Reliability

Updated October 25, 2009

Click here to go to our main page on heat

New for November 2009! And we apologize in advance for the lack of content. Microwave electronics reliability is such a big topic, you can make a career out of it. That is, if you aren't capable of designing a microwave circuit, you can take joy in testing others parts to failure.

For now, we'll just throw out some topics, which we'll get back to in more detail in the future. If anyone wants to contribute to these topics, or start some new ones, we would appreciate it! Contact us.

Eventually, every topic below will have its own Microwaves101 page.

Life test

Life test of microwave hardware usually involves accelerating the failures so that it doesn't take a lifetime to get data. The way to accelerate a life test is to increase the temperature of the hardware. If mean time to failure data can be determined at two temperatures, then some simple math can be done to extrapolate what the MTTF will be at any temperature. This is a universally accepted idea of life testing.

MTTF: mean time to failure, used in the context of a repairable system.

MTBF: mean to between failures, used in the context of a non-repairable system. This what we are evaluating in a life test.

DCOL: operating lifetime with DC-only excitation.

RFOL: operating lifetime with RF (and DC) excitation.

Infant mortality

Failure often occur very early in the lifetime of an electronic product. These are called infant mortalities. The mechanisms for infant mortalities are quite different from end-of-lifetime issues, the former usually due to sloppy workmanship, the latter usually being related to temperature failure.

In order to eliminate infant mortalities in hardware, often a burn-in procedure is used, prior to delivering hardware.

Failure modes

Failure is often a relative term. In a power amplifier, a failure criteria might be 1 dB reduction in saturated power, which might be called a "soft" failure. It all depends on what you can tolerate. Catastrophic failure means that the part is completely unusable (called a "hard" failure). Most life tests use a degraded performance criterion.

DC versus RF operating lifetime

Arrhenius plot

The MTTF data are plotted on a log/log plot of time.

The Arrhenius equation is an empirical observation of chemical reaction rate versus temperature. It was first explained by Swede Svante Arrhenius, in 1884, based on earlier work by Dutch Jacobus Henricus van 't Hoff.

k=Ae^-[Ea/RT]

where k is the rate constant

Ea is the activation energy, typically expressed in electron volts

T is in absolute units (degrees Kelvin)

R is the ideal gas constant 8.31 J/k-mol

More to come!

Graceful degradation

Stress analysis and derating

Stress analysis means considering all of the components in a circuit, evaluating their dissipated power, peak voltage, DC current, etc. Typically a spreadsheet is used. One of these days we'll create an Excel file as an example of this. The end result are columns that provide indication of stress levels as a percentage of the derated stress. A smart designer will be sure not to exceed 100% of ratings, and in military designs, often you must be less than 50% or ratings.

Derating must be done over temperature, components can handle a lot more stress at 25C than at 85 C. Usually the manufacturer will provide derating guidelines, which might amount to a straight-line curve from 100% at 85C to 0% at 150C

Here's some example components and the parameters you need to think about:

Resistors: dissipated power, DC current density (mA/mm), maximum temperature. More information here.

Capacitors: peak voltage (includes the sum of RF and DC voltages, and must take into account VSWR interactions which can double the RF voltage), dissipated power (yes, capacitors dissipate power, here's how to calculate it)

Inductors: DC current.

Transistors: maximum dissipated power density, maximum channel or junction temperature

Transmission lines: DC current density

MMICs: you can just pretend that the manufacturer's data sheet tells you the thermal resistance of the integrated circuit, or you can actually do your own calculation and know a lot more. We'll post a page on this soon.

Stress testing

Stress testing means applying a stress to a component, stepping up the stress, applying it again, while monitoring the device for failure or degradation. That maximum power that the device can stand without (or with minimal) damage is called survival power, or power handling.

Classic examples of components that you might want to subject to stepped stress tests are limiter circuits, and low noise amplifier circuits. A stepped stress might consist of:

1. Measure baseline small-signal S-parameters, or noise figure for an LNA
2. Expose unit to 10 seconds of RF power (10 dBm at first, for example)
3. Remeasure Small signal S-parameters
4. Increase power by 1 dB, return to step 1.

The failure of a part during stepped stress may be subtle (and be less than the measurement repeatability), or catastrophic. Often MMIC suppliers will tell you that their low noise amplifier will survive exposure to 20 dBm (but rarely will they put this in writing...) It might "survive", but its gain might change by one dB or more, and its transmission phase angle might change 10 degrees. In a cell phone, you might not care about this. But in a radar pre-monopulse receiver, 1 dB or 10 degree changes might ruin the monopulse null depth. So you have to carefully consider what "survival power" means in your application.

The power handling of a low noise amplifier might be very different in biased (on) and unbiased (off) states. It seems intuitive that the power handling might be higher in the unbiased state, because the input would not be well matches, and therefore a larger fraction of the incident power would be reflected. However, it is always best to evaluate the power handling in the state where high incident power is required, and that often means measurements in both states.

So, how much power handling can you expect an LNA to provide? The is is heavily dependent on the technology platform, the geometry of the first stage transistor, and the bias circuitry. As a general rule of thumb, a GaAs pHEMT LNA might be good for 50 milliwatts (17 dBm), while a GaN HEMT LNA will be good for 500 milliwatts. If anyone has some measured data to contribute, we'd appreciate it!