Home / Blog / Sizing for ride-through

Engineering notes

Sizing a supercapacitor module for ride-through

Oct 2026 · 9 min read

Engineering workbench with a supercapacitor module, calipers and an oscilloscope under cyan task lighting

Energy first, capacitance second

Most ride-through specifications arrive as a request for a capacitance in farads. That is the last number the design needs, not the first. Capacitance only becomes meaningful once two other quantities are fixed: the energy the load needs, and how much of the voltage window you are willing to let it sweep.

Energy is the easy half. Multiply the power you must carry by the time you must carry it. A 120 kW rack holding a 500 ms sag needs 60 kJ. A 5 kW substation protection load holding 4 s needs 20 kJ. Those two numbers are the entire demand side of the problem, and neither one mentions farads at all.

The window is where the engineering starts. A capacitor does not deliver a flat voltage like a battery; it delivers a falling one, and the useful energy is only what comes out between the top of the sweep and the bottom. That is why two designs with identical modules and identical loads can differ by a factor of three in how long they hold.

The voltage window decides how much of the capacitance you can use

The usable relationship is E = ½ C (V²max − V²min). Swept energy scales with the square of both voltages, and the difference between those squares grows fast as you let the floor drop. The table below runs it for a 48.6 V-class module — sixteen 3.8 V LIC cells in series, 166 F class — against five candidate floors.

Bus floorPer-cell floorDroopSwept energy, 166 F module
44 V2.75 V9.5%≈ 35 kJ
42 V2.63 V13.6%≈ 50 kJ
40 V2.50 V17.7%≈ 63 kJ
38 V2.38 V21.8%≈ 76 kJ
34 V2.13 V30.0%≈ 100 kJ

Read the table as a trade, not a menu. Going from a 44 V floor to a 34 V floor nearly triples the energy you can pull from the same hardware — but the deep rows also buy you a lower voltage at the load, a higher current for the same power, and a shorter cycle-life class if the sweep is a routine event rather than a once-a-year bridge. The 500,000+ cycle class our modules are qualified against is quoted on a 3.5 → 2.5 V per-cell window, which is 56 → 40 V across a 16S string. A single deep sag that dips below that is a different duty from a rack that sweeps to 38 V three hundred times a day, and the two should not be specified with the same number.

Hold-up time is a load question, not only an energy question

A capacitor's voltage falls as it discharges, so the current needed to hold constant power rises all the way to the bottom of the sweep. That has two consequences. First, the resistive drop across the string's ESR is worst at exactly the moment you can least afford it — at the end, when the voltage margin is thinnest. Second, the peak current the module must carry is set by the floor voltage, not by the nominal one.

A constant-resistance load is gentler: current falls with voltage, so the discharge curve is a clean exponential and the string never sees its worst-case current. Constant-current loads sit in between. Constant-power loads — which is what a server PSU or a motor drive presents — are the demanding case, and they are the reason the current check has to be run after the energy check rather than assuming the energy answer is final.

The current check is one line: divide the load power by the floor voltage, then compare that against the module's continuous rating and its surge class. A 130 A continuous rating with a surge class above 500 A for events under 5 s covers two very different styles of duty, and a design that only ever tests at nominal voltage will miss which one it bought.

Worked example one: 120 kW rack, 500 ms

This is the case our design guide walks through, and it is worth repeating because it shows how quickly a rack-scale number lands on a module count.

StepValue
Rack transient power120 kW
Ride-through window500 ms = 0.5 s
Energy demand120 kW × 0.5 s = 60 kJ
Swept energy per module (48.6 → 38 V, 166 F)≈ 76 kJ
Usable after derate (depth of discharge, ESR, conversion ≈ 50%)≈ 38 kJ per module
Energy-driven module count60 ÷ 38 ≈ 1.6 → 2 modules in parallel

Two modules is the energy floor, and it is the right answer to the question the table asks. It is not automatically the final BOM. The current check adds modules whenever the transient is long enough or large enough to leave the surge envelope: at rack scale, a full-load step on a 48 V bus is a kiloamp-class event, so the LIC tier is normally built as a parallel fleet, with the battery BBU tier carrying the seconds-to-minutes tail behind it. Run the current line before you freeze the count — and treat the energy table as the lower bound it is.

