╳ThinkPower
Constraints → topology recommendation
NGSpice ↗

Compare converter candidates for your design.

Choose a conversion ratio and compare coupled-inductor cells at their nominal ripple-cancellation points.

2 · DESIGN CONSTRAINTS

Your available resources

Ron × area defaults to an illustrative 1 mΩ·mm², not a characterized TSMC layout value. Capacitor budget covers flying/clamp caps; winding limits count power windings. Driver area and Cin/Cout are outside these budgets. Lk = 50 nH and Lm = 1 μH are editable illustrative defaults, not characterized TSMC magnetics.

3 · COMPARE ALL QUALIFYING CANDIDATES

Topology candidates

Ripple-cancellation criterion and recommendation scope

Output voltage and output current are the operating-point inputs. The selected conversion ratio N:1 sets the nominal required input voltage Vin = N Vout. Output power and load resistance are calculated: Pout = Vout Iout and Rload = Vout/Iout. Converter Rout is the modeled output-referred series resistance (RFET + RCAP + RIND + the non-double-counted charging increment), not the load resistance. Each panel shows Rout in mΩ as well as its per-unit value. Rout/Rload gives the nominal constant-current droop fraction when requested Vout = Vin/ratio. The specified load current is used in the loss model; this is not a solved resistive-load equilibrium. Only coupled-inductor cells with a verified nominal two-level excitation pattern at this ratio are listed. The admitted patterns are matched complementary winding inputs at 50% effective duty, or the three-level buck’s ideal constant-midpoint state at 2:1. A two-phase buck reaches complementary excitation at ideal 2:1; balanced half-input hybrid cells reach it at ideal 4:1. A three-level buck also belongs to 2:1: with a balanced flying capacitor and 50% gate duty, its two redundant states both apply Vin/2 to the winding input, ideally giving zero winding voltage and ripple. This midpoint result does not require magnetic coupling, although the drawn implementation retains coupled windings. It differs from its coupling-assisted 4:1 operation. Real flying-capacitor voltage ripple, imbalance and dead time leave residual ripple that is not quantified by the midpoint-hold estimate. A four-pulse or three-level sequence must preserve the complementary winding-input pattern; arbitrary pulse permutations are not admitted by this estimate.

Lself = Lk + Lm; M = −Lm; LDM = Lk + 2Lm
Δiphase,pp ≈ Vstep / (4 feff LDM)

Vstep is Vin for buck cells and Vin/2 for the admitted hybrid cells. feff is fs for the two-command cells, and 2fs for the specified four-pulse / three-level excitation. Phase ripple percentage is Δiphase,pp divided by Iout/2. The symmetric common-mode switching excitation cancels ideally; individual phase ripple is finite and approaches zero only with sufficiently large differential-mode inductance. This is not a measured zero-ripple claim.

We screen the entered phase-ripple threshold, winding/capacitor counts, capacitance budget and estimated open-loop droop. Total FET width is searched under the silicon budget with each model’s relative allocation held fixed. Die-area mode uses RΣ = (Ron × area)/(usable transistor area); the 1 mΩ·mm² default is an illustrative editable layout assumption. Shortlist resistances share the full-budget RS base. Loss ranking remains the flat-current model plus gate drive; ripple-loss corrections and Coss energy are not inferred.

Ripple estimates assume equal winding turns, inverse polarity, balanced capacitor voltages, stiff ideal switching levels and matched timing. Capacitor ripple, imbalance, dead time, resistive voltage errors and control action can spoil cancellation. Vin is derived from the selected ideal cancellation ratio and requested output voltage. Changing ratio preserves both output voltage and current. Open-loop output under the specified current is predicted as Vin/ratio − Iout Rout, so matching the ideal target does not prove loaded regulation. Correcting loaded output voltage by duty adjustment can move away from the nominal cancellation point. Saturation, core loss and magnetic implementation need separate verification.

The current library establishes only the 2:1 and 4:1 cancellation-point families. The 6:1, 8:1 and 16:1 tabs remain empty until a higher-order switching and magnetic model is available. Off-duty buck comparisons and the nonmagnetic SC option are omitted. The stacked buck is grouped at its electrical 2:1 cancellation point while retaining its existing 4:1 loss-sizing allocation.

