Designing Single-Switch Forward Converters

By Suresh Hariharan, Senior Corporate Applications Engineer, and David Schie, Director of IC Design, Maxim Integrated Products Inc., Sunnyvale, Calif. Often used in dc-dc converter modules for power levels below 100 W, single-transistor, resonantreset forward converters are also useful for dc-dc converters with adjustable output voltages.
mong power-converter topologies, the single-transistor forward converter is one of the most common for power levels below 100 W. This article, however, focuses on the improvements to the circuit known as the single-transistor, resonant-reset forward converter, which eliminates the reset winding and a diode (DTR) while offering several other advantages. Its duty cycle can exceed 50%, making it suitable for low-cost dc-dc converters that operate from a wide range of input voltages and deliver widely varying outputs. The absence of a reset winding reduces costs by simplifying the transformer, especially for the planar transformers widely used in high-density dc-dc converter modules. Finally, the resonantreset circuits sinusoidal reset voltage reduces EMI. To properly appreciate the resonant-reset topology, we must rst understand the conventional single-switch forward converter (Fig. 1). When switch Q1 turns on, the transformer current rises from zero and the diode DTR is reverse biased. Transformer magnetizing current builds up to a value IM=VINTON/LM, where TON is the on time per switching cycle and LM is the magnetizing inductance. During the switchs on period, the load current I O is
VIN TR DR LO DF IO VO VIN IO C P CRS
TR

reected in the primary as IP=IONS/NP, where NS is the number of secondary turns and NP is the number of primary turns. Output voltage is VO=VINDNS/NP, where D=TON/ TS and 1/TS is the switching frequency. Magnetizing current in the transformer primary just before turnoff is VINTON/ LM. When Q1 turns off, the transformer voltage tends to reverse. Voltage on the DTR cathode keeps decreasing until DTR turns on. For typical applications, the NP/NR turns ratio is 1, where NR is the number of turns in the primary reset winding. The transformer magnetizing current now decreases from IM to zero. When it reaches zero, the transformer is fully reset and voltage across the transformer remains at zero until the start of the next switching cycle. The maximum duty cycle (DMAX) in these applications is limited to 50%. On the other hand, single-switch, resonant-reset forward converters (Fig. 1) are characterized by the absence of a reset winding. During the off time, the transformer resets (without loss) through a resonant circuit consisting of the magnetizing inductance and the combined capacitance of the switch (CS), primary winding (CP) and all reected secondary capacitances (CRS), including the rectifyingdiode capacitance.

Description of Operation
VO

As su m p t i on s a re made in the following circuit analysis: C IN CO The circuit has DR CD CIN reached steady-state operation. D D Q1 Q1 L and C (fairly O O CS DTR large) can be considered G innite. S G S Leakage induc(a) (b) tance is neglected. Fig. 1. A conventional single-transistor forward converter (a). A single-switch, resonant-reset forward converter (b). Drops due to the
LO CO

Power Electronics Technology October 2005

38

www.powerelectronics.com

FORWARD CONVERTERS
diode and switch on-resistance are neglected. Steady-state operation for the circuit comprises three intervals in each switching cycle: Interval 1. Initially, t=0 and Q1 is on ( Fig. 2a ). The transformer is magnetized with a ramp current during this interval, dened as TON. Secondary current ows through
VIN IP + I MAG IR + DF + IS LO IO VDS VIN O I MAG CD G S DR + I2 T=O I1 TON T

the secondary diode DR, and the voltage across capacitance CD is approximately zero. CD includes the internal diode capacitance and the external capacitance added across diode DR. The primary magnetizing current has a value of I1 at the start of this interval and I2 at the end of the interval: I2 = I1 + VIN TON LM

CIN

LM

NP D

NS

CO

TON

(a)
VIN LO IO VDS VIN NS D G SC
S

+ CR I MAG LM CT NP

IP I R IS

VIN + I 2

L M / CR

C IN

DF +

CO

+ O I MAG T

CD + DR

I2 T=O I1=-I2 TON TR

(b)
VIN + IP = 0 IS

LO

IO VDS

V IN + I 2 L M / C R

I MAG CIN LM NP D G CD NS

DF +

CO

VIN O I MAG I2

(Eq. 1) The primary current IP is the sum of the reected current IR (equal to IONS/ NP) and the primary magnetizing current IMAG. In t e r v a l 2 . Wh e n t h e switch is turned off, the switch Q1 drain-to-source voltage beg ins to r ise (Fig. 2b). When that voltage exceeds V IN , the polarity a c ro s s t h e s e co n d a r y winding is reversed. Then the secondary diode DR turns off and the freewheeling diode DF turns on. A sinusoidal demagnetization current star ts to flow through the resonant circuit formed by the parallel combination of transformer magnetizing inductance L M and the capacitance C R reflected across the transformer primary. The capacitance CR is the sum of all capacitances across the primary including the reflected capacitance C D, the internal plus external capacitance across diode DR (internal diode capacitance of DR<<CD): N CR = CD S + CS + CT N
P 2

