Datasets:

jgammack

/

MTL-abstracts

like 0

Modeling Discharge Pathways in Li-O2 Batteries to Optimize Capacity

Li-O2 batteries offer the possibility of storing twice the gravimetric energy density of Li-ion batteries. Li-O2 batteries operate by reacting oxygen with lithium ions in a non-aqueous solvent to form Li2O2 on a conductive cathode material. However, Li2O2 has poor electronic conductivity and passivates the electrode area. Achiev-ing high capacity requires careful attention to Li-O2 dis-charge mechanisms in order to optimize cathode void space filling by Li2O2.Li-O2 discharge occurs by two competing mechanistic pathways which are responsible for two possible morphologies of Li2O2 discharge product. The surface pathway involves two consecutive electron transfers to form a ~10 nm thin film of Li2O2. The solvent pathway involves the solvation of the reaction intermediate Li+-O2-, which then reacts in solution to form ~100 nm in diameter toroids of Li2O2. Since toroids allow for greater volumes of Li2O2 to form with less electrode area coverage, toroids are preferable to maximize capacity. However, the exact dependence of each pathway on different discharge conditions and solvent properties to promote toroid formation is not fully understood.Rotating ring-disk electrode (RRDE) experiments were performed to understand these pathway trends. A rotating rod creates convection currents that sweep reactants to the central disk electrode (Figure 1). Li2O2 film and soluble Li+-O2- are formed at the disk. Soluble Li+-O2- is swept to the ring electrode and oxidized, providing a measure of the relative size of the solvent pathway. By comparing ring and disk currents, the separate contribution of each discharge pathway can be determined.We then developed a model based on nucleation and growth of the Li2O2 film to explain potentiostatic discharge curves collected from RRDE experiments under different discharge conditions, such as varying solvent water content (Figure 2). The model demonstrates that high Li+-O2- solvent solubility inhibits the surface pathway and that this effect is primarily responsible for toroid promotion.

Multi-cell Thermogalvanic Systems for Harvesting Energy from Cyclic Temperature Changes

Technologies for the Internet of Things (IoT) are be-ing developed. An IoT network consists of large quan-tities of networked sensors that are often in remote or difficult to access locations, which drives the need for self-powered systems. Here, we come up with two types of multi-cell thermogalvanic systems that gener-ate electrical power through temperature cycles. The dual-temperature, dual-stack, self-powered electrochemical system is depicted in Figure 1. This dual-temperature system uses two identical electrochemical stacks, which can be a single battery or multiple batteries connected in series; however, each electrochemical stack is held at a different temperature. On the other hand, a single-temperature system works similarly, with the electrochemical stacks having similar operating potentials but oppositely signed temperature coefficients. Its operation is illustrated in Figure 2. Both systems can harvest energy from temperature cycles.We have tested both dual-temperature systems and single-temperature systems with different cathode/anode materials, load resistances, and frequencies of temperature cycles. The largest energy conversion efficiency was obtained from the dual-temperature experiment with two homemade LiCoO2/Li coin cells in which the cathodes with composition Li0.85CoO2 were cycled between 20 °C and 50 °C. The loads were two 100Ω resistors. The current is shown in Figure 3, and the efficiency was calculated to be 0.22%. This value is comparable to the efficiency obtained using charging-free thermally regenerative electrochemical cycles (TRECs), thermocapacitive cycles and ionic thermoelectric supercapacitors, but with more flexibility of material selection. In the meantime, we have also tested two single-temperature systems with four LiV2O5/Li-Al and three LiCoO2/Li cells, and one LiMnO2/Li-Al and one LiV2O5/Li-Al cell, respectively. Although the efficiency and power were still limited, they confirmed the feasibility of this concept. These systems can be further optimized by using materials with higher temperature coefficients and decreasing internal resistance at the same time.

