Decades ago I worked on a satellite project. Lotta good stuff to learn; most of my circuits worked first time, so they let me handle integration of the entire "box" we were
One of my circuits was a stripped-down version of the customer-suggested relay-driver schematics; they suggested an AC_coupled positive-feedback 2-transistor circuit. I thought "We don't need that."
Turns out the SLOW switching of 2N2222s at 28 volts and 0.2 amps (for latching relays) was killing the transistors. At 50% point, that's 14 volts and 0.1 amps, or 1.4 watts.
Why is that bad? The die is about 1mm cube; the thermal timeconstant of that is 11.4 milliSeconds (a cubic meter of silicon has 11,400 seconds thermal Tau). The thermal capacity of silicon is 1.6 picoJoules/cubic_micron per degree C. The thermal capacity is thus 1.6 picoJoules/cubic_micron * 1000 * 1000 * 1000 === 1.6 milliJoules per degree C.
The heat rise in a millisecond was 1.4 milliWatts / 1.6 milliJoules/degree C, or about 0.8 degree C.
Why was 2N2222 killed? All the heat was dissipated in the 10 micron thick collector implant, not in the 1,000 micron thick die (maybe only 300 micron, of a typical wafer, but does not matter). Our heat rise was not 0.8 degree C per millisecond, but 30x to 100X higher.
And any current imbalance in the emitter region will lead to thermal runaway.
Cure? We kludged the PCB to use the extra transistors and the positive feedback. The customer did not want to respin the PCB. To teach me a lesson.
I learned my lesson on Safe Operating Area.
That PCB/box is still in orbit today. Kludged, to avoid SOA failure.
The 2N2222 was really 2N2219, a large metal case device to dissipate 0.5 watt.
I made the mistake; I'd striped out the switch-the-device-quickly extra components. Thus the 10 micron thick active region (the collector implant) had to dissipate the 1.4 watts, briefly, and the device developed hot spots due to slight emitter variations.
By reducing the "briefly" from milliseconds to nanoseconds, by positive feedback,
the switching event became non-destructive. I needed to learn about SOA.
The 2N22xx family has a thin base and achieves 100+MHz bandwidth at moderate current.