(Eq. 2) where CS is the primary S T=O switch capacitance and CT + T DR is the transformer primary I =-I TON TR capacitance. Interval 2 TS equals TON + TR, where TR (c) is one-half of a resonant interval: Fig. 2. From Fig. 1b, an equivalent circuit for the voltage on Q1 and the primary magnetizing-current 1 f RES = waveform during Interval 1not to scale (a). From Fig. 1b, an equivalent circuit for the voltage on Q1 and the 2 L M C R primary magnetizing-current waveform during Interval 2not to scale (b). From Fig. 1b, an equivalent circuit
1 2

for the voltage on Q1 and the primary magnetizing-current waveform during Interval 3not to scale (c). www.powerelectronics.com

39

Power Electronics Technology October 2005

FORWARD CONVERTERS
C41 .1 F, 16 V -48 VOUT C53 1000 pF NDRV R49 VCC 1 2 T4

Q9 4 CMLT3946E CMLT3946E PO368

ISENSE Circuit
3 L4 47H 47 R53 Q11 Q1 1 Si3458DV C55 2.2 nF nF, 250 Vac c -48 VOUT C56 220 pF pF, , 100 V
4.7 F, 25 V

VIN

15T -48 VOUT 3 4

19T 10 9

D23

1N4148 W

Primary PWM Control Circuit

75 1N4148 W D20

R51 22

100 pF , 200 V

PWRAMP_TRNS 12 1 T5 PWRAMP_TRNS

VOUT

Q10 Q1 0 C54 R52 Si3440DV 22

C59

C60

C61

4.7 F, 25 V 4.7 F, 25 V 0.1 F, 16 V

C62

D21 1N4148 W D22 1N4148 W

15T 6 Q14 Q1 4 FQD4N25 R60 .47 REG5 C64 0.1 F SOURCEQ9 SOURCEQ

6T 3T

8 7

15 R62 Gate Gat

Q1 Q15 INDS351AN

-48 VOUT -48 VOUT R59

U9 5 4 3 VCC /OUT GND PS9121 CA AN 1 2

D24 4.99 k IN4148 W

-48 VOUT

Secondary VSENSE and Feedback and Control

VRE REF F C42 D17 B0520 Gate R40 10 k

R37 NDRV 22

C40

T3

0.1 F, 16 V -48 VOUT 8 PAO184

0.1 F. 16 V 5

Fig. 3. Resonant-reset forward converter with an input range of ground to 48 VOUT (36 V to 56 V) and output range 4 V to 18 V.

(Eq. 4) The external capacitance CR charges from zero to a peak value of: L I2 M CR during this interval, and then discharges back to zero. The magnetizing current I1 at the end of the interval should therefore equal I2. The drain-to-source voltage (VDS) on the primary switch Q1 at the end of this interval is VIN, but reaches a peak of: L M halfway through Interval 2. VIN + I2 C
R

TR = L M C R

is possible if L M C R > TR. In that case, a full half-cycle of resonance has not been completed before the next switching cycle begins, and therefore the voltage across the primary switch exceeds VIN at the start of each switching cycle. That condition increases the switching loss.

Transient Operation
Transient stresses on the primary switch and secondary output diodes can vary greatly depending on the type of controller used in the application. If the design is not optimal, transients can cause failure in the primary switches or the secondary diodes. Consider operation with a current-mode PWM controller. Initially, the power supply operates at no-load and high-line voltage. A load transient is applied (minimum load to full load), which causes an immediate duty-cycle step to maximum duty cycle. In turn, that event causes a large increase in the transformers magnetizing current and may saturate the transformer unless its design accounted for such transients. The resonant-reset voltage is much higher than that during steady-state operation and may cause failure in the forward diode or the primary switch. To combat this problem, we introduce a volt-sec clamp. Consider the controller above with a maximum duty-cycle clamp that is inversely proportional to the input voltage. That arrangement limits the maximum ux excursion along the B-H loop of the transformer during a transient, which allows the use of a smaller transformer. Transient-voltage stress on the forward diode and the primary switch is

Interval 3. During this interval, diodes DR and DF are both on, and the primary switch is off (Fig. 2c). Voltage across the transformer primary is held to zero by the reected virtual short across diode DF, and the magnetizing current is held to -I2 for the entire interval. The end of Interval 3 denes the end of a switching cycle, and because the circuit is at steady state, the current I1 equals -I2. Substituting for I1 in Eq. 1, we see that the primary magnetizing current at the start of each switching cycle is: I1 = (VIN TON ) 2 LM (Eq. 5)