Ising-Model-Based Computation by using Block Copolymer Self-assembly

Directed self-assembly of block copolymers can gener-ate complex and well-ordered nanoscale patterns for lithography. Previously, self-consistent field theory has been commonly used to model and predict the block co-polymer morphology resulting from a given template. In this work, we map block copolymer self-assembly onto an Ising model using two-dimensional post lattice template. We describe a simple and fast Ising-model-based simulation method for block copolymer self-as-sembly. With the Ising lattice setup, we demonstrate Ising-model-based logic gates.Figure 1 shows a diagram of the Ising lattice setup. To define the Ising lattice, we used a post lattice template with horizontal and vertical pitch equal to the equilibrium block copolymer periodicity, L0. After block copolymer processing, we defined a binary state, +1 or −1, between each adjacent pair of posts. We assigned +1 to a state when two adjacent posts were connected by a block copolymer structure, and −1 otherwise. The Ising Hamiltonian is given bywhere J’s and h’s were assumed to be independent of lat-tice location. We calculated the minimum Hamiltonian configuration using simulated annealing and compared the simulation results with previously reported results.To perform Ising-model-based computation, we encoded Boolean operations into the ground states of Ising lattices by designing specific Hamiltonians. Figure 2 shows a template design for a buffer where a boundary was defined by incommensurate double posts. Inside the boundary, an input state and an output state were defined. Prior to block copolymer processing, the input state was determined by the orientation of double posts while the output state was undetermined. After block self-assembly, the output state was set equal to the input state, performing the buffer operation.

Initiated Chemical Vapor Deposition (iCVD) of Cross-linked Polymer Films for the Directed Self-assembly of Block Copolymers

Directed self-assembly (DSA) of block copolymer (BCP) thin films is a promising approach to enable next-genera-tion patterning at increasingly smaller length scales. DSA uses a combination of physical and chemical constraints to force the BCP domains to self-assemble with the de-sired orientation with respect to the substrate. Physical constraints, such as holes and trenches, are formed using conventional lithographic techniques. Chemical con-straints, or wetting layers, are thin films that are either sandwiched between the BCP film/substrate interface or coated on top of the BCP film. These wetting layers ensure the pattern formed upon self-assembly has the appropriate orientation with respect to the substrate. Controllable chemistry combined with facile processing is key to the integration of these wetting layers. Here, we demonstrate that initiated chemical vapor deposited (iCVD) polydivinylbenzene (pDVB) ultra-thin films can direct the self-assembly of poly(styrene-block-methylmethacrylate) (PS-b-PMMA). iCVD allows for the simultaneous synthesis and formation of polymer thin films via a surface free radical polymerization. Additionally, methyl radicals formed at increased filament temperatures can change the chemical structure of the growing pDVB film in situ. By tuning the degree of backbone methylation, we systematically changed the wetting properties of iCVD pDVB from weakly PMMA preferential to complete PS preference. Conformal coatings of weakly preferential iCVD pDVB films on topographical line and space patterns produced self-assembled BCP films with both perpendicular orientation and long-range alignment (Figure 1a and 1b). Current research efforts aim to use iCVD pDVB films to enable contact hole shrinkage. To minimize the diameter of the template hole, the BCP should assemble with a central, cylindrical PMMA domain. Preliminary experiments examined the effects of strongly PS-preferential, conformal iCVD pDVB films on a contact hole shrink template. The dark spot within each hole in Figure 1c corresponds to the PMMA domain. These results indicate that iCVD pDVB films are a viable method to enable contact-hole shrinkage.