TI describes the three-level buck’s ideal D = 0.5 zero-inductor-voltage point in SLYT807. For coupled-inductor modal behavior, see ADI’s phase- and output-ripple analysis. The formulas above follow directly from the stated symmetric inductance matrix.

Inspect schematics, die allocation, and loss derivations

Reference drawings use normalized resistances. Inspect a shortlisted design to load its loss weights.

Per-unit loss weights
180 nm device reference · uses the specified output voltage and current

User-supplied device example from XFLY_Power_Analysis.md; not independently verified foundry data. FOMg = 1.51102 Ω·pC; FOMoss = 0.67424 Ω·pC. First-order width scaling; same device technology for all FETs. Device voltage rating and PVT limits require a PDK check.

Optional Eoss entry: dissipated Coss energy in pJ per 1 mm of total FET width per cycle, averaged over the actual device transitions after recovery. Leave blank when unknown. Qoss is not converted to Eoss.

Derivation and model scope

RΣ is the resistance of all available FET width in parallel. With the existing twelve-unit area budget, RΣ = RS/12. A device assigned fraction α has width αWtot and on-resistance RΣ/α. FOMg = Ron·Qg and FOMoss = Ron·Qoss are width independent under the supplied scaling.

Pg = VGS fs ng FOMg / RΣ; Rg = Vx² RΣ / (VGS fs ng FOMg)

Vx = M Vin is the ideal secondary voltage, ahead of the series loss resistances. Both loss shunts use Vx, rather than the loaded Vout. Gate power is a separate supply demand unless the driver is explicitly powered from the main input. The shunts are accounting equivalents; their current does not enter the conduction-loss calculations.

Poss = fs Σ νi Eoss,diss,i; Ross = Vx² / Poss

ng is the width-weighted number of gate-charge events per period, assumed uniform in this example. Actual four-pulse sequences require event counts for each physical device. No exact Eoss follows from FOMoss alone: Q(V), Eoss(V), switching swings, and recovery are needed. The optional aggregate Eoss input must already include these effects.

The numerical table uses matched flat winding currents for inductive converters. The added SP-SC uses an exact linear periodic RC model, with no power winding. Width allocations remain fixed at each topology’s reference ratio (2:1 for the added cells; 4:1 for the paper cells). These allocations are not automatically reoptimized at another sizing duty. The 4 μF budget covers flying and clamp capacitors only; Cin and Cout are excluded. Adopted allocation: equal main capacitors for two-capacitor cells; equal main and clamp capacitances in four-capacitor cells, consistent with the existing 2SBUCK model. This is a disclosed comparison assumption, not a capacitor-allocation optimum. Additional XFLY4P clamp recharge is not derived. Turn-on/off overlap, dead time, core loss, interconnects and ripple corrections remain outside these estimates.

RFET, RCAP and RIND use the existing closed-form duty expressions below. The charging increment uses the paper's asymptotic interpolation for the paper cells. The added SP-SC has its own exact linear periodic RC model. Unknown Coss loss is excluded from the displayed partial loss, so no total efficiency is claimed.

Finite-RC charge-event check

For a single linear RC charging event lasting d/fs, let q = a Iout/fs be the required transferred charge. Integrating i²R for an exponential current and eliminating its initial amplitude gives:

Revent = a²/[2 C fs] coth[d/(2 fs Rpath C)]
RSSL = a²/(2 C fs); RFSL = a² Rpath/d

Here C is the effective capacitance seen by that event. The exact result approaches both limits; they must not be added to another RMS model of the same event. It assumes a fixed linear path, settled initial conditions specified by the event, and a driving step chosen to transfer q. Each converter needs its own event capacitance, duration, charge multiplier, and periodic initial conditions before this expression can replace the existing interpolation.

Loading topology models…
CONVERTER TOPOLOGYAVERAGED DC EQUIVALENT CIRCUIT
Effective resistance versus conversion ratio

The horizontal axis is the step-down ratio M = Vout/Vin. Only supported duty ranges are plotted.

