Mechanism and Impact of Bipolar Current Voltage Asymmetry in Computational Phase-Change Memory.

Nanoscale resistive memory devices are being explored for neuromorphic and in-memory computing. However, non-ideal device characteristics of read noise and resistance drift pose significant challenges to the achievable computational precision. Here, it is shown that there is an additional non-ideality that can impact computational precision, namely the bias-polarity-dependent current flow. Using phase-change memory (PCM) as a model system, it is shown that this "current-voltage" non-ideality arises both from the material and geometrical properties of the devices. Further, we discuss the detrimental effects of such bipolar asymmetry on in-memory matrix-vector multiply (MVM) operations and provide a scheme to compensate for it.

Optimised weight programming for analogue memory-based deep neural networks.

Analogue memory-based deep neural networks provide energy-efficiency and per-area throughput gains relative to state-of-the-art digital counterparts such as graphics processing units. Recent advances focus largely on hardware-aware algorithmic training and improvements to circuits, architectures, and memory devices. Optimal translation of software-trained weights into analogue hardware weights-given the plethora of complex memory non-idealities-represents an equally important task. We report a generalised computational framework that automates the crafting of complex weight programming strategies to minimise accuracy degradations during inference, particularly over time. The framework is agnostic to network structure and generalises well across recurrent, convolutional, and transformer neural networks. As a highly flexible numerical heuristic, the approach accommodates arbitrary device-level complexity, making it potentially relevant for a variety of analogue memories. By quantifying the limit of achievable inference accuracy, it also enables analogue memory-based deep neural network accelerators to reach their full inference potential.

Gel-on-a-chip: continuous, velocity-dependent DNA separation using nanoscale lateral displacement.

We studied the trajectories of polymers being advected while diffusing in a pressure driven flow along a periodic pillar nanostructure known as nanoscale deterministic lateral displacement (nanoDLD) array. We found that polymers follow different trajectories depending on their length, flow velocity and pillar array geometry, demonstrating that nanoDLD devices can be used as a continuous polymer fractionation tool. As a model system, we used double-stranded DNA (dsDNA) with various contour lengths and demonstrated that dsDNA in the range of 100-10 000 base pairs (bp) can be separated with a size-selective resolution of 200 bp. In contrast to spherical colloids, a polymer elongates by shear flow and the angle of polymer trajectories with respect to the mean flow direction decreases as the mean flow velocity increases. We developed a phenomenological model that explains the qualitative dependence of the polymer trajectories on the gap size and on the flow velocity. Using this model, we found the optimal separation conditions for dsDNA of different sizes and demonstrated the separation and extraction of dsDNA fragments with over 75% recovery and 3-fold concentration. Importantly, this velocity dependence provides a means of fine-tuning the separation efficiency and resolution, independent of the nanoDLD pillar geometry.

Characterizing fluorocarbon assisted atomic layer etching of Si using cyclic Ar/C4F8 and Ar/CHF3 plasma.

With the increasing interest in establishing directional etching methods capable of atomic scale resolution for fabricating highly scaled electronic devices, the need for development and characterization of atomic layer etching processes, or generally etch processes with atomic layer precision, is growing. In this work, a flux-controlled cyclic plasma process is used for etching of SiO2 and Si at the Angstrom-level. This is based on steady-state Ar plasma, with periodic, precise injection of a fluorocarbon (FC) precursor (C4F8 and CHF3) and synchronized, plasma-based Ar+ ion bombardment [D. Metzler et al., J. Vac. Sci. Technol., A 32, 020603 (2014) and D. Metzler et al., J. Vac. Sci. Technol., A 34, 01B101 (2016)]. For low energy Ar+ ion bombardment conditions, physical sputter rates are minimized, whereas material can be etched when FC reactants are present at the surface. This cyclic approach offers a large parameter space for process optimization. Etch depth per cycle, removal rates, and self-limitation of removal, along with material dependence of these aspects, were examined as a function of FC surface coverage, ion energy, and etch step length using in situ real time ellipsometry. The deposited FC thickness per cycle is found to have a strong impact on etch depth per cycle of SiO2 and Si but is limited with regard to control over material etching selectivity. Ion energy over the 20-30 eV range strongly impacts material selectivity. The choice of precursor can have a significant impact on the surface chemistry and chemically enhanced etching. CHF3 has a lower FC deposition yield for both SiO2 and Si and also exhibits a strong substrate dependence of FC deposition yield, in contrast to C4F8. The thickness of deposited FC layers using CHF3 is found to be greater for Si than for SiO2. X-ray photoelectron spectroscopy was used to study surface chemistry. When thicker FC films of 11 Å are employed, strong changes of FC film chemistry during a cycle are seen whereas the chemical state of the substrate varies much less. On the other hand, for FC film deposition of 5 Å for each cycle, strong substrate surface chemical changes are seen during an etching cycle. The nature of this cyclic etching with periodic deposition of thin FC films differs significantly from conventional etching with steady-state FC layers since surface conditions change strongly throughout each cycle.