Geometry-dependent Properties of Synthetic Monolayer MoS2

Two-dimensional transition metal dichalcogenides (2-D TMDs) have shown great promise to be an ideal candi-date for post-silicon technology. Their atomic thickness-es and large carrier effective masses can offer excellent electrostatic gate control, a reduced source-to-drain leakage current, and a higher on-current in the ballistic regime, potentially enabling ultra-scaled devices, tunnel field-effect transistors, and ballistic transistors. Howev-er, the intricacy and diversity of the structural defects in 2-D TMDs significantly affect their electrical and optical properties, in either beneficial or detrimental ways. In the case of monolayer MoS2, several challenging issues including Femi level pinning at metal/MoS2 interface, unintentional n-type doping, and carrier scatterings, non-radiative excitonic recombinations, etc., have been attributed to a considerable amount of sulfur vacancy in monolayer MoS2. On the other hand, specific types of defects, if controlled carefully, also offers the access to engineer the nature of monolayer MoS2, such as channel polarity modification for realization of low-power MoS2–based CMOS integrated circuits, exciton reservoirs to prolong the excitonic lifetime for high-performance optoelectronic and photonic devices.This work explores the correlation between the domain geometries and the presence of different types of defects in monolayer MoS2 synthesized by chemical vapor deposition through transport and spectroscopy measurements. We show that the shapes of MoS2 domain can modulate the photoluminescence intensity and work function of MoS2 monolayers and the threshold voltage in the MoS2 field-effect transistors. Based on a two-defect-state model, the geometry-modulated behavior can be explained. This work not only offers a strategy to engineer the nature of MoS2 from the synthesis perspective, but also pave a path to realize low-power MoS2 CMOS integrated circuits.

Atmospheric Microplasma Sputter Deposition of Interconnects

We have preliminarily developed an apparatus that allows for the continuous, direct writing of interconnect-quality conductive lines. An atmospheric-pressure microplasma obviates the need for a vacuum while allowing for fine resolution imprints. We tested and characterized a novel focusing mechanism in which collisions with the working gas are harnessed to transfer electrostatic force to neutral sputtered atoms. This method compresses the deposit’s width in one dimension while expanding its length in the perpendicular dimension. We find that for an ideal set of parameters, the imprint is narrower than the sacrificial sputtering target (i.e., 9 µm wide imprint from a 50 µm diameter target). Other sets of parameters lead to other results, as computer simulation predicted, ranging from an unfocused spot 400 µm in diameter to a narrow line with 20:1 compression in the direction of focus, i.e., width, and 20:1 expansion in length (Figure 1), as compared to the unfocused spot.The microstructure of the deposit is of particular interest. As is typical of sputterers, the deposit could be smooth (55 nm roughness), and the resistivity can be as low as 1.1 µΩ·m (with no annealing). However, the resistivity greatly depends on the microstructure, which in turn depends on the deposition conditions. It is well known that sputtering at high-pressure results in a grain structure, as the early deposits shadow parts of the bare substrate, keeping sputtered material from fully coating the substrate. Traditionally, vacuum sputtering prevents this problem by allowing the sputtered material to impact the substrate normal to the surface; however, we sputter at atmospheric pressure, and thus, the sputtered material is redirected by random collisions. In our case, we use a combination of directed gas flow and electrostatic forces to prevent this shadowing effect (Figure 2).

Influence of TMAH Development on Niobium Nitride Thin Films

Patterning of superconducting thin films at the na-noscale has enabled numerous technologies used in signal detection and digital circuits. For instance, super-conducting nanowire single photon detectors (SNSPDs) and, more recently, the nanocryotron (nTron) both make use of the ability to pattern niobium nitride films at dimensions < 100 nm. Electron beam lithography of these devices often employs the negative tone resist hy-drogen silsesquioxane (HSQ) due to its high resolution and superior line edge roughness. Development of HSQ and adhesion promotion of HSQ to the substrate sur-face are both facilitated by tetramethylammonium hy-droxide (TMAH), making it an integral chemical in the fabrication process. However, despite the prevalent use of TMAH in patterning superconducting films, its influ-ence on the film itself has yet to be fully studied.Here we have investigated the effects of exposing NbN thin films to 25% TMAH. We show that TMAH modifies the surface chemistry of the film by reacting with the NbN to form niobium-based clusters, which are visible by scanning electron micrograph inspection (Figure 1). In addition to thinning the overall NbN film and reducing its critical temperature, the formation of niobium clusters creates a barrier to reactive ion etching in CF4, threatening the lithographic pattern transfer (Figure 2). While characterization such as FTIR has been employed to identify the compounds created by this reaction, future work is needed to study the mechanism through which the hexaniobate species interfere with the reactive ion etch chemistry.

