A key difference which is most of the times left aside is that most ACTIVE electronic devices are designed, manufactured and TESTED (accepted / rejected) for meeting a very specific set of requirements:
- We can call the above set of target requirements PRIMARY or MUST, which means we really need to achieve a very good performance at these requirements in order to differentiate our device and make it better than a "standard" or baseline device.
- Then, there is a second group of requirements, SECONDARY or NICE TO HAVE, which cannot be overlooked, or our device may be under the "standard" device in these other parameters. Most often, the secondary requirement are at odds with the primary ones, meaning that getting better at one of the primary parameters will make worse the secondary parameter. In other occassions, the secondary requirements are simply expensive to improve and not really needed for our targe market or applications.
The above happens simply because it is not feasible to create an active device which is best suited for all (many) intended applications.
For instance, and refering to BJT design, for a given manufacturing technology, "high voltage switching" (higher avalanche collector-base breakdown) will need a higher diffusion dopants area, which in turn will make the input and output parasitic capacitances higher, and so the resulting BJT will be slower than if we decide not to improve the BVcb. In this simple example, the desired characteristics "higher BVcb" and "fastest switching times" cannot be improved simultaneously. As a result, when designing a very linear device I will sacrifice higher BVcb in order to get a higher Ft (unity gain bandwith).
Returning to the original question, there are THREE main reasons which explain why manufacturers sometimes "label" or subtitle a device with such adjectives as "designed for switching applications" or "general purpose linear amplifier":
- Some of the target parameters you have to optimize in order to get the "best" switching device under a given manufacturing technology are of little use or work against the best linear amplifier behaviour: ruggedness of parasitic internal diodes/SCRs, very high peak current, ESD protection, storage and delay time optimization, high BVcb, thermal stability...
- Nowadays, it is common to build discrete power/switching devices as many internally devices connected in paralell. This technique naturally improves many of the above parameters which make a "good switching device", however, will also make the device much less linear, literally.
- Price! Improving a parameter which is not-needed-for-the-target-application will surely rise costs up! Why? Because the manufacturer will now have to characterize the device also for the not-really-needed parameters and, worse, REJECT the manufactured devices which do not satisfy the named parameter during the testing phase. This will lower the yield of the manufacturing process and drive the prices up.
The last item, characterizing and testing for a not-really-needed parameter is easy to spot on many datasheets. You will notice many general purpose (lineal amplifier) BJTs do not guarantee nor even state the expected values for storage and delay times. On the other hand, switching BJTs will most of the times fully characterize switching times, waveforms and related-parameters, but will not get into much detail nor depict the variability of hie/hfe/hoe curves.