Energy Storage, EV Charging, and Bidirectional Converters

Ask for the duration. Power alone says nothing about energy, and energy is what dominates the cost.

The two extremes of what "5 MW" might mean:

• 5 MW / 1.25 MWh (4C) — fifteen minutes of discharge. A frequency-response asset. Sized by power; the cells are chosen for rate capability rather than energy density.
• 5 MW / 20 MWh (0.25C) — four hours. An energy-shifting asset. Sized by energy; the cells are chosen for cost per kWh and cycle life.

Sixteen times the energy for the same power, and since installed cost is dominated by cells, roughly an order of magnitude in price.

What else changes with the C-rate, and is often overlooked:

1. Chemistry. High-rate duty pushes towards LTO or high-power NMC; energy duty towards LFP or high-nickel NMC.
2. Thermal design. Loss goes as \( I^2R \), so a 4C system dissipates sixteen times the heat of a 1C one for the same pack. Cooling may move from air to liquid.
3. Round-trip efficiency. Loss fraction is \( IR_{int}/V \), so it is proportional to rate. The same cells might give 95% at 0.25C and 88% at 2C.
4. Degradation. High rates and deep cycles both age cells faster, and a frequency-response asset may perform thousands of shallow cycles a year against a shifting asset's few hundred deep ones.
5. The converter's duty. Fast reversal and good light-load efficiency for one; peak efficiency at sustained full load for the other.

And the same confusion in the other direction is just as common: a "100 kWh home battery" that can only deliver 5 kW cannot start a heat pump, however much energy it holds. Both numbers, always.

Because the limit is inside the cell, not in the charger. The charger is a voltage source that has run out of voltage difference.

Trace it through the model. The terminal voltage is \( V_{oc}(SoC) + IR_{int} \), and there is a hard ceiling — about 4.2 V per cell for NMC — above which lithium plates onto the anode irreversibly.

• Early in the charge, \( V_{oc} \) is low, so a large \( IR_{int} \) can be added before reaching the ceiling. Full current is available.
• As the cell fills, \( V_{oc} \) rises towards the ceiling and the permitted \( IR_{int} \) shrinks in proportion. The current must fall.
• In the CV phase, \( I = \left(V_{limit}-V_{oc}(SoC)\right)/R_{int} \), and the numerator collapses as the cell fills. The decay is roughly exponential, so the last few percent take disproportionately long.

Two effects make it worse than the simple picture: \( R_{int} \) itself rises towards the top of the range, and lithium diffusion into the electrode becomes the limiting process rather than the external circuit. Pushing harder does not fill the cell faster; it plates lithium.

So a better charger cannot help, because the charger is not the constraint. What can help:

1. Warm the pack. \( R_{int} \) can be three times higher at 0 °C than at 25 °C, so a cold pack reaches the voltage ceiling at a much lower current. Pre-conditioning is worth several minutes and costs only some energy.
2. Higher pack voltage. An 800 V architecture halves the current for the same power, cutting \( I^2R \) heating fourfold — and heat is what forces the taper in the first place.
3. Better cell design — thinner electrodes, better electrolytes — which raises the rate at which lithium can intercalate. This is where the real progress is being made.
4. Stop at 80%. Usually correct, and it also reduces calendar ageing.

And the practical consequence for comparing vehicles: peak charging power is close to meaningless. The useful figure is average power from 10% to 80%, which captures the taper.

Because each switch position already conducts in both directions — the transistor one way, its body diode the other — and the direction of power flow is decided by the sign of the average inductor current, which is a control decision.

The two views of the same circuit:

• From the high-voltage side: a buck. \( V_{low} = DV_{high} \), power flows down.
• From the low-voltage side: a boost. \( V_{high} = V_{low}/(1-D') \), power flows up.

The switches, the inductor and the capacitors are identical. Nothing is added.

The general rule this illustrates: a converter is bidirectional when every switch position can conduct in both directions and block in one. A transistor with an antiparallel diode does exactly that, so replacing every diode with a transistor makes the power stage bidirectional. Since synchronous rectification is now standard for efficiency reasons, most modern converters are bidirectional whether or not it was intended.

What genuinely has to change, and where the engineering effort goes:

1. Continuous conduction through zero. In one direction the converter would normally enter discontinuous conduction at light load. A bidirectional converter must stay continuous, allowing the inductor current to go negative — which changes the small-signal model and therefore the compensator.
2. Deciding what is regulated. Charging may regulate current; discharging may regulate the DC-link voltage. The controller must switch objective without a transient at the handover.
3. Dead time hurts twice. With current flowing either way, the dead-time-induced voltage error changes sign with the current, distorting the transfer characteristic near zero — the same effect that causes zero-crossing distortion in inverters.
4. Protection is symmetric now. Overcurrent, precharge and fault response must all work in both directions, and a fault during reverse flow is not the mirror image of one during forward flow.

The summary worth remembering: the power stage is nearly free; the control is the product.