Roll-to-Roll Transfer of Conductive Graphene Sheets

Graphene technology has been widely explored to pro-duce large sheets of conductive film to facilitate the manufacturing of flexible transparent photovoltaics. Monolayer-thick graphene has 97% transmittance in the visible regime and outstanding mechanical and electrical properties: that makes graphene suitable for transparent electrodes in order to replace the current state-of-the-art ITO electrodes, which are less flexible and are limited by the low indium supply on earth. However, scaling up the graphene manufacturing is tricky since it is typically grown on copper foils by chemical vapor deposition (CVD), and therefore, an ad-ditional transfer step is required to insert the graphene sheet into practical devices. The success of the transfer process is critical for the performances and the scal-ability of the graphene film. Given the compatibility with the manufacturing processes in organic and flexible electronics, we explore roll-to-roll (R2R) to enable the deployment of large area graphene on plastic substrates. We investigate how to avoid defects and fractures in the graphene film upon transfer. We scan over several options in order to figure out how the interplay of adhesion forces between the graphene and the host substrate works out. These investigations will advance the progress of the application of graphene in future flexible electronics.

Additive Manufacturing of High-temperature Compatible Magnetic Actuators

Various MEMS devices require large displacement and large force actuation to be efficient, such as miniature pumps. Magnetic actuation delivers large displacement and large force in a compact form factor. Additive man-ufacturing has recently been explored as a processing toolbox for MEMS; researchers have reported additive-ly manufactured microsystems with performance on par or better than counterparts made with standard microfabrication. In this work, miniature actuators are printed in pure Nylon 12 using the fused filament fab-rication method where a thermoplastic filament is ex-truded from a hot nozzle to create layer by layer a solid object. The actuators have embedded magnets that are not demagnetized by the heated nozzle (@ 250 °C) while being sealed in place midstream in the printing process.We have demonstrated the first miniature, additively manufactured, monolithic magnetic actuators compatible with high temperature (>200 °C) operation (Figure 1). The displacement of a 150 μm-thick, single-layer membrane actuator is characterized by various DC coil bias voltages, resulting in a maximum membrane displacement of 302 μm with 20V DC applied to the driving coil; in addition, the magnetic force is proportional to the square of the current drawn by the coil as expected from theory (Figure 2).

“Soft” Epitaxy in DNA-Nanoparticle Thin Films

The programmability of DNA makes it an attractive structure-directing ligand for the assembly of nanoparticle superlattices that mimic atomic crystallization. However, synthesizing multilayer single-crystals of defined size remains a challenge. This work studies growth temperature and interfacial energetics to achieve epitaxial growth of single crystalline nanoparticle thin films over arbitrarily shaped 500 × 500 μm2 areas on lithographically patterned templates. Both surface morphology and internal structure are examined to provide an understanding of particle attachment and reorganization (Figure 1).Importantly, these superlattices utilize a “soft,” elastically malleable building block, resulting in significant strain tolerance when subjected to lattice mismatch. Calculations of interaction potentials, small-angle X-ray scattering data, and electron microscopy images show that the oligomer corona surrounding a particle core can deform to store seven times more elastic strain than atomic films. DNA-nanoparticles dissipate strain both elastically through coherent relaxation of mismatched lattice parameter and plastically (irreversibly) through formation of dislocations or vacancies (Figure 2). Additionally, the DNA cannot be extended as readily as compressed, and thus, the thin films exhibit distinctly different relaxation behavior in the positive and negative mismatch regimes. These observations provide a more general understanding of utilizing rigid building blocks coated with soft compressible polymeric materials to control nano- and microstructure through “soft heteroepitaxy.”