Worked example two: 12 kW for 2 s

A smaller industrial case makes the same two-step method obvious, because here the energy answer and the current answer disagree outright.

StepValue
Load power and window12 kW for 2 s
Energy demand12 kW × 2 s = 24 kJ
Usable energy per 166 F module (48.6 → 38 V, 50% derate)≈ 38 kJ
Energy-driven count24 ÷ 38 ≈ 0.6 → 1 module
Current at the floor voltage12 000 W ÷ 38 V ≈ 316 A
Against the 130 A continuous class316 A on one module is out of class; three modules hold ≈ 105 A each
Against the >500 A surge class, under 5 s316 A on one module is inside the surge envelope for an isolated event

So the honest answer is conditional. If the 2 s event happens once in the equipment's life and nothing else pulls hard, one module carries it inside the surge class. If it happens twice a shift, the same event is a cycling duty and the continuous rating governs — three modules, each at roughly 105 A, with the extra capacitance also pushing the sweep shallower and the cycle-life class longer. That distinction is the difference between a warranty claim and a design margin, and it costs nothing to write down.

The derating stack between nameplate and delivered energy

Nameplate energy is ½ C V² at full rated voltage. Almost nothing in a real installation runs there. Four multipliers stand between the datasheet and the joules your load actually sees, and each one is a line you can defend in a design review.

StageWhat it accounts forPlanning value
Nameplate½ C V² at rated voltage, one 166 F-class module at 48.6 V≈ 196 kJ
Swept windowEnergy above the floor your load still regulates at (48.6 → 38 V)≈ 76 kJ
Delivery lossesDepth of discharge limits, ESR heating, DC-DC conversionusable ≈ 50% → 38 kJ
End of lifeCapacitance drift and ESR rise across a 15-year design-life classhold 20–30% capacitance margin
Cold startCapacitance falls and ESR rises as temperature drops toward −40 °Csize current from the cold ESR, not the 25 °C value

End-of-life derating is the line buyers most often skip. A module qualified for a 15-year design-life class is not a module that holds its first-day capacitance for fifteen years; it is a module whose aging curve stays inside an agreed envelope. If the load must ride through on year fifteen exactly as it did on day one, the count has to be built on the end-of-life capacitance and the end-of-life ESR — which usually means one more module in the parallel fleet than the beginning-of-life arithmetic suggests. The same logic applies to the cold case: a rack in an unheated hall in a cold climate should have its current check run at −40 °C, where the string's ESR is a multiple of its 25 °C value and the sag at the same current is proportionally larger.

Write the brief as five lines

Sizing conversations stall when the brief is a paragraph of adjectives. Five lines are enough to return a real module count and a cell-level quote, and they map one-to-one onto the arithmetic above.

  • Power: the transient or protection load in kW, and whether it behaves as constant power, constant current or constant resistance.
  • Window: how long it must hold — 1–50 ms transients, 500 ms sags and 4 s protection windows are three different products.
  • Bus: nominal voltage, charger output (V-max) and the lowest voltage the load still regulates at (V-min).
  • Duty: events per day or per year, which decides whether the 3.5 → 2.5 V cycle class or a shallower window applies.
  • Environment: ambient range, coldest start, airflow available, and the service-life class the site has to reach.

Send those five lines and the reply names a module family, a series count and a capacitance class within 48 hours, with the derating assumptions written out so your own review can challenge them. If the numbers land between the standard lines — the 16–108 V band where a lot of industrial ride-through actually lives — the custom and ODM route builds cell-to-cell on the same sorted, laser-welded line, first article in 8–12 weeks.

Check the arithmetic

Send the five-line brief. Get the count back.

Power, window, bus, duty, environment — that is enough for an engineering reply that shows its derating assumptions, within 48 hours.

Get a Quote