Wafer-scale integration of sacrificial nanofluidic chips for detecting and manipulating single DNA molecules.

Wafer-scale fabrication of complex nanofluidic systems with integrated electronics is essential to realizing ubiquitous, compact, reliable, high-sensitivity and low-cost biomolecular sensors. Here we report a scalable fabrication strategy capable of producing nanofluidic chips with complex designs and down to single-digit nanometre dimensions over 200 mm wafer scale. Compatible with semiconductor industry standard complementary metal-oxide semiconductor logic circuit fabrication processes, this strategy extracts a patterned sacrificial silicon layer through hundreds of millions of nanoscale vent holes on each chip by gas-phase Xenon difluoride etching. Using single-molecule fluorescence imaging, we demonstrate these sacrificial nanofluidic chips can function to controllably and completely stretch lambda DNA in a two-dimensional nanofluidic network comprising channels and pillars. The flexible nanofluidic structure design, wafer-scale fabrication, single-digit nanometre channels, reliable fluidic sealing and low thermal budget make our strategy a potentially universal approach to integrating functional planar nanofluidic systems with logic circuits for lab-on-a-chip applications.

Nanoscale lateral displacement arrays for the separation of exosomes and colloids down to 20†nm.

Deterministic lateral displacement (DLD) pillar arrays are an efficient technology to sort, separate and enrich micrometre-scale particles, which include parasites, bacteria, blood cells and circulating tumour cells in blood. However, this technology has not been translated to the true nanoscale, where it could function on biocolloids, such as exosomes. Exosomes, a key target of 'liquid biopsies', are secreted by cells and contain nucleic acid and protein information about their originating tissue. One challenge in the study of exosome biology is to sort exosomes by size and surface markers. We use manufacturable silicon processes to produce nanoscale DLD (nano-DLD) arrays of uniform gap sizes ranging from 25 to 235 nm. We show that at low Péclet (Pe) numbers, at which diffusion and deterministic displacement compete, nano-DLD arrays separate particles between 20 to 110 nm based on size with sharp resolution. Further, we demonstrate the size-based displacement of exosomes, and so open up the potential for on-chip sorting and quantification of these important biocolloids.

Hydrodynamics of diamond-shaped gradient nanopillar arrays for effective DNA translocation into nanochannels.

Effective DNA translocation into nanochannels is critical for advancing genome mapping and future single-molecule DNA sequencing technologies. We present the design and hydrodynamic study of a diamond-shaped gradient pillar array connected to nanochannels for enhancing the success of DNA translocation events. Single-molecule fluorescence imaging is utilized to interrogate the hydrodynamic interactions of the DNA with this unique structure, evaluate key DNA translocation parameters, including speed, extension, and translocation time, and provide a detailed mapping of the translocation events in nanopillar arrays coupled with 10 and 50 µm long channels. Our analysis reveals the important roles of diamond-shaped nanopillars in guiding DNA into as small as 30 nm channels with minimized clogging, stretching DNA to nearly 100% of their dyed contour length, inducing location-specific straddling of DNA at nanopillar interfaces, and modulating DNA speeds by pillar geometries. Importantly, all critical features down to 30 nm wide nanochannels are defined using standard photolithography and fabrication processes, a feat aligned with the requirement of high-volume, low-cost production.