Aligned CNT-based Microstructures and Nanoengineered Composite Macrostructures

Materials comprising carbon nanotubes (CNTs), such as hierarchical nanoengineered advanced composites for aerospace applications, are promising new materials thanks to their mechanical and multifunctional properties. We have undertaken a significant experimentally based program to understand both microstructures of aligned-CNT nanocomposites and hierarchical nanoengineered advanced composites macrostructures hybridized with aligned CNTs.Aligned nanocomposites are fabricated by mechanical densification and polymer wetting of aligned CNT forests. Here the polymer is typically an unmodified aerospace-grade epoxy. CNT forests are grown to mm-heights on 1-cm2 Si substrates using a modified chemical vapor deposition process. Following growth, the forests are released from the substrate and can be handled and infiltrated. The volume fraction of the as-grown CNT forests is about 1%; however, the distance between the CNTs (and thus, the volume fraction of the forest) can be varied by applying a compressive force along the two axes of the plane of the forest to give volume fractions of CNTs exceeding 20% (see Figure 1a). Variable-volume fraction-aligned CNT nanocomposites were characterized using optical, scanning electron (SEM), transmission electron (TEM) microscopy, 3-D TEM, and X-ray computed tomography (CT) to analyze dispersion and alignment of CNTs as well as overall morphology. Extensive mechanical property testing and modeling are underway, including 3-D constitutive relations and fracture toughness.Nanoengineered hierarchical composites hybridized with aligned CNTs are prepared by placing long (>20 μm) aligned CNTs at the interface of advanced composite plies as reinforcement in the through-thickness axis of the laminate (see Figure 2). Three fabrication routes were developed: transplantation of CNT forests onto pre-impregnated plies (“nanostitching”), placement of detached CNT forests between two fabrics followed by subsequent infusion of matrix, and in situ growth of aligned CNTs onto the surface of ceramic fibers followed by infusion or hand-layup. Aligned CNTs are observed at the composite ply interfaces and give rise to significant improvement in interlaminar strength, toughness, and electrical properties. Extensions of the CNT-based architectures to ceramic-matrix nanocomposites and towards multifunctional capabilities are being developed, including structural health monitoring and deicing.

Electrospray-printed Physical Sensor

Electrospray deposition (ESD) has recently gained at-tention as a manufacturing technology to develop novel nanostructured composites to produce low-cost micro- and nano-devices. ESD is also a remarkably versatile printing technique due to its capability to create ultra-thin films made from a great variety of liquid feedstock (e.g., suspensions of polymeric, dielectric, metallic par-ticles) that can be doped with organic nanostructures to modulate the physical properties of the imprint. No-tably, the resulting nanoreinforced composites might show enhanced transduction, which, in combination with printing on flexible substrates, might be relevant for exciting applications such as wearable biomedical devices.This project aims to develop an additively manufactured, low-cost, flexible physical sensor based on an ultrathin nanocomposite film doped with functionalized carbon nanostructures. The Taylor cone on an electrospray emitter fed with nanocomposite feedstock is shown in Figure 1a, while an electrospray-deposited imprint on a substrate is shown in Figure 1b. Essentially, this project is divided in (i) down-selecting and optimizing the formulation of the liquid feedstock, (ii) optimizing the fabrication of the ultrathin (~100 nm) nanostructured composite, and (iii) demonstrating a flexible physical sensor with transducing component made of the optimized nanostructured composite (see Figure 2).

