Nonideality-Aware Training for Accurate and Robust Low-Power Memristive Neural Networks.

Recent years have seen a rapid rise of artificial neural networks being employed in a number of cognitive tasks. The ever-increasing computing requirements of these structures have contributed to a desire for novel technologies and paradigms, including memristor-based hardware accelerators. Solutions based on memristive crossbars and analog data processing promise to improve the overall energy efficiency. However, memristor nonidealities can lead to the degradation of neural network accuracy, while the attempts to mitigate these negative effects often introduce design trade-offs, such as those between power and reliability. In this work, authors design nonideality-aware training of memristor-based neural networks capable of dealing with the most common device nonidealities. The feasibility of using high-resistance devices that exhibit high I-V nonlinearity is demonstrated-by analyzing experimental data and employing nonideality-aware training, it is estimated that the energy efficiency of memristive vector-matrix multipliers is improved by almost three orders of magnitude (0.715 TOPs-1 W-1 to 381 TOPs-1 W-1 ) while maintaining similar accuracy. It is shown that associating the parameters of neural networks with individual memristors allows to bias these devices toward less conductive states through regularization of the corresponding optimization problem, while modifying the validation procedure leads to more reliable estimates of performance. The authors demonstrate the universality and robustness of this approach when dealing with a wide range of nonidealities.

Simulation of Inference Accuracy Using Realistic RRAM Devices.

Resistive Random Access Memory (RRAM) is a promising technology for power efficient hardware in applications of artificial intelligence (AI) and machine learning (ML) implemented in non-von Neumann architectures. However, there is an unanswered question if the device non-idealities preclude the use of RRAM devices in this potentially disruptive technology. Here we investigate the question for the case of inference. Using experimental results from silicon oxide (SiO x ) RRAM devices, that we use as proxies for physical weights, we demonstrate that acceptable accuracies in classification of handwritten digits (MNIST data set) can be achieved using non-ideal devices. We find that, for this test, the ratio of the high- and low-resistance device states is a crucial determinant of classification accuracy, with ~96.8% accuracy achievable for ratios >3, compared to ~97.3% accuracy achieved with ideal weights. Further, we investigate the effects of a finite number of discrete resistance states, sub-100% device yield, devices stuck at one of the resistance states, current/voltage non-linearities, programming non-linearities and device-to-device variability. Detailed analysis of the effects of the non-idealities will better inform the need for the optimization of particular device properties.

Memristor-Based Edge Detection for Spike Encoded Pixels.

Memristors have many uses in machine learning and neuromorphic hardware. From memory elements in dot product engines to replicating both synapse and neuron wall behaviors, the memristor has proved a versatile component. Here we demonstrate an analog mode of operation observed in our silicon oxide memristors and apply this to the problem of edge detection. We demonstrate how a potential divider exploiting this analog behavior can prove a scalable solution to edge detection. We confirm its behavior experimentally and simulate its performance on a standard testbench. We show good performance comparable to existing memristor based work with a benchmark score of 0.465 on the BSDS500 dataset, while simultaneously maintaining a lower component count.

The interplay between structure and function in redox-based resistance switching.

We report a study of the relationship between oxide microstructure at the scale of tens of nanometres and resistance switching behaviour in silicon oxide. In the case of sputtered amorphous oxides, the presence of columnar structure enables efficient resistance switching by providing an initial structured distribution of defects that can act as precursors for the formation of chains of conductive oxygen vacancies under the application of appropriate electrical bias. Increasing electrode interface roughness decreases electroforming voltages and reduces the distribution of switching voltages. Any contribution to these effects from field enhancement at rough interfaces is secondary to changes in oxide microstructure templated by interface structure.

Controlling and modelling the wetting properties of III-V semiconductor surfaces using re-entrant nanostructures.

