Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
A circuit that incorporates 2,048 memristors and 30,080 transistors can implement Bayesian inference via an approach that performs computation locally and with minimal data movement. The optical microscopy image on the cover shows the die of the Bayesian system.
A skin-like sensory system, consisting of a substrate-less nanomesh strain sensor and an unsupervised meta-learning framework, enables the rapid recognition of various hand movements with minimal training and can work for any user. The device is able to complete various tasks, including virtual keyboard typing and object recognition.
This Perspective explores the potential of large-area electronics in wirelessly powered sensor nodes for the Internet of Things, considering low-power circuits for digital processing and signal amplification, as well as diodes and printed antennas for data communication and radiofrequency energy harvesting.
The magnetic anisotropy of the van der Waals ferromagnet Fe5GeTe2 can be continuously tuned—from an initial out-of-plane orientation to a canted orientation and then finally to an in-plane orientation—using electrical gating.
By combining p-type transistors made with silicon-on-insulator technology and n-type transistors made with two-dimensional molybdenum disulfide, heterogeneous complementary field-effect transistors can be fabricated on the wafer scale.
An activity-difference training approach, which employs 64 × 64 memristor arrays with integrated complementary metal–oxide–semiconductor control circuitry, can be used to train a deep neural network to efficiently classify Braille words.
A Bayesian machine can be implemented in a system with distributed memristors, allowing it to locally perform computation with minimal energy movement.
A nanomesh sensor that is directly printed on a person’s hand and is coupled with an unsupervised meta-learning framework can provide user-independent and data-efficient recognition of different hand tasks.
Three-layer heterostructures consisting of an indium gallium arsenide semiconducting film, a lithium niobate piezoelectric film, and a silicon substrate can be used to create acoustoelectric amplifiers that operate at gigahertz frequencies with large non-reciprocal gain and low noise in continuous operation.