Empirical Modeling of Copper Semi-additive Electro-chemical Plating

Semi-additive electro-chemical plating (ECP) is a common process for fabricating copper interconnects in many advanced packaging technologies, such as Wafer Level Integrated Fan Out (InFO) packaging. While cost efficient, this process suffers from thickness variations in the height of the plated copper. The most significant of these variations are layout dependent, where areas with dense interconnects plate slower than sparse areas (Figure 1). If left unchecked, these variations can lead to significant complications in later stages of the fabrication process, and ultimately to decreased electrical performance of the final packaged device. Previously, there were limited methods for predicting these variations, and foundries had to rely on experimentally determining which layouts would perform acceptably. Recently, we have developed a model that predicts these variations and allows errors to be predicted and corrected without the need to first fabricate the layout in question. While a model based on fundamental physics could in principle be developed to predict these variations, we instead develop an empirical model based on experimental data. This approach is well suited for many industrial applications, as empirical models can often be developed more quickly, without a significant loss of accuracy, and can be rapidly tuned or adapted to accommodate effects whose causes are uncertain. Our ECP model is divided into four stages as summarized in Figure 2. First, the effective pattern density of each point on the layout is determined with a learned spatial filter. These pattern densities are then mapped to effective conductances using a ratio-of-polynomials approximation. Next, these conductances are masked with the original layout, as photoresist prevents copper from growing in unwanted areas. Finally, the current flowing through each point of the wafer is solved for, and these currents are then converted to the plating height at each point in the layout.

150 °C Copper Bonding Technology with Graphene Interlayer

Bonding technology plays a significant role in elec-tronic packaging as it provides physical and electrical connections between semiconductor chips. Reliability of bonding joints affects the energy consumption and speed of an electronic system. Hence, it is important to have a reliable bonding technology. Copper (Cu) bonding technology is one of the most frequently-used bonding technologies nowadays. However, two critical issues have been limiting the reliability of lead-free Cu bond-ing technology: high bonding temperature (~260 °C) and aging degradation.We have devised a graphene-based Cu bonding technology that is of low bonding temperature and high reliability. By integrating nanoscale graphene/Cu composite on the Cu substrate prior to thermocompression bonding, Sn-Cu joints can be fabricated at a bonding temperature as low as 150 °C, which is the lowest reported value to date for Cu bonding technology. Specifically, we electrochemically deposit a layer of Cu nanocone array on the Cu substrate and cover it with a graphene sheet, prior to the bonding process. When subjected to heat, microscale Sn solder deforms and replicates the Cu nanocone array morphology, and hence transforming into nanoscale Sn. Compared to microscale Sn, nanoscale Sn has much lower melting points and facile surface diffusion. This phenomenon effectively contributes to the low bonding temperature observed in our bonding technology. The presence of the graphene layer prevents the formation of Cu-Sn intermetallic compounds thus significantly slows down the aging degradation. With the advancement in graphene synthesis and transfer technology, we anticipate the graphene-based Cu bonding technology presented in this work can be integrated into the existing commercial Cu bonding technology for industrial applications in the near foreseeable future.

Chemical Vapor Deposition of Multiple Transition Metal Disulfides in One Synthesis Step

Recently, transition metal disulfides (TMD) have received tremendous attention due to their exceptional optical and electrical properties. Many techniques have been explored to obtain monolayer TMD and chemical vapor deposition (CVD) synthesis using transition metal oxide, and chalcogenide solid precursors is the most common method used in laboratories now. However, the quantity of solid precursors used is usually surplus giving rise to chemical reactions between precursors in each of their crucibles, as a result of precursors’ diffusion at growth temperature. Hence, a CVD setup is normally dedicated for the growth of only one type of TMD to avoid cross-contamination (except for hetero-structures synthesis), and it is impossible to grow multiple monolayer TMD in one synthesis step. Here, we report a new technique to synthesize MoS2 and WS2 monolayer films in one CVD process. We first disperse a minuscule amount of metal oxide precursor on targeted substrates, which were then loaded to the furnace in slanting position, rather than horizontal, followed by a sulfur annealing to concurrently grow monolayer MoS2 and WS2 on separate substrates. The synthesized TMD films exhibit good properties as confirmed by Raman, PL, XPS, STEM analyses, and electrical measurements.

