CMOS-based conventional Von-Neumann architectures are immensely scalable, but add considerable delay due to data transfer between the core and
memory units. This is particularly time-consuming in case of computations involving matrix multiplication such as Principal Component Analysis
(PCA). PCA is gaining increasing importance for large data classification and is useful in analysis of bioinformatics data such as protein sequences and gene expression. Given the high quantity of input genetic data, hardware level accelerators are becoming essential. In this talk, a
Pd/WO3/W based memristive accelerator network, termed memPCA, is trained using Sanger’s rule for PCA of an example large data set of the ‘Pfam04237’ protein sequence family. The speedup obtained is compared to a direct PCA computation in Scilab software on Intel quad-core i7-4770 machine.