Reconfigurable Mixed-Kernel Heterojunction Transistors for Personalized Support Vector Machine Hardware

Mixed-dimensional van der Waals heterojunctions have enabled electrostatically tunable response functions in electronics, photonics, and optoelectronic devices. Reconfigurable responses can be further controlled by two factors. First, dual-gated p-n heterojunctions between one- and two-dimensional materials allow asymmetric screening effects.^[1,2] Second, the self-aligned fabrication method is the key to well-controlled carrier modulation and resulting current pathways through constituent heterojunction and p- and n-type series transistors.^[1,2] Here we report reconfigurable mixed-kernel transistors based on dual-gated 1D/2D van der Waals heterojunctions that can generate fully tunable Gaussian, sigmoid, and mixed Gaussian/sigmoid response functions (kernels).^[3] The resulting heterojunction-generated kernels are employed in support vector machine (SVM) algorithms for arrhythmia detection from electrocardiogram (ECG) signals with high classification accuracy. Hardware implementation of SVM for edge applications is currently impractical due to the complexity and high power consumption needed for kernel optimization using conventional complementary metal-oxide semiconductors (CMOS) circuits. In addition, the reconfigurable mixed-kernel heterojunction transistors also allow for personalized detection using Bayesian optimization. A single mixed-kernel heterojunction device can generate the equivalent response function of a CMOS circuit comprised of several dozens of transistors, making this approach an ultralow-power hardware kernel generator with broad applicability to SVM classification. ^[3]

[1] Nano Letters 18, 1421 (2018)

[2] Nature Communications 11, 1565 (2020)

[3] Nature Electronics DOI:10.1038/s41928-023-01042-7 (2023)