A Nanofabricated, Path-separated, Grating Electron Interferometer

Recent progress in focused-ion-beam (FIB) technol-ogy has enabled the fabrication of electron optical elements such as zone-area plates, phase plates, and beamsplitters. These nanofabricated elements can be used to perform Zernike phase-contrast imaging, ho-lography and beam aberration correction in a conven-tional transmission electron microscope (TEM). We have fabricated a grating-Mach-Zehnder-electron-in-terferometer, using FIB milling of a single-crystalline silicon workpiece. As shown schematically in figure 1(a), the interferometer uses two thin layers of silicon as dif-fraction gratings; the first to split the incident electron beam, and the second to recombine two of the diffract-ed beams. The gap between the gratings in our interfer-ometer was 20 µm. Fabrication of the gratings from a single crystalline silicon workpiece ensures alignment and precise positioning. We obtained a rotational align-ment of ~100 µrad and a grating positioning accuracy of 100 nm. Figure 1(b) is a scanning electron micro-graph of this interferometer. We inserted this interferometer in the sample holder of a 200 kV TEM (JEOL 2010F). We used an electron beam with a diameter of 240 nm on the first grating and convergence semi-angle of 4 mrad in our experiment. As shown in figure 2(b), when imaging the second grating (figure 2(a), sample z-height z1) at high-resolution (Ψ0), we obtained a lattice-resolved image of silicon. As we raised our sample holder z-height to move the imaging plane below the second grating (z2-z5), the first-order diffracted beam from this grating (Ψ0g) moved closer to the first-order diffracted beam from the first grating (Ψgg), and the two beams overlapped 20 µm below the second grating (z6). Figure 2(c) is a high-resolution image of the overlapping beams, showing interference fringes of period 0.32 nm, which was expected from the interference of first-order silicon diffracted beams. This interferometer could be used to perform electron holography in any TEM, as well as interaction-free imaging using the Elitzur-Vaidman scheme.

A Scheme for Low-dose Imaging via Conditional Sample Re-illumination

Recently, several electron-beam-based low-damage im-aging schemes for radiation-sensitive samples (such as proteins and biomolecules) have been investigated. It is now possible to incorporate a Mach-Zehnder inter-ferometer (MZI) in a standard transmission electron microscope (TEM) to perform Elitzur-Vaidman Inter-action-free imaging (IFI). We are theoretically inves-tigating the performance of an MZI-based IFI with a Poisson source. We combined IFI with a conditional re-illumination scheme that reduced the probability of imaging errors at low illumination doses.As a first step, we considered imaging of purely black-and-white pixels. As shown in figure 1, we considered two schemes: classical and IFI, with various imaging detectors. We quantified error as the probability of incorrectly inferring the transparency of a pixel (Perr), and damage as the mean number of electrons that scatter off a black pixel (ndamage), respectively. At the start of our calculations, we assumed a prior probability q of a given pixel being black. Then, we found expressions to update q based on the electron detection statistics, assuming a Poisson beam with mean λt. If the value of q was within a pre-defined minimum acceptable error threshold ∈, we made an inference on whether the pixel was black or white. If this condition was not met, we re-updated q using a second round of detection statistics. This process was repeated a maximum of Nmax times. Figure 2 shows the results of imaging utilizing conditional re-illumination, for both classical and IFI. These results were calculated with Nmax=1 (circles with dotted line) and Nmax=50 (crosses with dashed lines) illuminations. For both schemes, conditional re-illumination offered a reduction in ndamage at 50 illuminations as compared to single-stage illumination. For classical imaging, Ndamage was reduced to 1, and for IFI, ndamage saturated to 0.67, at Nmax=50.We are now working on extending these calculations to semi-transparent samples, as well as implementing this illumination scheme in a scanning TEM.