ZVS is a property of the operating point, not of the topology. It is lost near zero phase shift, and asymmetrically whenever the voltage ratio departs from unity.

The requirement. Before a device turns on, the inductor current must already be flowing in the direction that discharges its output capacitance. From the edge currents:

with \( d = nV_2/V_1 \). At \( d = 1 \) both reduce to \( |\varphi| > 0 \) and ZVS holds everywhere except exactly zero power. Any mismatch opens a dead band around zero, and it hurts bridge 1 when \( d>1 \) and bridge 2 when \( d<1 \).

Now why a charger lives there. Both bad conditions arrive together:

• Low power means small \( \varphi \) — and the CV tail is a long, low-power phase.
• High battery voltage at the end of charge means \( d \) well above unity — and that is exactly when the CV tail happens.
• Low battery voltage at the start means \( d \) well below unity, costing bridge 2 its ZVS during the initial low-power ramp.

So the converter is soft-switched precisely where it was measured for the datasheet — full power, mid-range battery — and hard-switched over a substantial part of a real session.

The remedies, and their prices:

1. Extended (dual or triple) phase-shift modulation — use the bridge duty ratios as extra control freedoms. Restores ZVS over a much wider range; the controller becomes considerably harder and the modulation boundaries must be handled carefully.
2. Choose \( n \) for the middle of the battery range, so \( d \) is near unity for most of the session and mismatched only at the ends.
3. Add a regulating stage in front so the DAB always sees \( d \approx 1 \). Costs an extra conversion.
4. Use a CLLC resonant converter, which retains soft switching more naturally across load — at the cost of frequency control and a harder magnetic design.

And the habit worth forming: when a converter is quoted at 98% efficiency, ask at what power and what voltage ratio, and ask to see the curve. Peak efficiency is a single point on a surface.

Because charging current, not charging voltage, is what everything else is limited by — and doubling the voltage halves it.

What 875 A would require. A cable of that rating is too heavy and too stiff for a person to handle, and needs active liquid cooling along its whole length. Connectors, contactors and fuses at that current are large and expensive. The practical ceiling for an air-cooled cable is around 200–250 A; liquid cooling extends it to about 500 A.

But the stronger argument is the losses, because they go as the square:

• Halving the current cuts \( I^2R \) by four — in the cable, the connector, the contactors, the busbars and the pack's own internal resistance.
• The loss inside the pack appears as heat in the cells, and cell heating is precisely what forces the charging taper. Less heat means the high-current phase can be held longer.
• The same benefit applies to the traction inverter and the motor for the whole life of the vehicle, not only while charging.

What it cost, honestly:

1. New semiconductors. 650 V silicon IGBTs no longer suffice; 1200 V devices are needed, and silicon carbide became the practical choice — initially at a significant price premium.
2. Insulation and creepage requirements rise throughout, affecting motor windings, connectors and the entire high-voltage distribution.
3. Backward compatibility. Most installed fast chargers were 400 V, so an 800 V vehicle needed a boost stage — often implemented cleverly by reusing the traction inverter and motor windings as the boost converter rather than adding hardware.
4. Service and safety training across the whole aftermarket.

And the direction of travel: heavy vehicles are already looking beyond 800 V, because the same argument applies with more force at megawatt charging. The reasoning is not specific to cars — it is the same reason transmission is at 400 kV rather than 4 kV.

Because none of the remaining obstacles are power electronics problems. Both stages of an on-board charger — the active front end and the isolated DC–DC — are inherently bidirectional. Exporting requires a control change, not a hardware change.

What is actually blocking it:

1. Degradation, and who bears it. Grid cycling consumes cycle life the owner paid for. Whether the service payment exceeds the degradation cost depends on tariff, chemistry and cycle depth — and many warranties simply exclude it, which ends the discussion before the economics start.
2. Standards. ISO 15118 defines the communication; CHAdeMO supported bidirectional operation years before CCS did; and every national grid code adds its own protection and certification requirements. A capable vehicle may be prohibited in a given jurisdiction.
3. Metering and settlement. Export needs an export meter, a licence or exemption, and a market that will pay for a resource that can drive away. This is slow institutional work with no technical content.
4. Protection. Everything in Chapter 28's anti-islanding discussion applies, plus the awkward case of deliberately islanding to power a house during an outage — a different mode with different requirements.
5. Availability. A grid operator contracts for a resource that will be present. A car is not guaranteed to be.

Where it already works, because these objections weaken:

• Vehicle-to-home and vehicle-to-load. No export, so no licence and no settlement. Commercially available for several years.
• Fleets. Predictable schedules, one owner bearing both the degradation and the revenue — every objection weakens at once, which is why almost every serious V2G deployment is a bus or van fleet.
• Smart charging. Simply controlling when to charge captures much of the grid value with none of the difficulty, and is being deployed at scale now.

And the general lesson, which is a good one to leave this chapter with: for a mature technology the binding constraint is frequently not technical. Recognising when your converter is no longer the problem — and that the remaining work is standards, contracts and warranties — is part of engineering judgement, not a departure from it.