The continuing advance of artificial intelligence requires initiatives to address the potential harms of the technology and efforts to develop new energy-efficient electronic hardware.

The researchers used a 12-nm silicon process to fabricate their chip, which contains 22 billion transistors, 256 cores and 224 MB of on-chip memory in an area of 800 mm2. Each core contains memory and processing elements, and can execute up to 8,192 operations per cycle at a precision of 2 bits. Software was co-designed for the chip to ensure that the full capabilities of the architecture were utilized

The researchers — who are based at the Beijing Institute of Technology, Weizmann Institute of Science and the Holon Institute of Technology — stacked crossed one-dimensional tungsten disulfide (WS2) nanotubes to create a zero-dimensional interface of size 10 nm × 10 nm × 2 nm. A spontaneous electric polarization switching occurred due to van der Waals sliding. The ferroelectric diodes had an on/off

The researchers — who are based at various institutions in the United States, South Korea and Germany — grew nanometre-scale islands of barium titanate on a strontium titanate substrate coated with few-layer graphene; the nuclei were sufficiently small that transmission electron microscopy could be used to preclude any pinholes at the growth sites of the islands. The barium titanate grown in this way

The researchers — who are based at the University of Washington and the Université Grenoble Alpes — dropped the fliers from a commercial drone at a height of 40 m. The robots got carried laterally by a light breeze and could operate in wind speeds of more than 5 m s–1. Free-fall trajectory and the lateral distance travelled depended on the flier shape: the microflier travels further as it tumbles through

The researchers — who are based at ShanghaiTech University and Pusan National University — used an electric field together with a double-layer gas flow (one with and one without metal nanoparticles) to print architectures containing multiple materials. Charged nanoparticles are guided by the coupled electrical field and aerosol flow, and are deposited on the printing area. The team show that the method

Computing hardware that can find the ground states of the Ising model could provide a powerful route to solving difficult combinatorial optimization problems.

The researchers — who are based at Northwestern University, Tsinghua University, University of Illinois at Chicago and the University of Chicago — used a thin (100 nm) patterned gold disk encapsulated in polyimide as a soft thermal sensor for interfacing with the kidney surface. The biosensor monitors the local temperature and thermal conductivity of the kidney, which are used as a proxy for inflammation

The researchers — who are based at the University of St Andrews, the University of Glasgow and the Institute of Physical-Organic Chemistry and Coal Chemistry, Ukraine — developed a device consisting of an organic gain medium that is excited by electroluminescence emitted from an organic light-emitting diode integrated on top. By physically separating the regions where charge injection and laser population

The researchers — who are based at various institutions in China — developed a modified bidirectional gate recurrent unit neural network that was used to mitigate the signal distortion typically observed when electro-optical modulation occurs within silicon microring modulators. Using this approach, along with discrete multitone modulation with bit and power loading, the maximum spectral efficiency

With micro- and nanoelectronic devices and circuits, the most important technical requirements are related to performance metrics (mobility and subthreshold swing for transistors, and switching energy and endurance for memristors), reproducibility (yield, variability, reliability and stability) and fabrication processes (the use of a competitive device structure and size, the use of industry-compatible

Francis Willett and colleagues — who are based at various institutes in the USA — have now created a speech-to-text brain–computer interface that relies on four intracortical microelectrode arrays to collect signals, and a recurrent neural network and language model to decode them into words. With the system, the attempted speech of a participant, who can no longer speak clearly due to amyotrophic

The researchers — who are based at IBM Research — fabricated a 14-nm inference chip containing 34 tiles, each of which contains a crossbar array of phase-change memory devices that can perform matrix–vector multiplications. Overall, the chip has 35 million phase-change memory devices. The chip also contains digital-to-analogue input, analogue-to-digital output, analogue peripheral circuitry and parallel-2D-mesh

With advances in capabilities coming hand in hand with calls for regulation, 2023 is proving to be a critical year for brain–computer interfaces.

The researchers — who are based at Delft University of Technology, QuTech and the Netherlands Organisation for Applied Scientific Research — created their crossbar architecture using strained germanium quantum wells with double-barrier gates and a diagonal plunger line layout. This approach allows a single gate to control multiple quantum dots with a sublinear number of control terminals, thus reducing

The researchers — who are based at the National Institute for Materials Science, Mie University, Keio University, Osaka University and TDK Corporation — studied 252 combinations of different magnesium (Mg2+) and gallium (Ga3+) cation occupations in the spinel material. The simulations were performed using a machine-learning factorization machine developed by the team in combination with D-Wave’s Advantage