In almost every electric vehicle on sale, the cells inside the pack are wired in series and joined by fixed pieces of metal, usually copper bars. Pulsetrain takes the metal out and puts a semiconductor in its place: a low-voltage MOSFET on every interconnection, between every pair of cells.
That substitution is the whole company.
A copper bar is a passive component. A MOSFET is not, so the connection between two cells stops being a fixed piece of geometry and becomes something software controls. Any cell can be switched into series, into parallel, or bypassed entirely, in what Niclas Lehnert calls a more or less arbitrary way. The pack stops being a stack and becomes a matrix.
Two things follow immediately. The matrix can synchronise itself to almost any incoming signal, AC or DC, fast charging or slow, which removes the need for an onboard charger. And it can generate a sine wave for the motor straight from the cells, which removes the need for an inverter.
A conventional electric powertrain is three components: the pack with its battery management system, an onboard charger so the car can take AC at home, and an inverter to turn the pack's DC into AC for the motor. A multi-level inverter collapses all three into one. Lehnert, who co-founded Pulsetrain and runs its operational side without a technical background of his own, says the description depends on which direction the listener comes from. It is either a battery management system with inverter functionality included, or an inverter with battery management ability included.
The weight saving is measured against those three parts, excluding the motor, because the motor stays. Combining onboard charger, battery management system and inverter into one electronic component takes up to 40% off their combined mass, which Lehnert puts at roughly 15 kilos on average.
Why this was not buildable ten years ago
The blocker was the parts. A multi-level inverter needs semiconductors that handle very low voltage and quite high current, and that was a requirement nobody was asking for. Semiconductor producers had no interest in developing or selling such a device because there was no demand for it. They could have been made to order, but as the sole user of a component nobody else wanted, the price would have been enormous, and the system would not have been competitive no matter how much better it was on the technical side.
Crypto mining broke that, roughly a decade ago. The graphics cards it ran on needed exactly the class of component Pulsetrain needs, and once the semiconductor makers were holding them in stock off the shelf, the price fell sharply. Lehnert calls crypto a kickstarter for the technology.
The second thing that had to change was where the industry was putting its attention. The last decade of electromobility was spent getting electric cars onto the road at all, using off-the-shelf technology proven to work, rather than optimising the technical side. Lehnert thinks that stage is finished and that OEMs and Tier 1 suppliers are turning back towards the technology.
The third barrier has not gone away. A multi-level converter is a very close connection between hardware and software, and in the Western hemisphere, he says, those are two fields of know-how that rarely come together, particularly where they have to be genuinely connected. Some OEMs still have to build the software expertise. He is direct about locating his own team's advantage in exactly that gap.
The pack is limited by its weakest cell
A conventional battery management system treats the several hundred, sometimes several thousand cells in a pack as a single entity. It has no real access to them individually and no precise information on how they differ, so it manages all of them the same way. Which means it has to protect the worst one.
Lehnert's illustration is deliberately extreme. Take a thousand cells, all but one at 100% capacity, and a single cell at 75%. The pack can only be charged to 75%, because the system has to look after the weak point.
Reality is less dramatic and works the same way. Cells in a pack already differ slightly in state of health and state of charge, through environmental effects or minor production variation, and cells that start out different age at different rates, so the spread widens.
In automotive, end of life is usually taken as 75 to 80% remaining capacity. Current cars on the road reach that point at roughly 10 to 12 years on average, with outliers in both directions, and newer generations do better. The date is set by the weakest cells in the pack.
Per-cell control changes the arithmetic. If the software can choose which cells carry the load at a given moment, weaker cells can be used less often and stronger ones more. Lehnert's example is a car holding a constant speed on the motorway, where there is latitude to pick. Ageing then spreads almost evenly across the whole pack, and the moment it hits 75 to 80% arrives later.
How much later is the open question, and he is careful with it. Simulations in the research sector show lifetime extensions of between 60 and 80%, sometimes beyond 80%. He points to Professor Lienkamp at the Technical University of Munich, to Professor Weyh at the University of the Federal Armed Forces in Munich, the institute Pulsetrain spun out of, and to his co-founder Dr Manuel Kuder, who publishes in the field. Pulsetrain has a first real-life prototype running and is now checking those figures against real data. While the tests are still going he "would not lean too wide out of the window", but the first results point the same way the research does.