Inorganic semiconductors such as III-V materials are very important in our everyday life as they are used for manufacturing optoelectronic and microelectronic components with important applications span from energy harvesting to telecommunications. In some applications, these components are required to operate in harsh environments. In these cases, having waterproofing capability is essential. Here we demonstrate design and control of the wettability of indium phosphide based multilayer material (InP/InGaAs/InP) using re-entrant structures fabricated by a fast electron beam lithography technique. This patterning technique enabled us to fabricate highly uniform nanostructure arrays with at least one order of magnitude shorter patterning times compared to conventional electron beam lithography methods. We reduced the surface contact fraction significantly such that the water droplets may be completely removed from our nanostructured surface. We predicted the wettability of our patterned surface by modelling the adhesion energies between the water droplet and both the patterned surface and the dispensing needle. This is very useful for the development of coating-free waterproof optoelectronic and microelectronic components where the coating may hinder the performance of such devices and cause problems with semiconductor fabrication compatibility.

Morphological Variations of Explosive Residue Particles and Implications for Understanding Detonation Mechanisms.

The possibility of recovering undetonated explosive residues following detonation events is well-known; however, the morphology and chemical identity of these condensed phase postblast particles remains undetermined. An understanding of the postblast explosive particle morphology would provide vital information during forensic examinations, allowing rapid initial indication of the explosive material to be microscopically determined prior to any chemical analyses and thereby saving time and resources at the crucial stage of an investigation. In this study, condensed phase particles collected from around the detonations of aluminized ammonium nitrate and RDX-based explosive charges were collected in a novel manner utilizing SEM stubs. By incorporating the use of a focused ion beam during analysis, for the first time it is possible to determine that such particles have characteristic shapes, sizes, and internal structures depending on the explosive and the distance from the detonation at which the particles are recovered. Spheroidal particles (10-210 µm) with microsurface features recovered following inorganic charge detonations were dissimilar to the irregularly shaped particles (5-100 µm) recovered following organic charge firings. Confirmatory analysis to conclude that the particles were indeed explosive included HPLC-MS, Raman spectroscopy, and mega-electron volt-secondary ionization mass spectrometry. These results may impact not only forensic investigations but also the theoretical constructs that govern detonation theory by indicating the potential mechanisms by which these particles survive and how they vary between the different explosive types.

Electrospray synthesis and properties of hierarchically structured PLGA TIPS microspheres for use as controlled release technologies.

Microsphere-based controlled release technologies have been utilized for the long-term delivery of proteins, peptides and antibiotics, although their synthesis poses substantial challenges owing to formulation complexities, lack of scalability, and cost. To address these shortcomings, we used the electrospray process as a reproducible, synthesis technique to manufacture highly porous (>94%) microspheres while maintaining control over particle structure and size. Here we report a successful formulation recipe used to generate spherical poly(lactic-co-glycolic) acid (PLGA) microspheres using the electrospray (ES) coupled with a novel thermally induced phase separation (TIPS) process with a tailored Liquid Nitrogen (LN2) collection scheme. We show how size, shape and porosity of resulting microspheres can be controlled by judiciously varying electrospray processing parameters and we demonstrate examples in which the particle size (and porosity) affect release kinetics. The effect of electrospray treatment on the particles and their physicochemical properties are characterized by scanning electron microscopy, confocal Raman microscopy, thermogravimetric analysis and mercury intrusion porosimetry. The microspheres manufactured here have successfully demonstrated long-term delivery (i.e. 1week) of an active agent, enabling sustained release of a dye with minimal physical degradation and have verified the potential of scalable electrospray technologies for an innovative TIPS-based microsphere production protocol.

Mid-infrared optical coherence tomography.

A time domain optical coherence tomography (OCT) system is described that uses mid-infrared light (6-8 microm). To the best of our knowledge, this is the first OCT system that operates in the mid-infrared spectral region. It has been designed to characterize bioengineered tissues in terms of their structure and biochemical composition. The system is based upon a free-space Michelson interferometer with a germanium beam splitter and a liquid nitrogen cooled HgCdTe detector. A key component of this work has been the development of a broadband quantum cascade laser source (InGaAs/AlInAs containing 11 different active regions of the three well vertical transition type) that emits continuously over the 6-8 microm wavelength range. This wavelength range corresponds to the so called "mid-infrared fingerprint region" which exhibits well-defined absorption bands that are specifically attributable to the absorbing molecules. Therefore, this technology provides an opportunity for optical coherence molecular imaging without the need for molecular contrast agents. Preliminary measurements are presented.