Not every inverter failure develops slowly. A recent FMEA study of PV inverters identified surge-induced varistor short circuits, capacitor degradation, and cooling fan failures as the dominant risks, alongside relays and power semiconductors [1]. Some of these occur without warning. However, a significant proportion of ageing announces itself over months. Usually, however, nobody is listening.

One easily measurable signal is thermal: the thermal resistance between the semiconductor and the ambient temperature rises slowly, causing the device to run hotter at the same load. Eventually, derating or shutdown kicks in. The customer notices a fall in yield, the installer a service call and the distributor a warranty claim.

However, this process can be monitored using sensors that are already incorporated into many devices.

The Thermal Path and Its Weak Links

Power semiconductors are among the most critical components in power converters. In a widely cited industry survey, 31 per cent of respondents ranked them as the most fragile component, ahead of capacitors [2].

Their losses have to travel through a stack of layers: chip, die attach, substrate, baseplate or package, thermal interface, heat sink and ambient air. Each layer contributes to the total thermal resistance (Rth). Two links in particular age:

The thermal interface. Thermal grease is mechanically worked by every temperature cycle. As the device and heatsink expand at different rates, each load change creates shear and peeling stress at the interface. With each cycle, the grease is squeezed out of the contact area (pump-out). At high temperatures, the filler can also separate from the matrix (dry out) [3]. For power modules, pump-out is documented as a distinct mechanism of long-term thermal degradation.

The solder layers. Die attach and substrate solder also fatigue under thermal cycling. Cracks and delamination reduce the effective heat conduction area.

An inverter in a residential storage system undergoes such cycles daily: from sunrise to end of charge and evening discharge. Over 20 years, this equates to approximately 7,300 large daily cycles, in addition to numerous smaller cycles resulting from passing clouds and load changes.

What the automotive industry defines as end of life

The automotive sector offers an established reference. The ECPE guideline AQG 324 defines two end-of-life criteria for power cycling tests: a 5 per cent increase in forward voltage or a 20 per cent increase in thermal resistance [5]. The same 20 per cent criterion is used in lifetime studies on three-phase MOSFET inverters to assess die attach fatigue.

However, these are test bench values, not field limits, and their consistency is actively debated in the research community. However, they provide a clear indication: R_(th) is a recognised indicator of ageing, and relevant changes lie in the double-digit percentage range. With good signal processing, this can be resolved in the field.

How Rth trending works in the device

The principle is simple: Rth is temperature difference divided by power loss. Implementing this is more challenging.

Power loss must be estimated. The firmware knows the current, voltage, switching frequency and temperature. A loss model uses these to calculate the power dissipated in the semiconductor. For SiC MOSFETs, the model must account for the temperature dependence of the on-resistance.

Temperatures are captured. Multiple NTC sensors on the semiconductor, the heat sink and inside the enclosure provide temperature readings along the thermal path.

Comparable operating points must be found. This is the critical part. A passive heatsink has high thermal inertia and responds far more slowly than load changes. Only quasi-steady states at a similar load and ambient temperature are meaningful. The firmware collects these, sorts them into classes and compares each class against its own history.

Setting a reference: The first weeks after commissioning provide the baseline. Every subsequent deviation is assessed relative to this baseline, rather than against a datasheet value.

The result is not an exact junction temperature, but rather a trend. And that trend is exactly what matters.

What the trend reveals:

A slow, steady rise over a period of years indicates material ageing in the interface or solder layers. A sudden increase within a few days usually indicates an external cause, such as the device being enclosed after installation, convection paths being blocked, or the mounting loosening.

The first case is otherwise difficult to detect. The FMEA study mentioned above shows that severe ageing mechanisms, such as die attach fatigue resulting from thermal cycling, are also among the most difficult to detect [1]. Trending Rth makes this measurable remotely and before the customer notices any loss of yield.

Why This Matters for Distribution

For distributors and installers, Rth trending shifts the discussion from assumptions to data.

Service visits can be planned instead of reacting to failures.

Installation errors become apparent long before they result in warranty issues.

In the event of a claim, there is a thermal history of the device instead of conflicting statements.

It also answers a question that everyone should ask over a 20-year service life: how can I be sure how the device is performing in year twelve?

Materials and measurement belong together.

Monitoring does not replace robust design. For the Ampereq Gen3, we have therefore combined both, using ceramic thermal pads instead of grease and passive cooling without a fan, as well as Rth trending in the firmware. The pads significantly reduce the risk of pump-out; eliminating the fan removes one of the dominant risks from the FMEA entirely; and the trending monitors whatever happens to the device over its lifetime.

The question is not whether an inverter ages. The question is whether it informs you of this.

Sources in the first comment.