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.

Toward Software-Equivalent Accuracy on Transformer-Based Deep Neural Networks With Analog Memory Devices.

Recent advances in deep learning have been driven by ever-increasing model sizes, with networks growing to millions or even billions of parameters. Such enormous models call for fast and energy-efficient hardware accelerators. We study the potential of Analog AI accelerators based on Non-Volatile Memory, in particular Phase Change Memory (PCM), for software-equivalent accurate inference of natural language processing applications. We demonstrate a path to software-equivalent accuracy for the GLUE benchmark on BERT (Bidirectional Encoder Representations from Transformers), by combining noise-aware training to combat inherent PCM drift and noise sources, together with reduced-precision digital attention-block computation down to INT6.

A three-dimensional micro-electrode array for in-vitro neuronal interfacing.

OBJECTIVE: In this paper, we report on the development of an easy-to-fabricate three-dimensional Micro-Electrode Array (3D-MEA) specifically designed for brain-on-a-dish applications. APPROACH: The proposed device consists of pillar-shaped gold microelectrodes realized by electroplating directly on top of a standard MEA, making this approach highly versatile and convenient for batch fabrication. Moreover, with this simple technique, it is possible to obtain electrodes with a height of more than 100 µm onto different kind of substrates, ranging from glass to flexible plastic ones. MAIN RESULTS: This novel 3D-MEA structure has been validated with acute brain slices, successfully recording both epileptiform-like discharges (upon the administration of 4-AP), and electrically-evoked neuronal activity. The preliminary validation showed a substantial improvement in the signals amplitude with respect to both commercial and custom planar electrodes thanks to a better coupling offered by the peculiar shape of the three-dimensional electrodes. SIGNIFICANCE: Beside the versatility of the fabrication approach, which allows to obtain 3D MEA devices onto both rigid and flexible substrates, the reported validation showed how the pillar approach can outperform standard planar MEA recordings in terms of signal amplitude. Moreover, thanks to the possibility of obtaining multi-level 3D structures within the same device, the proposed fabrication technique offers an interesting and flexible approach for the development of a new family of electrophysiological tools for 3D in vitro electrophysiology, in particular for acute brain slices and 3D neuronal cultures for brain-on-a-dish applications.

Surface Initiated Polymer Thin Films for the Area Selective Deposition and Etching of Metal Oxides.

The area selective growth of polymers and their use as inhibiting layers for inorganic film depositions may provide a valuable self-aligned process for fabrication. Polynorbornene (PNB) thin films were grown from surface-bound initiators and show inhibitory properties against the atomic layer deposition (ALD) of ZnO and TiO2. Area selective control of the polymerization was achieved through the synthesis of initiators that incorporate surface-binding ligands, enabling their selective attachment to metal oxide features versus silicon dielectrics, which were then used to initiate surface polymerizations. The subsequent use of these films in an ALD process enabled the area selective deposition (ASD) of up to 39 nm of ZnO. In addition, polymer thickness was found to play a key role, where films that underwent longer polymerization times were more effective at inhibiting higher numbers of ALD cycles. Finally, while the ASD of a TiO2 film was not achieved despite blanket studies showing inhibition, the ALD deposition on polymer regions of a patterned film produced a different quality metal oxide and therefore altered its etch resistance. This property was exploited in the area selective etch of a metal feature. This demonstration of an area selective surface-grown polymer to enable ASD and selective etch has implications for the fabrication of both micro- and nanoscale features and surfaces.

Growing Patterned, Cross-linked Nanoscale Polymer Films from Organic and Inorganic Surfaces Using Ring-Opening Metathesis Polymerization.

The ability to modify substrates with thin polymer films allows for the tailoring of surface properties, and through combination of patterning finds use in a large variety of applications such as electronics and lab-on-chip devices. Although many techniques can be used to afford polymer-modified surfaces such as surface-initiated polymerization or layer-by-layer methodologies, their stability in a wide range of environments as well as their ability to target specific chemistry are critical factors to enable their successful application. In this paper, we report a facile technique in creating nanoscale polymer thin films using solid-state continuous assembly of polymers via ring-opening metathesis polymerization (ssCAPROMP) directly from surfaces functionalized through silanization. Using a polymeric precursor that includes norbornene moieties, a highly dense cross-linked network of polymer can be grown in a bottom-up fashion to afford thin films from an olefin-terminated silanized planar surface. Such nanotechnology affords films retaining the desirable qualities of previously reported methods while, at the same time, being covalently bound to the substrate: they are virtually pinhole free and can be reinitiated multiple times. By combining this process with microcontact printing, patterned films can be created by either the patterned deposition of a catalyst or by controlling the surface silanization chemistry and placement of olefin-terminated and nonreactive silanes. Additionally, patterned ssCAPROMP films were grown from SU-8 by selectively functionalizing the surface through masking and lift-off processes after the silanization step, thereby spatially controlling the surface-initiation, and subsequent polymer film formation. These patterned films expand the capabilities of the CAPROMP process and offer advantages over other film formation techniques in processes where patterned substrates and modified but robust surface chemistries are utilized.

Fifteen Nanometer Resolved Patterns in Selective Area Atomic Layer Deposition-Defectivity Reduction by Monolayer Design.

Selective area atomic layer deposition (SA-ALD) offers the potential to replace a lithography step and provide a significant advantage to mitigate pattern errors and relax design rules in semiconductor fabrication. One class of materials that shows promise to enable this selective deposition process are self-assembled monolayers (SAMs). In an effort to more completely understand the ability of these materials to function as barriers for ALD processes and their failure mechanism, a series of SAM derivatives were synthesized and their structure-property relationship explored. These materials incorporate different side group functionalities and were evaluated in the deposition of a sacrificial etch mask. Monolayers with weak supramolecular interactions between components (for example, van der Waals) were found to direct a selective deposition, though they exhibit significant defectivity at and below 100 nm feature sizes. The incorporation of stronger noncovalent supramolecular interacting groups in the monolayer design, such as hydrogen bonding units or pi-pi interactions, did not produce an added benefit over the weaker interacting components. Incorporation of reactive moieties in the monolayer component that enabled the polymerization of an SAM surface, however, provided a more effective barrier, greatly reducing the number and types of defects observed in the selectively deposited ALD film. These reactive monolayers enabled the selective deposition of a film with critical dimensions as low as 15 nm. It was also found that the selectively deposited film functioned as an effective barrier for isotropic etch chemistries, allowing the selective removal of a metal without affecting the surrounding surface. This work enables selective area ALD as a technology through (1) the development of a material that dramatically reduces defectivity and (2) the demonstrated use of the selectively deposited film as an etch mask and its subsequent removal under mild conditions.

Simple and versatile methods to integrate directed self-assembly with optical lithography using a polarity-switched photoresist.

We report novel strategies to integrate block copolymer self-assembly with 193 nm water immersion lithography. These strategies employ commercially available positive tone chemically amplified photoresists to spatially encode directing information into precise topographical or chemical prepatterns for the directed self-assembly of block copolymers. Each of these methods exploits the advantageous solubility and thermal properties of polarity-switched positive tone photoresist materials. Precisely registered, sublithographic self-assembled structures are fabricated using these versatile integration schemes which are fully compatible with current optical lithography patterning materials, processes, and tooling.