During the entirety of Interval 3, the voltage across the transformer primary is held at 0 V, so the primary switch voltage VDS remains at VIN. Note that at the end of TS, I2I1 40

Power Electronics Technology October 2005

www.powerelectronics.com

FORWARD CONVERTERS
signicantly less, but is still higher than during steady-state operation. Now consider the operation of this converter type with a very light load using diodes for rectication. Magnetizing current is very close to zero during this mode of operation, and the duty cycle is low. If we now apply a load transient (from no load to full load), the duty cycle immediately increases to the maximum value allowed by the adaptive duty-cycle clamp. Before application of the transient, the magnetizing current is zero. The transient peak duty cycle at high-line voltage is: VIN min DMAXtr VIN max where VINmin is the low-line input voltage, DMAXtr is the maximum duty cycle at low-line voltage set by the adaptive duty-cycle clamp, and VINmax is the input voltage at high-line voltage. When a transient occurs, the magnetizing current increases from zero to: VIN min DMAXtr L M fS in the rst switch-on cycle after the transient, where LM is the primary magnetizing inductance and fS is the switching frequency. After the switch turns off, the magnetizing current reverses in a sinusoidal fashion set by the magnetizing inductance LM and capacitance CR. Peak voltage on the switch is: V DMAXtr L M VP max TR = VIN max + IN min (Eq. 6) L M fS CR For steady-state operation at full-load and high-line voltage, the peak steady-state voltage on the switch is: VP max S = VIN max + VIN min DMAXs 2 L M fS LM CR (Eq. 7) duty cycle, controlling any buildup of magnetizing energy while alleviating voltage stress.

Design Example
The working power supply of Fig. 3 accepts dc input voltages in the range 36 V to 56 V, and produces an isolated variable output voltage in the range of 4 V to 18 V, controlled by an adjustable external reference. The maximum output current is 0.4 A and the switching frequency is 500 kHz. The resonant-reset forward converter is most suitable for this design because it lets us maximize the duty cycle. That capability is necessary if the output voltage is to be properly controlled from high levels all the way down to 4 V. Otherwise, the PWM controllers minimum on time is a limitation that could introduce problems. Synchronous rectiers should be included to maximize efciency and enable the PWM controller to control the output voltage down to 4 V at light loads. The current-mode PWM controller shown also includes an adaptive volt-sec clamp. Because the power supply must turn on at 36 V and provide full power at 36 V, we set its turn-on point at 34.2 V. That value of turn-on voltage includes a 5% margin to compensate for component tolerances. We then set the maximum duty cycle corresponding to the turn-on point (set by the adaptive duty cycle) at 75%. That leaves 25% of

where DMAXs is the steady-state duty cycle at full load and low line. In practical applications, we try to set DMAXtr slightly higher than DMAXs. We also see that the peak transient reverse voltage on the diode DF is more than twice as high as the peak steady-state reverse voltage with this type of pulsewidth modulated (PWM) controller. For PWM controllers without the volt-sec clamp, the transient voltage can be even higher. If the circuit includes synchronous rectiers, the inductor current does not become discontinuous, and the magnetizing currents at light load and at full load are almost the same. For PWM current-mode controllers with volt-sec clamps, the transient-voltage stress on the primary switch and the secondary diode DF is closer to the peak steady-state voltage stress. The behavior of voltage-mode controllers is similar to that of current-mode PWM controllers. Again, the use of an adaptive volt-sec clamp can reduce stress. These converter types often include a duty-cycle soft-start that ramps up the

www.powerelectronics.com

41

Power Electronics Technology October 2005

FORWARD CONVERTERS

Fig. 4. From Fig. 3, VDS on Q14 at an input of 48 Vdc, with output voltage at 4 V (a) and at 8 V (b).

the switching time available for resetting the transformer at the converters lowest operating voltage. At the lowest operating voltage, the maximum-available reset time for the transformer is: TR = (1 DMA MAX X) fS (Eq. 8)

where D MAX =0.75 and f S =5 10 5 . These values yield a reset time of 0.5 s. To minimize switching loss, the magnetizing current should complete one half-cycle of sinusoidal resonant ringing as given by Eq. 4. Therefore, L M C R = 0.5 10 6 sec , and the peak steady-state voltage stress on the primary switch (obtained by substituting values in Eq. 7) is 217.2 V. Thus, for this design we choose a switch rated at 250 V. Primary-to-secondary turns ratio for the transformer is: n VIN min DMAX MA VOUT (Eq. 9)

