Forward Converter

An isolated buck

If you remember one thing about the forward converter, remember this: it is an isolated buck. Replace the inductor in a buck converter with a transformer and an output inductor, and you have a forward. Same volt-second balance, same continuous output current, same LC output filter — but with a galvanic barrier between input and output.

The contrast with the flyback is sharp. In a flyback, the transformer is really a coupled inductor: it stores the cycle's full energy in an air gap during the switch-on phase, then dumps it through the secondary during the switch-off phase. Primary and secondary current never flow at the same time. In a forward, the transformer is a real transformer — primary current and secondary current flow simultaneously during the on-time, and energy passes through the core rather than being stored in it. The output inductor stores whatever transient energy the LC filter needs to maintain continuous current to the load.

The pay-off is that the peak primary current is set by the output current and turns ratio, not by the cycle's full energy. Lower peak currents mean smaller FETs, smaller transformers, and lower conduction loss. The forward converter dominates the 100 W to 1 kW range where the flyback runs out of breath.

The single-switch forward — minimum part count

                                    L_out
       N_p             N_s     D1 ────UUUU──┬── V_out
   V_in ●───┐         ┌──●───┴───────┐      │
           │         │              │     ═ C_out  ⏚ load
           │         │              │      │
   V_in ●  N_reset   │   D2 (catch) ╲      │
       │      ●──D_r┤         freewheel    │
       │       │   │            │           │
       └─[SW]──┘   └────────────┴───────────┴── GND_sec
        primary    secondary

   SW       = primary-side MOSFET
   N_reset  = reset winding (typically N_p turns)
   D_r      = reset clamp diode
   D1       = output forward rectifier (Schottky for V_out < 60 V)
   D2       = output catch (freewheel) diode
   L_out    = output inductor
   C_out    = output capacitor

Three windings on the transformer core: primary (N_p turns), secondary (N_s turns), and reset (N_reset turns, usually = N_p). Two diodes on the secondary side and one inductor — the output LC filter. The primary side has one switch and one reset clamp diode.

The two phases of every switching cycle

Switch closed (on-time, fraction D, with D ≤ 0.5). The MOSFET ties the primary bottom to ground; V_in is now dropped across the primary. The secondary winding immediately sees V_in × (N_s/N_p), the forward rectifier D1 forward-biases, and current flows from the secondary, through D1, through L_out, into the load. Primary and secondary currents flow at the same time. Inside the transformer, the primary's MMF and the secondary's MMF nearly cancel — the core sees only a small net magnetising current (a few percent of the load current). On the load side, the output inductor sees V_in·(N_s/N_p) − V_out at its terminals; its current ramps up linearly. The output capacitor charges.

Switch open (off-time, fraction 1 − D). The primary FET turns off. The secondary winding's voltage flips polarity, D1 reverse-biases (no more energy crossing the transformer), and the load is now supplied only by the output inductor freewheeling through the catch diode D2 — exactly like a buck converter in its off-phase. Meanwhile on the primary side, the small magnetising current is reset: it flows through the reset winding and the clamp diode D_r back into V_in, ramping the magnetising flux back to zero before the next switch turn-on.

The reset is a hard constraint. If the next on-time starts before the flux has reset, the core walks asymmetrically and eventually saturates. With a 1:1 reset winding, the reset voltage equals the input voltage, so the reset time equals the on-time — capping the maximum duty cycle at 0.5. A 1:2 reset winding allows D up to 0.66 but forces 1.5·V_in onto the FET drain. Most single-switch designs accept the D ≤ 0.5 limit.

The voltage transfer relation

Volt-second balance on the output inductor (it is just a buck-converter inductor with V_in replaced by V_in · N_s/N_p):

on-time:   V_L = V_in · (N_s/N_p) − V_out
off-time:  V_L = −V_out

V_L_avg = 0  ⇒  V_out = V_in · (N_s/N_p) · D

So the output voltage is the input scaled by the turns ratio and the duty cycle. A 400 V DC bus into a 4:1 step-down transformer at D = 0.45 gives V_out = 400 · 0.25 · 0.45 = 45 V — a typical telecom rectifier output. The turns ratio sets a fixed step; the duty cycle is the regulation knob.

Worked example: 400 V bus → 12 V / 30 A server PSU

Design a 360 W server PSU stage: V_in = 400 V (from PFC pre-regulator), V_out = 12 V at 30 A, f_sw = 200 kHz, target D_nom = 0.4, η_target = 90 %.

Turns ratio. Pick N_p/N_s so V_out = V_in · (N_s/N_p) · D gives 12 V at D = 0.4:

N_p / N_s = (V_in · D) / V_out = (400 · 0.4) / 12 = 13.3

Round to N_p:N_s = 26:2 = 13:1.
Use N_p = 26 turns, N_s = 2 turns on an EER42 ferrite core. N_reset = 26 turns.

