A rooftop solar array has maybe six parts a homeowner can name. Panels, racking, an inverter, a meter, some conduit, and, if the installer had a good day, a battery. The real bill of materials runs to a few thousand lines.

Most of those lines are unremarkable. Screws, gaskets, ferrite beads, resistors costing a fraction of a cent. But a fair number of them carry weight that has nothing to do with mechanical load. If they get two percent worse, the system gets noticeably worse. And if they had stayed as expensive and as bulky as they were twenty years ago, a good share of today’s clean-energy hardware would still be commercial-scale only, because the enclosure alone would have priced it out of a garage.
An inverter spends most of its life measuring, not converting
Strip one down and the power stage looks almost plain. Transistors, an inductor, a capacitor bank, a heatsink the size of a paperback. The difficulty lives in the loop wrapped around all of that.
A grid-tied inverter has to match the utility’s voltage and phase, track the array’s maximum power point as clouds pass, back off when the grid sags, and disconnect within a couple of seconds if the grid goes away entirely so it doesn’t energize a line a repair crew is working on. Every one of those behaviors is a control decision, and every control decision needs a current reading first. Tens of thousands of readings per second, from a hot box bolted to a south-facing wall, staying accurate for a decade.
That requirement is what quietly shapes the whole board.
The coil the size of a shirt button
The oldest trick in current sensing is still one of the best. Run the conductor through a ring of magnetic material, wind several hundred turns of thin wire around the ring, and the alternating field induces a proportional, much smaller current in the winding. A small resistor across it, called the burden, turns that into a voltage a microcontroller can read. Ratios of 1000:1 are typical, and isolation comes for free, since nothing connects the 240-volt side to the 3.3-volt side except a magnetic field.
What changed is the packaging. These used to be ring-shaped modules bolted to a panel and wired in by hand. Open a distributor catalog today, filter for a current transformer PCB mount part, and there are hundreds of variants: through-hole toroids a centimeter across, low-profile surface-mount packages, and designs where the primary is a copper trace on the board rather than a separate wire, though that last approach stays confined to particular current and isolation ranges. Assembly that once needed a technician, two wires and a terminal block now happens in a pick-and-place machine.
Two things surprise people who haven’t designed with one. Accuracy isn’t a single number: a metering coil has both a ratio error and a phase displacement, and the second matters more than it sounds, because real power depends on the angle between voltage and current. A few tenths of a degree becomes a percentage error in the reading, and it gets worse at the low end of the range. A house drawing forty watts at three in the morning is exactly where a cheap sensor is least trustworthy.
The other is saturation. Push DC through the primary and the core stops behaving linearly, which is one reason a utility meter and a hobby energy monitor can disagree in a house full of cheap half-wave rectified loads.
Shunt, coil, Hall, or Rogowski: what the choice actually costs
Four approaches dominate, and the trade-offs between them are the reason no single one has won.
| Method | How it senses | Main compromise | Where it usually lands |
| Shunt resistor | Voltage drop across a precise low-value resistor | No inherent isolation; self-heating shifts the reading; needs an isolated amplifier | Battery management, DC side of chargers, cost-driven single-phase meters |
| Current transformer | Magnetic coupling into a wound core | AC only; accuracy falls off at very low current; core can saturate on DC | AC revenue metering, inverter output stages, clamp-on retrofit monitoring |
| Hall-effect or fluxgate | Semiconductor or coil sensing field in a gapped core | Needs its own supply and calibration; Hall parts carry offset and drift specifications that compensation reduces rather than eliminates; fluxgate is far more precise and far more expensive | DC string sensing, EV charger leakage detection, DC fast charging |
| Rogowski coil | Air-cored winding, output proportional to rate of change | Signal needs integration; weaker at low currents | Retrofit metering on large busbars, switchgear, high-current testing |
In practice this decision gets made in the first week of a project and then almost never revisited, because it drives the board layout, the isolation strategy, the enclosure size and roughly half the certification paperwork.
Six milliamps, and why an EV charger costs more than it looks like it should

A home charger looks like it should be a relay and a bit of signaling. Physically, that is close to true. The cost is in the safety envelope.
Here is the wrinkle. The rectifier that lives inside the vehicle means a fault on the car side can push smooth DC current into the earth path. Conventional residual current protection is built around a differential transformer, and smooth DC saturates that transformer’s core, which means the DC leakage not only goes undetected but can blind the device to the AC leakage it was installed to catch. A protective device that fails silently is worse than no device at all.
So the standards force the issue. Installations in IEC-based markets are built either around a Type B device or around a Type A breaker paired with dedicated DC residual current detection at roughly 6 mA; North American equipment works to a 20 mA DC threshold. Either way, a plain current transformer cannot do the job. What goes on the board is a fluxgate or comparable sensor with its own excitation circuit, its own compensation, and a price to match. That component, sitting in a plastic box on a garage wall, is a meaningful fraction of what separates a certified charger from an extension cord with ambitions.
Silicon carbide gets the press release, the gate driver gets none
A SiC MOSFET switches far faster than the IGBT it replaced, and that speed is the entire point. Faster edges mean less energy burned per switching event, higher frequency, smaller magnetics, a smaller box. They also mean the voltage at the switching node can move at tens of volts per nanosecond, a couple of centimeters away from control electronics running at five.
The driver is what has to survive that, and it is asked to do three awkward things at once:
- deliver and remove a substantial gate charge in tens of nanoseconds, usually with a negative off-state bias so the device doesn’t partly switch itself back on through its own parasitic capacitance;
- carry commands across an isolation barrier while that barrier is being slewed at enormous rates, which is why digital isolators are specified for common-mode transient immunity in kilovolts per microsecond rather than by insulation voltage alone;
- detect a short circuit and shut the device down gently enough that the collapsing current doesn’t produce a destructive voltage spike, all within a few microseconds, because wide-bandgap devices have much less thermal buffer than silicon.
Get the transistor right and the driver wrong, and the result is not a product that performs poorly. It is a product that fails on the bench and a team that goes back to slower silicon.
Why a 2010 inverter weighs three times what its replacement does

