Fig. 6: System-level benchmarking of the proposed ternary content addressable memory (TCAM) while performing the language recognition task.

a Illustration of the HDC algorithm for language recognition. The encoded search vector of the sample text that needs to be recognized is compared to all pre-trained class vectors. The most similar (i.e., closest) vector is then inferred as the result. b The inference accuracy of language recognition over the vector size. c Energy consumption of a single vector comparison with 10,000 bits using ferroelectric superconducting quantum interference device (FeSQUID)-based and static random-access memory (SRAM)-based TCAM cells. d Confusion matrix of Hamming distance (HD) calculation in an FeSQUID TCAM block with 15 cells. We performed a 10,000-point Monte Carlo analysis with 5% standard deviation in the resistance levels and the bias current. e Inference accuracy loss due to the injected errors from the resistance variation.