Output inductor. The inductor sees V_in · N_s/N_p − V_out = 400/13 − 12 = 18.8 V during the on-time. For ΔI_L = 30 % of I_out = 9 A peak-to-peak ripple:

L_out = (V_in · N_s/N_p − V_out) · D / (ΔI_L · f_sw)
      = 18.8 · 0.4 / (9 · 200 000)
      = 4.2 µH

Choose 4.7 µH commercial inductor, I_sat ≥ 40 A (I_pk = 30 + 4.5 = 34.5 A).

Output capacitor. For 50 mV peak-to-peak ripple at full load:

C_out = ΔI_L / (8 · ΔV · f_sw) = 9 / (8 · 0.05 · 200 000) = 113 µF
Pick 2× 100 µF polymer caps in parallel (ESR < 5 mΩ each).

Primary current. Peak primary current = I_out · N_s/N_p + magnetising = 30/13 + 0.2 ≈ 2.5 A. RMS primary current ≈ 1.6 A.

FET selection. 800 V breakdown (margin over 800 V — twice V_in with reset overshoot), R_DS(on) ~ 100 mΩ, Q_g ~ 50 nC. SiC MOSFETs (1200 V class) are increasingly common to reduce switching loss.

Diode selection. D1 sees reverse voltage = V_in · N_s/N_p = 31 V plus margin; a 60 V Schottky (or synchronous FET) at 40 A rating works. D2 sees similar reverse voltage and freewheels at I_out = 30 A.

Efficiency. Combined switch + diodes + magnetics + filter losses total around 40 W on 400 W input — η ≈ 90 %. Replace the secondary diodes with synchronous FETs and η climbs to 93 %.

Forward vs flyback head-to-head

Property	Flyback	Forward
Power range	milliwatts – ~150 W	100 W – several kW
Transformer role	Energy storage (air gap)	Real-time transfer (no storage)
Air gap	0.1–0.5 mm — required	None — undesired
Output filter	Cap only — pulsed delivery	L + C — continuous delivery
Peak primary current	High — stores full cycle energy	Low — equals I_out · N_s/N_p
Output ripple	Higher — capacitor handles pulses	Lower — inductor smooths
Reset mechanism	Energy dumped to load each cycle	Reset winding or active clamp
Duty cycle range	10 – 80 %	≤ 50 % (single-switch); ≤ 70 % (active clamp)
Efficiency	75 – 88 %	80 – 93 %
Part count	~5 (cheapest)	~8 (slightly more)
Multiple outputs	Easy (extra secondary winding)	Possible but each needs its own LC filter

The forward wins on every metric that matters above ~100 W: efficiency, ripple, peak current, transformer size. The flyback wins on cost, parts count, and ease of multi-output below 100 W. The transition zone is roughly 75–150 W; engineering judgment, target margin, and EMI requirements decide which side a particular product lands on.

Single-switch, two-switch, active-clamp, half-bridge

• Single-switch forward with reset winding. The textbook design. One FET, one reset winding, D ≤ 0.5. Dominant in 100–300 W telecom and industrial supplies. FET drain stress: 2·V_in.
• Two-switch forward. Two FETs in series with the primary (one high-side, one low-side, gated together). Two clamp diodes resonate the magnetising current back to V_in during the off-time — no reset winding needed. FET drain stress: exactly V_in (no overshoot). Dominant in 300 W – 1 kW industrial PSUs.
• Active-clamp forward. Replaces the reset winding with an active-clamp circuit (a clamp FET + capacitor) that recycles the magnetising and leakage energy back to V_in. Pushes max duty to 0.7, enables ZVS turn-on, raises efficiency by 3–5 percentage points. Used in 100–500 W server and telecom supplies; the underlying topology of many GaN-based modern designs.
• Half-bridge / full-bridge forward. Two- or four-FET inverter on the primary applies bipolar AC across the transformer (instead of unipolar pulses). The core operates in both magnetic quadrants, doubling its energy throughput per cycle. Dominant above 1 kW. Common in 1.5 kW–10 kW server and telecom rectifiers.
• Push-pull forward. Centre-tapped primary with two FETs alternately driving the two halves. Symmetrical core utilisation like a half-bridge but at lower part count. Common in 500 W – 2 kW DC-DC converters in EV onboard chargers and industrial DC backup systems.
• Forward with current-doubler rectifier. Two output inductors and two rectifiers instead of one inductor with two diodes — halves the inductor current rating, reduces conduction loss in high-current outputs (typical: 12 V at 60 A+ server rails).

Where the 10–20 % efficiency loss comes from

• Switching loss in the primary FET. The FET turns on into a non-zero V_DS and turns off carrying primary current. Total switching loss scales as ½ · V_in · I_p · t_trans · f_sw. SiC FETs cut transition times to a few nanoseconds, reducing this loss by 60–80 % vs Si.
• Conduction in primary FET. I_p_RMS² × R_DS(on). For 800 V Si FETs in a 400 V design, R_DS(on) is typically 100–300 mΩ; for 1200 V SiC, 30–100 mΩ at higher cost.
• Transformer copper and core. The transformer carries real load current (unlike a flyback's pulsed primary current), so copper losses are significant. Litz wire is mandatory above ~200 kHz to control proximity effect and AC resistance.
• Output diode forward voltage. Each Schottky drops 0.4–0.7 V at full current; at 30 A output, that's 12–20 W per diode. Synchronous rectification (replacing diodes with low-R_DS(on) FETs) is mandatory for high-current outputs.
• Output inductor. DCR × I² and core loss. Same physics as a buck-converter inductor; same engineering trade-offs.
• Reset and clamp losses. The reset winding (or active clamp) circulates magnetising current back to V_in but loses a small fraction to diode forward drop and clamp resistance. Active-clamp implementations recover most of it but add 0.5–1 W of clamp-circuit dissipation.

