I read somewhere that PNP transistors have lower noise than NPN transistors. The majority carriers in NPN are electrons where as in PNP they are holes, so I thought the noise would be much be less in NPN transistors.
Which is correct?
I am very sure that the noise generated by holes and electrons is quite the same. If there are differences in the NPN and PNP transistor, they must be an indirect cause and not caused by the type of carriers.
Essentially there are only two dominant types of (white) noise sources
1) shot noise (this is due to the discreteness of carrier flow when passing a potential barrier such as a PN junction). This is independent on the device structure, doping levels, material compostition, carrier type and so on. It is fundamental and the same for holes and electrons (independent on temperature, only depends on current!).
2) thermal or resisitive noise (due to brownian motion of carriers in a resisitive layers) This is also fundamental and only depends on temperature and overall resistance.
Hence transisitors, having e.g. different layer resistances will also generate different amount of noise. This may be the cause for different behavior of PNP and NPN transisors (or more precisely, between any pair of transistors, not just different polarity). Shot noise is quite the same in all bipolar transistors.
BTW: The noise generated by surface states (as mentioned by rawbrawb) is always some kind of 1/f noise and vanishes at a certain frequency (in contrast to the above two white noise sources). Contrary to white noise (which is quite easily explained) 1/f noise is a vrey complicated phenomenon with many different effects that can cause it. It strongly depends on material quality, device strucure and so on. Strong differences between devices are expected for 1/f noise. So always look for the frequency the noise is specified at!
If you are referring to low 1/f noise , then it may be true that some PNP's are quieter than their complementary NPN's. Compare for example 2N2222A vs 2N2907A , 2N3904 vs. 2N3906 or 2N4124 vs. 2N4126 the PNP variants are a bit quieter. It is possible that other factors may play a larger role in determining Noise Figure. I use the 2N2907's in all my Direct Conversion Amateur Receivers. These amps are silent!
Without laboratory testing, there is no answer. I have the instinctual feeling that circuit layout and packaging (how the pins attach to the semiconductor) will have 100+ times greater effect on noise than NPN v PNP.
"Noise" is a tough subject, akin to "accuracy" in measuring; you must first define what your noise is. Anecdotally, the US Navy taught me current flows with the holes, not electrons. I mention this because I immediately think, oh, what if the holes are noisy and the electrons are quiet? (meaning to me that holes, electrons, doesn't matter as they are functionally equivalent opposites)
Invariably, noise in transistors arise because of the construction of the device and also because of native issues. You touched upon one native issue in identifying holes/vs. electrons. Other things that can influence noise are surface states and whether or not the carriers can interact with those states. There are many different ways of making a transistors so one can only talk in trends. So it's safe to say the PNP is lower but one needs to know fabrication details to know specifically WHY.
In general though, if the NPN generates higher Gm at a given current you can get a NPN amplifier to operate with lower noise simply by running it hotter and then filtering the response back to the BW you need. Higher Gm means more noise in total but lower noise per unit BW.