volume 2 , 23-27
Feb. 1997 Page(s):612-616 vol.2 Digital Object Identifier 10.1109/APEC.1997.575633
[8] Gezgin C., Heck B.S., Bass R.M., Simultaneous Design of Power Stage and Controller for Switching Power
Supplies, Power Electronics, IEEE Transactions on, Volume: 12, Issue: 3 page(s): 558-566 May 1997 ISSN: 0885-
8993, Digital Object Identifier: 10.1109/63.575683.
[9] Negebauer T.C., Perreault D.J., Computer Aided Optimization of dc/dc Converters for Automotive Applications,
IEEE Transactions on Power Electronics, Vol. 18, No. 3, May 2003
[10] Pingqiang Zhou; Won Ho Choi; Bongjin Kim; Kim, C.H.; Sapatnekar, S.S., "Optimization of on-chip switched-
capacitor DC-DC converters for high-performance applications", in Computer-Aided Design (ICCAD), 2012
IEEE/ACM International Conference on , vol., no., pp.263-270, 5-8 Nov. 2012
[11] Elkholy, M.M., Elhameed, M.A., Braking of three phase induction motors by controlling applied voltage and
frequency based on particle swarm optimization technique (2015) International Journal of Power Electronics and
Drive Systems, 5 (4), pp. 520-528.
[12] Sreedhar, M., Dasgupta, A., Ray, S., Harmonic mitigated front end three level diode clamped high frequency link
inverter by using MCI technique (2014) International Journal of Power Electronics and Drive Systems, 4 (1), pp.
91-99.
[13] Amirahmadi, A.; Dastfan, A.; Yassami, H.; Rafiei, S.M.R., "Multi-Objective optimum design of controller for PFC
rectifier using NSGA-II algorithm", in Power Electronic & Drive Systems & Technologies Conference (PEDSTC),
2010 1st , vol., no., pp.180-185, 17-18 Feb. 2010
[14] Larouci, C., Ejjabraoui, K., Lefranc, P., Marchand, C., Impact of control constraint on a multi-objective buck
converter design (2012) International Journal of Power Electronics and Drive Systems, 2 (3), pp. 257-266.
[15] Srinivas N., Deb K., Multiobjective optimization using non-dominated sorting in genetic algorithms, Evolutionary
Computation 2, no. 3, pp. 221-248, 1994.
[16] Mirjafari, Mehran; Balog, Robert S. 'Survey of modelling techniques used in optimisation of power electronic
components', IET Power Electronics, 2014, 7, (5), p. 1192-1203
[17] Singh, D., Singh, B., Singh, N., Performance indices based optimal tuning criterion for speed control of DC drives
using GA (2014) International Journal of Power Electronics and Drive Systems, 4 (4), pp. 461-473.
[18] Adaptation in Natural and Artificial Systems: An Introductory Analysis With Applications to Biology, Control, and
Artificial Intelligence- MIT Press/ Bradford Books edition - John H. Holland - 1992
[19] S. Kirkpatrick; C. D. Gelatt; M. P. Vecchi, Optimization by Simulated Annealing, Science, New Series, Vol. 220,
No. 4598., May 13, 1983
[20] Kennedy, J.; Eberhart, R. (1995). "Particle Swarm Optimization", Proceedings of IEEE International Conference on
Neural Networks. IV. pp. 1942–1948
[21] Altinoz, O.T.; Erdem, H., "Evaluation function comparison of particle swarm optimization for buck converter," in
Power Electronics Electrical Drives Automation and Motion (SPEEDAM), 2010 International Symposium on ,
vol., no., pp.798-802, 14-16 June 2010
[22] Rody Oldenhuis, “GODLIKE A robust single & multi-objective optimizer”, Available at:
http://www.mathworks.com/matlabcentral/fileexchange/24838-godlike-a-robust-single---multi-objective-optimizer

 ISSN: 2088-8694
386
BIOGRAPHIES OF AUTHORS
Angelo Ambrisi received the Ph.D. degree in Electrical Energy Conversion from Second
University of Naples in 2010. He his currently Teacher of Computer Science in Secondary
Technical School, on leave toward a Ph.D in Electrical Engineering at University of Naples
FEDERICO II.
His research interests are in the general areas of Optimization Problems, developing applications
for applied electromagnetic problems (Worst Case Analysis - Biomedical - Nuclear fusion).
Currently he is tackling problems in the optimization of electrical circuits, as well as the
application of optimization techniques to the identification of electromagnetic configurations in
Tokamak devices.
Massimiliano de Magistris received the Ph.D. degree in Electrical Engineering from University
of Naples FEDERICO II in 1994, where he is presently Associate Professor at Department of
Electrical Engineering and Information Technology, University of Naples FEDERICO II,
teaching courses on basic and advanced circuit theory.
His research interests have been in the general areas of applied electromagnetics and electrical
circuits, plasmas and accelerators engineering, electromagnetic compatibility. His current
research interests are in the circuital macro modeling of electrical and electronic systems, electro
thermal compact modeling of circuits and devices, non linear circuits and complex networks
analysis.
He has been author of more than 80 peer reviewed scientific papers for international journals,
conferences and books, and is a reviewer for several primary scientific journals and conferences.
Raffaele Fresa received the Ph.D degree in Electrical Engineering from University of Naples
"Federico II" in 1994. At present he is associate professor at School of Engineering of Basilicata
University, Italy, teaching courses on basic circuit theory and numerical modelling of circuit and
electromagnetic fields. In 2015 he received the Full Professor qualification in Electrical Sciences
His research interests include analytical and numerical methods for Electromagnetic fields and
coupled problems, nuclear fusion, non-destructive testing (NDT) and optimization of electrical
systems. He has been author of more than 70 peer reviewed scientific papers for international
journals, conferences and books; reviewer for several primary scientific journals and
conferences; co-author of several electromagnetic codes including the CARIDDI code used in
many international laboratories for the calculation of the eddy currents in the passive
components of nuclear fusion reactors.

Multi-Objective Optimization Based Design of High Efficiency DC-DC Switching Converters

More Related Content

What's hot (20)

Similar to Multi-Objective Optimization Based Design of High Efficiency DC-DC Switching Converters (20)

More from IJPEDS-IAES (20)

Recently uploaded (20)

Multi-Objective Optimization Based Design of High Efficiency DC-DC Switching Converters