968 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1952 



samples, particularly the fact that one surface of each filament was 

 cemented to a support, which would probably change the surface recom- 

 bination properties somewhat. Aside from the lack of symmetry, the 

 behavior of the two filaments with the higher recombination constants 

 is in reasonable agreement with the calculated curves. The filaments 

 with the lower recombination constants are in poor agreement with 

 calculated values, in that the noise does not fall off with increasing field 

 nearly as fast as calculated. The cause of this behavior is not under- 

 stood. These experimental curves may also be compared with the dotted 

 curve of Fig. 10, calculated on the assumption of volume generation and 

 surface recombination. The similarity is quite poor in all cases. The 

 somewhat better agreement with the surface generation calculations than 

 with the volume generation calculations is not the basis for anything 

 more than a very tentative feeling that the experimental results sup- 

 port the surface generation viewpoint. 



While there are many discrepancies in detail between the experi- 

 mental and calculated relations between noise and magnetic field, these 

 are at least partially understandable in terms of the differences between 

 the experimental setup and the theoretical model. The high degree of 

 qualitative agreement considerably strengthens the hypothesis of noisy 

 injection of minority carriers as an important element in the noise 

 process. 



V. NOISE CORRELATION PHENOMENA 



The noisy hole injection hypothesis leads one to expect certain corre- 

 lation phenomena in the noise voltage observed in neighboring portions 

 of a filament. Consider first noise measurements at a frequency so low 

 that the transit time of a hole* along the filament is negligibly small. 

 This might be a frequency of one kilocycle in a typical experiment. 

 The holes have an average lifetime, from which can be determined an 

 average life path, which is defined as the product of the lifetime by the 

 drift velocity under the existing electric field. Noise voltage measure- 

 ments across segments of the filament much shorter than a life path 

 should be highly correlated, since nearly all the holes which make a 

 transit of one segment will make an almost simultaneous transit of the 

 other segment. On the other hand, noise voltages across segments much 

 longer than a life path should show little correlation, because most of 

 the holes appearing in the two segments are from different sources, and 

 the sources have been assumed to be statistically independent. 



* As before, the concepts apply equally well to electrons in p-type material. 