We choose a transformer with an EFD15 core of 3F3 material, and obtain n 1.35 by substituting values in Eq. 9. The actual primary turns (30) and secondary turns (24) yield a turns ratio of 1.25. The magnetizing inductance for this transformer, wound using ungapped cores, is 702 H 25%. Tolerance in the magnetizing inductance could produce a tolerance of (+11%)/(13.4%) in the transformers self-resonant frequency, not accounting for tolerance in the total capacitance appearing across the primary in the actual circuit. The measured self-resonant frequency of a sample transformer was lower than 1 MHz. We must guarantee that the actual circuits demagnetizing self-resonant frequency is higher than fS/(1-DMAX). We therefore gap the core, both to reduce the transformers measured self-resonant frequency and to reduce the variation in magnetizing inductance. Using a gapped core with Al tolerance of 10% yields a magnetizing inductance of 144 H. The self-resonant frequency measured for the new transformer sample is 4 MHz, and the transformer capacitance calculated from the expression for self-resonant frequency
Power Electronics Technology October 2005

is 11 pF. Based on the available reset time, the maximum allowable primary capacitance is 176 pF. That value allows a maximum of 165 pF for the sum of switch capacitance and reected diode capacitance (CR). Because MOSFET capacitance is not easily determined, we must build the circuit and adjust the value of added capacitance across the synchronous MOSFET (QR) to get the appropriate reset time. In the actual power supply, the added capacitance across MOSFET QR is 100 pF. The output inductor and capacitor are chosen to optimize efciency and ensure compliance with the output-ripple specication. Thus, the inductor value is 47 H, and CO is formed by connecting three ceramic capacitors in parallel, each rated 4.7 F and 25 V. For the primary MOSFET Q1 (voltage rating of 250 V), we choose an FQD4N25 from Fairchild Semiconductor (South Portland, Maine) for its low inherent capacitance and low on-resistance. This MOSFET also minimizes the gate-drive loss, conduction loss and switching loss. Peak stress on the synchronous rectier QR is: VQR DMA MAX X VIN min 2 n A (1 DMAX MAX ) (Eq. 10)

where nA is the power transformers actual primary-tosecondary turns ratio. In this case, nA is 1.25 and the calculated value of VQR is 122 V. Therefore, we choose a 150-V MOSFET for QR. The peak voltage stress on the freewheeling MOSFET QF is: VQF VIN max nA (Eq. 11)

where nA is 1.25 and VINmax is 56 V. The calculated value is 44.8 V, so for QF we choose a MOSFET rated at 60 V. (The control circuit and synchronous MOSFET drives are shown in Fig. 3 but not discussed further.)

Experimental Results
Figs. 4 and 5 show voltage waveforms on the primary MOSFET of Fig. 3 at different input voltages and various 42
www.powerelectronics.com

FORWARD CONVERTERS

Fig. 5. From Fig. 3, output voltage at 18 V, with VDS on Q14 at an input of 36 Vdc (a) and 56 Vdc (b).

output voltages, with an output load of 400 mA. The drain-voltage waveforms clearly show that the resonantreset voltage does not vary with line voltage, but is proportional to the output voltage. Peak voltage on the primary MOSFET is equal to the input voltage plus the resonant-reset voltage. We conclude that resonant-reset forward converters are quite suitable for power supplies operating from wide-range

dc-voltage inputs. They also are suitable for applications requiring a wide range of adjustable output voltage. In designing resonant-reset forward converters, you should take care to minimize the stress of transient voltages on the devices (the use of synchronous rectication reduces transient-voltage stress on the power semiconductors). For optimum performance, you also should choose an appropriate controller. PETech

Kooler Inductors

Inductors made from Magnetics Kool M E cores run cooler than those made with gapped ferrite cores. Eddy currents, caused by the fringing flux across the discrete air gaps of a gapped ferrite, can lead to excessive heat due to heavy copper losses. The distributed air gaps inherent in Kool M can provide a much cooler inductor. Kool M E cores are available in many industry standard sizes. Magnetics now offers cores in 14 sizes (from 12 mm to 80 mm) and four permeabilities (26, 40, 60, and 90). New sizes are being added. Standard bobbins are also available. If you are using gapped ferrite E cores for inductor applications, see what Kool M E cores can do for you. You may even be able to reduce core size in addition to having a cooler unit. Production quantities are now in stock. For more information, contact Magnetics.

P.O. Box 11422 Pittsburgh, PA 15238-0422 Phone 412.696.1333 Fax 412.696.0333

New Sizes Available!

www.powerelectronics.com

1-800-245-3984

email: magnetics@spang.com www.mag-inc.com

43

Power Electronics Technology October 2005