Analogue circuits are theoretically more effective than digital architectures due to their lower power consumption and areal footprint, but even the simplest implementations of analogue Gaussian functions require a significant number of circuit elements, which is made worse when implementing tunable and mixed kernel functions.
In this work, we developed a new type device that can generate tunable kernel functions. Its reconfigurable nature allows for personalized detection using Bayesian optimization. A single mixed-kernel heterojunction device can generate the equivalent transfer function of a CMOS circuit comprised of dozens of transistors and thus provides a low-power approach for SVM classification applications.