Where forward converters show up

• Server power supplies (1 + 2 stage architectures). Below 1 kW, server PSUs typically use a two-switch forward as the DC-DC stage after PFC. The flagship 80-Plus Gold and Titanium PSUs in Dell PowerEdge, HPE ProLiant, and Supermicro servers up to about 800 W use this topology. Above 1 kW LLC has taken over, but two-switch forwards still ship in massive volume.
• Telecom 48 V rectifiers. Two-switch and half-bridge forwards converting 400 V PFC bus to 48 V at 1–3 kW. Eltek, Delta, and Vertiv all ship telecom rectifier modules built around forward variants.
• Industrial DIN-rail PSUs above 100 W. Phoenix Contact Quint 4, TDK-Lambda DRB, Mean Well DR series — when they spec output power above 100 W they almost universally implement forward topologies internally.
• EV onboard charger DC-DC. The 12 V auxiliary DC-DC inside a Tesla, BMW i3, or Rivian (converting 400 V or 800 V traction battery to 12 V at 1–3 kW) is typically a half-bridge forward, sometimes with current-doubler output. Continuous output current and high efficiency are both critical.
• Telecom auxiliary supplies. Smaller (100–300 W) auxiliary supplies inside larger telecom equipment, generating intermediate bus voltages for downstream point-of-load converters.
• Welder and induction-heating power stages. Mid-power industrial welders use half-bridge forwards for their primary-to-secondary isolation while delivering hundreds of amps to the work piece at low voltage.

Common design pitfalls

• Flux walking. If the reset is incomplete — too little reset voltage, too short an off-time, or asymmetric switching — the flux walks each cycle and eventually saturates the core. Symptom: primary current ramp climbs cycle by cycle, then explodes when the core saturates. Cure: oversize the reset winding margin, never push D above 0.45 in a single-switch design, current-limit the FET to catch saturation early.
• Leakage spike on FET drain. Leakage inductance between primary and secondary stores energy that has nowhere to go when the FET opens, ringing the drain voltage up. Mitigate with an RC snubber on the drain or — in active-clamp designs — recover the energy through the clamp FET.
• Reset diode reverse recovery. The reset clamp diode carries magnetising current during the off-time; at turn-on it reverse-recovers and dumps energy into the FET. Use ultrafast or SiC Schottky diodes for the reset clamp on high-frequency designs.
• Output inductor saturation. Same as a buck converter — check inductor I_sat at maximum peak current including transient load steps. Underrated cores saturate, dI/dt explodes, and the output diodes pop.
• Secondary diode reverse recovery (D2 catch). When D1 turns on at the start of the on-time, D2 must turn off. Standard silicon diodes take 50+ ns to recover; during that time both diodes conduct, shorting the secondary and dumping energy into the primary FET as a current spike. Use Schottky for V_out < 60 V or SiC Schottky for higher rails, or synchronous rectifiers with adaptive timing.
• Slow soft-start. Forward converters need a slow duty-cycle ramp at startup to avoid overcharging the output capacitor (which behaves as a near-short at t = 0). A 1–5 ms soft-start ramp is standard; many controller ICs include this on-chip.
• Skipping the secondary rectifier reverse-voltage budget. The forward diode D1 must block V_in · N_s/N_p plus any reset overshoot. In a 1:13 design with V_in = 400 V, D1 sees up to 80 V reverse during reset — easy to forget and pop the diode in the first prototype.

Historical context

The forward converter emerged in the mid-1970s alongside the flyback as the two foundational isolated-DC-DC topologies. Where the flyback won the cheap-and-small market segment (early television flyback transformers and the first generation of switching wall warts), the forward converter took the high-power industrial and aerospace segment. Mil-spec textbooks from the late 1970s (such as Pressman's "Switching and Linear Power Supply, Power Converter Design") treat the single-switch forward as a default starting point for any isolated DC-DC above 50 W.

The two-switch forward variant was popularised in the 1980s in telecom-grade rectifiers, eliminating the reset winding while halving the FET voltage stress. Active-clamp forwards entered the mainstream in the 1990s with the publication of Vlatković et al.'s seminal active-clamp papers, enabling soft switching and pushing efficiency above 90 % in compact server PSUs. The 2010s saw GaN-based forward converters demonstrate 96 % efficiency at 1 kW in research benches; commercial GaN forwards now ship in the highest-density telecom and server PSUs from Delta and Eltek.