The data to run it on barely exists
Every vehicle with a multi-level inverter generates cell-level data continuously while it drives. That data goes up to the cloud at regular intervals, gets evaluated there, and the patterns found in it come back to the individual vehicle as changes to how its pack is operated. Lehnert calls that the real difference from conventional systems: they can generate data, but they can never proactively change the operating state of a specific vehicle, or of a fleet.
The awkward part is that the raw material is scarce, and he concedes it. Cell manufacturers test standard charge and discharge protocols, not this pattern of use, so the datasets for it are almost not available. Pulsetrain has to create its own baseline.
Beyond current, voltage and temperature, which he treats as the standard measurements, the hardware has an intrinsic ability the conventional stack does not. It can put a pulse signal through a cell rather than only DC, something like pumping an AC signal through the battery, and read what comes back. The response is a Nyquist plot, and its shape gives a precise indication of what is going on inside that cell.
Construction machines come before cars
Automotive is the obvious market, huge and visible. It is also, in Lehnert's experience, extremely complex to enter, with extremely high entry barriers and severe cost sensitivity.
That leaves two routes. Go in through the premium and high-performance segment, where the cost of the system matters less, or prove the technology in a non-automotive market first and use the momentum of early systems to drive costs down. Pulsetrain took the second.
Construction was chosen for two reasons. Buyers and operators of construction machinery calculate on total cost of ownership, so the longer a machine can be used the cheaper it becomes for them, and the initial price is correspondingly less decisive. And despite the size of the machines, they run at far lower power levels than a conventional electric car, which makes a new technology easier to implement. The company was at Bauma in Munich the week before recording. Automotive comes in a second step, with better cost performance and proof points that reduce an OEM's fear of an unproven system.
Aviation is a third or fourth option rather than a near-term one, despite fitting the technology well. Redundancy is the reason it fits: a cell failure on the ground means stopping and getting out, and in the air, as he puts it, getting off is maybe a bit challenging. But the certification and development cycles are worse than automotive. Even Airbus, Boeing and MTU face a significant challenge implementing a new technology, he says, even if it is only a screw they are replacing. Pulsetrain would rather "stay on the ground before we go and head for the sky".
Only the software decides which way the power flows
Because direction of flow is a software decision, the system is bidirectional by construction, and not at the three or three and a half kilowatts of a bidirectional onboard charger. It is bidirectional at full motor power. A vehicle with a 100 kilowatt motor has, in theory, 100 kilowatts available to a house or to the grid. Lehnert is aware that most plugs cannot handle that, and neither can the discharge C-rates of the cells, but there is no bidirectional charger in the chain and no sequence of interfaces each adding losses.
Second life gets simpler in the same way. Today a pack has to be removed, opened, emptied of cells, the cells rematched, and a new battery management system fitted on top: a new pack, effectively, built from old parts. It takes time and it is expensive because it takes time, which is a large part of why second life struggles to compete with first life. With a multi-level inverter the pack comes out and the software is updated to tell it that it is now stationary storage. No hardware changes, no rematching.
The same knowledge helps at resale. Someone buying a used electric car cannot really know whether the pack is at 87%, 90% or 97%. A multi-level inverter gives an almost perfect indication of the capacity still in it, and therefore a defensible price.
Lehnert also thinks the technology is a precondition for the packaging trend rather than a casualty of it. Cell to pack, and pack to chassis beyond it, is a non-repairable throwaway concept: a minor failure finishes the pack, and in the extreme case the car. Those designs only deliver their advantages, he argues, on top of a system that can absorb a certain level of failure and keep operating.
The objection he gets is that cells might get good enough to make all of this unnecessary. His answer is that these are chemical systems, extremely complex whenever they are scaled, and that perfect cells are by definition more expensive ones. Once a cell is in a pack it is subject to heat, cold, the driver and temperature differences across the pack, and ageing is driven by all of those variables regardless of how uniform the cells were leaving the factory.
Which brings the argument to fire. Better and more precise data makes the anomalies that precede a thermal runaway easier to identify, and easier to spot earlier than a conventional system would. And when a cell is seen going critical, the multi-level inverter can short circuit it, removing almost a third of the energy in that cell immediately. The chemical energy stored in it is still there, so this is not a cure.
What it produces is time. Enough to tell the driver the pack is in a critical state and they should stop and walk away, and, for someone in an accident who can no longer move, every second added to that timeline is a second in which a rescue unit can reach them.
This piece draws on the full conversation, which is available with a complete transcript on the episode page.