A Pattern Recognition Approach to Transistor Array Parameter Variance

Luciano da F. Costa,Filipi N. Silva,Cesar H. Comin

DOI: https://doi.org/10.48550/arXiv.1708.00469

2017-08-02

Abstract:The properties of bipolar junction transistors (BJTs) are known to vary in terms of their parameters. In this work, an experimental approach, including pattern recognition concepts and methods such as principal component analysis (PCA) and linear discriminant analysis (LDA), was used to experimentally investigate the variation among BJTs belonging to integrated circuits known as transistor arrays. It was shown that a good deal of the devices variance can be captured using only two PCA axes. It was also verified that, though substantially small variation of parameters is observed for BJT from the same array, larger variation arises between BJTs from distinct arrays, suggesting the consideration of device characteristics in more critical analog designs. As a consequence of its supervised nature, LDA was able to provide a substantial separation of the BJT into clusters, corresponding to each transistor array. In addition, the LDA mapping into two dimensions revealed a clear relationship between the considered measurements. Interestingly, a specific mapping suggested by the PCA, involving the total harmonic distortion variation expressed in terms of the average voltage gain, yielded an even better separation between the transistor array clusters.

Instrumentation and Detectors,Computational Physics,Data Analysis, Statistics and Probability