Most of that weight was magnetics, and most of the reduction came from materials rather than from cleverness in the schematic.
Energy stored per switching cycle falls as frequency rises, so a converter that switches faster needs less core and less copper for the same power. That is the mechanism. What made the higher frequencies survivable was a generation of core materials with lower losses in the relevant range: manganese-zinc ferrites for the high-frequency transformers, powdered iron and sendust for inductors that need to saturate gracefully rather than abruptly, and nanocrystalline alloys where very high permeability matters.
That last family shows up in two unrelated places. One is the common-mode choke, which exists because fast switching pushes noise through the parasitic capacitance between a PV array and the earth beneath it, and a system that doesn’t manage that noise does not pass EMC testing. The other is the metering core discussed earlier, where the same high permeability is what lets a small ring stay accurate at low current.
Boring material science, in other words, is a large part of why the box got small.
The board is a component, not just what components sit on
At 1000 volts DC a printed circuit board stops being neutral. It becomes an insulator with a rating, and that rating depends on creepage across its surface, clearance through the air, and the tracking index of the laminate, meaning how readily the material forms a conductive carbon path once it is dirty and electrically stressed. Designers buy the distance they need with milled slots between high-voltage nets, with conformal coating, sometimes with a physically larger board than the circuit requires. None of it is visible in the finished product. All of it separates a unit that ages quietly in coastal humidity from one that develops a tracking fault in year four.
The half-second before a battery connects

There is a moment at the start of every battery system’s day that receives more engineering attention than the rest of the cycle put together.
Close a main contactor directly onto a discharged DC-link capacitor and the capacitor looks, briefly, like a short circuit. The resulting inrush is limited only by wiring resistance, it can reach hundreds of amps, and its usual consequence is a pair of contacts welded shut. A welded contactor in a battery system means the pack cannot be disconnected, which turns a maintenance job into a hazard.
The fix is a resistor and a smaller relay. Precharge closes first, the capacitor fills over a few hundred milliseconds, the main contactor closes into a voltage it can handle, and the precharge path opens again. Cheap parts, unglamorous sequence, and the reason the expensive parts survive.
DC contactors themselves deserve a mention here too. Alternating current gives a switch a free opportunity to extinguish an arc a hundred times a second when the current passes through zero; direct current offers no such thing, so an arc will happily sustain itself across opening contacts. Sealed, gas-filled chambers and magnetic blowout arrangements exist to solve a problem that simply doesn’t arise on the AC side.
Ask a maintenance technician what fails first
Nobody says the panels. Modules are the most conservatively engineered part of the system, warranted for decades and degrading slowly enough that the number is measured in fractions of a percent per year.
The answers you get are fans, contactors, and capacitors, roughly in that order of annoyance. The DC-link capacitor bank is the interesting one. It absorbs the ripple current from the switching stage, it lives inside a hot enclosure, and electrolytic types age in a way that follows temperature closely, with a rule of thumb among designers that useful life roughly halves for every ten degrees Celsius of extra core temperature. That single relationship explains a startling number of design decisions: the shift toward metallized polypropylene film capacitors on the DC link, the obsession with derating, and the installer’s insistence that the inverter not be mounted where it gets afternoon sun.
The five-dollar connector that keeps appearing in fire reports
PV connectors of the MC4 style all look alike. They are not. Different manufacturers use different contact geometries, plating and tolerances, and a plug from one brand mated to a socket from another can feel perfectly solid while making poor contact. Resistance means heat, heat accelerates oxidation, oxidation raises resistance, and the loop runs away over years on a roof nobody inspects. Investigators looking into array fires return to cross-mated connectors often enough that electrical codes now carry explicit language about intermatability, and careful installers treat brand mixing as a hard rule rather than a preference.
Where the price actually came down
Module prices get the headlines. Panels really did become dramatically cheaper, and that story is well told.
Less discussed is what happened to everything behind the panel. A metering function that once required a wired-in transformer, a separate analog conditioning board and a calibration procedure now consists of a board-mounted coil, a metering IC and a firmware routine. That change removes wiring, removes connectors, removes assembly labor, shrinks the enclosure, cuts the copper, and lowers the shipping weight. No single step is dramatic. Each one is a few percent.
Compounded across fifteen years and several hundred component categories, those few percent are the difference between a product that needs an electrician for two hours and one that clips onto a DIN rail in ten minutes.
Distributed control followed from several of these curves at once rather than from any single one. Microcontrollers with enough compute for real-time metering became commodity items. Wireless communication stopped requiring a dedicated module and a specialist. Power semiconductors improved. Sensing got smaller and, at ordinary accuracy grades, considerably cheaper, though high-accuracy measurement remains a real line item and revenue-grade metering has never been close to free. Take away any one of those and the smart meter, the load-balancing charger and the utility-dispatchable inverter all become harder to justify commercially.
None of these parts is going to be the subject of a press release, and the difference between a system that runs for fifteen years and one that gets quietly replaced after six sits in them rather than in anything printed on the spec sheet. Six parts on that roof have names the owner knows. The other few thousand lines on the bill of materials are what decide whether those six were worth installing.