How the effective resistance is calculated

The ideal DC transformer represents the averaged conversion ratio; it does not imply a physical transformer or galvanic isolation. The series resistances are referred to the output. The three green blocks show the coefficient-weighted switch, capacitor-ESR and winding contributions, grouped as RFSL.

Vout ≈ M Vin − Iout Req
Req = RIND + √[(RFET + RCAP)² + RSSL²]
RFSL = RFET + RCAP + RIND
RSSL,add = √[(RFET + RCAP)² + RSSL²] − (RFET + RCAP)
Req = RFSL + RSSL,add

The orange inset gives the estimated charge-sharing limit RSSL. The orange series block is its incremental contribution after interpolation. FSL and SSL are limits of the same charging process. Paper cells use the paper’s interpolation; the added SP-SC uses its exact linear periodic RC expression below. The limits are not added as independent series resistors. All displayed numbers divide resistance by RS. Device allocations are fixed at their reference points: 2:1 for BUCK2Φ and SP-SC, 4:1 for the paper cells; all use total area 12 A₀.

Assumptions: two matched flat phase currents, identical device technology, and zero clamp-capacitor ESR. The separate gate-drive and Coss accounting shunts sit at the ideal secondary node Vx = M Vin, ahead of the series losses: Rg = Vx²/Pg and Ross = Vx²/Poss. Their physical values are in the separate technology example; they are excluded from the per-unit resistance sweep. Gate demand is tracked separately from the main input unless its supply is specified. Gate drive, magnetic core loss, winding ripple, and dead time are excluded. XFLY2P uses the paper’s effective six-switch model with the physical split-middle area normalization. XFLY4P applies only to the specified four-pulse realization. Unsupported operating points receive no numerical resistance.

2:1 family: circuit operation and derivation

BUCK2Φ has two synchronous half bridges, with 180° phase displacement and two inverse-coupled windings. Db = M. Each winding carries Iout/2 in the flat-current approximation. At the 2:1 sizing point, all four switches receive 3 A₀: Ron = RS/3. Summing both upper/lower duty-weighted losses gives RFET = RS/6 at any duty with those fixed widths, plus RIND = RL/2. This assumes CCM and neglects triangular ripple. The opposite winding dots specify inverse coupling. Coupling changes ripple and modal inductance; without self/mutual inductance and core-loss data, no coupling-dependent loss reduction is assigned. RL remains the resistance of each winding, so the flat-current winding term is RL/2.

SP-SC has two interleaved four-switch cells. In each phase, S1/S3 (or Q1/Q3) connect CF between Vin and Vout; S2/S4 connect it between Vout and ground. A and B are 50% clocks shifted 180°. The two equal flying capacitors share the assigned budget; there is no power inductor. Ideal conversion is fixed at 1/2. The exported comparison index D = 2M is not its gate duty; SP-SC gate duty is 0.5. The eight equal switches each receive 1.5 A₀, giving Ron = 2RS/3.

RFET = 2RS/3; RCAP = RC/2; RFSL = 2RS/3 + RC/2; RSSL = 1/(8 CF fs)
Req = RSSL coth(RSSL/RFSL); RHC,add = Req − RFSL

For one phase, each half-cycle transfers q = Iout/(4fs). Integrating the exponential i²R loss through Rpath = 2Ron + RC over both half-cycles, and summing both phases, gives the expression above. It assumes matched cells, constant linear Ron/ESR, stiff input/output ports and no dead time. RFET/RCAP are FSL components; RHC,add is the finite-RC increment, not a second RMS loss. The finite-RC result tends to RFSL at high frequency and RSSL at low frequency. Only its fixed 2:1 point appears in the conversion-ratio sweep; voltage droop does not create another ideal conversion ratio.

The single-cell coth relation is also documented in Berkeley EECS-2021-236, Eq. (3.13). The two-phase reduction and normalized coefficients shown here are derived for these drawn cells.

Upload a topology

Vision connection

Your inference token and image go directly from this browser to the Hugging Face inference provider for this request. Neither is saved.

Choose a schematic to begin.