Towards Dislocation-free GaN

The performance of advanced GaN-based electronics and optoelectronics can rely heavily on the structural quality of the epilayer used in its fabrication. The lay-er’s characteristics, such as dislocation density or sur-face roughness, are largely inherited from the initial GaN growth. Due to the limited availability and the cost of high-quality bulk GaN substrates, heteroepitaxy of GaN on foreign substrates such as Al2O3, SiC, and Si is conventionally used. The lattice and thermal-expan-sion-coefficient mismatch of these substrates to GaN unavoidably lead to the formation of dislocations, as well as potential cracks and wafer bow. In addition, the majority of the substrate material is usually removed from state-of-the-art devices to lower the thermal resistance of the packaged devices and improve performance. The removal of GaN devices from bulk/foreign substrates is very challenging and is an ongoing subject of research. Existing removal processes involving photoelectrochemical etching, mechanical spalling, and laser interface decomposition suffer from slow processing speed and/or significant surface roughening and cracking, limiting the process yield and practicality of substrate reusing.Recently, we discovered that the epitaxial registry of adatoms could be determined by the underlying substrate remotely without direct contact with the substrate, but through a narrow gap defined by monolayer graphene. Therefore, homoepitaxial growth can be performed remotely through the single-atom-thickness gap, with the dislocation density of the epitaxial thin film at the same level as the high-quality substrate. In addition, because of the van der Waals interaction at the graphene interface, the epitaxial thin film can be precisely and rapidly exfoliated from the substrate, demonstrating the atomic flatness at the released surface mimicking the morphology of graphene surface. We performed the remote epitaxy of GaN on GaN/sapphire substrate with monolayer graphene as an interlayer to demonstrate high-quality, low dislocation density GaN thin films. We obtained GaN epilayer with material quality identical to the GaN/sapphire substrate in terms of surface morphology and dislocation density. We further exfoliated the GaN epitaxial thin film from the substrate achieving free-standing GaN of 300nm thick.Ultimately, we will develop the process of GaN remote epitaxy on bulk GaN substrate with minimal defects, enabling the GaN-based electrical and optoelectronic devices approaching intrinsic performance without the limitation from material quality. On the other hand, the cost of such high-performance devices will be significantly reduced since expensive substrates will be reused.

Fabrication of Small-pitch Gratings for Smith-Purcell Radiation from Low-energy Electrons

Swift-moving electrons carry evanescent near-field, which can be coupled to far-field radiation when the electrons move closer to a periodic structure and in parallel to the periodic structure plane. This effect was named after Smith and Purcell, following their first experimental demonstration of the effect. The wavelength of Smith-Purcell radiation depends on the grating pitch and the electron energy. Here, we demon-strate Smith-Purcell radiation in the optical regime by using gratings with 50-60 nm pitch and electrons with 1.5-6 keV kinetic energy. These results have potential applications in tunable nanoscale light sources.Our gratings were fabricated on gold-coated silicon substrates. The 200-nm-thick gold coating layer was used to suppress cathodoluminescence from silicon. The grating patterns were defined using electron beam lithography in PMMA resist, followed by 0 °C cold development in 3:1 IPA:MIBK. 20 nm gold was then deposited via electron-beam evaporation and lifted-off in hot NMP. Figure 1 shows an SEM image of a 50-nm-pitch grating.To measure Smith-Purcell radiation, the grating samples were mounted inside a modified SEM with an optical attachment to collect the radiated light and measure its spectrum. Electrons with 1.5-6 keV kinetic energy were used to induce the Smith-Purcell radiation. Figure 2 shows the measured Smith-Purcell radiation spectra from a 50-nm-pitch grating using electron beams with different kinetic energies. The peaks of the radiation spectra match well with the theoretical predictions (vertical dashed lines). We demonstrate the Smith-Purcell radiation wavelength decreases as we increase the electron kinetic energy or decrease the grating pitch.

Remote Epitaxy through Graphene for Two-dimensional Material Based Layer Transfer

Van der Waals epitaxy (vdWE) has gained great inter-est for crystalline growth as it substantially relaxes the strict lattice matching requirements in conventional heteroepitaxy and allows for facile layer release from the vdWE surface. In recent studies, vdWE was inves-tigated on two-dimensional (2-D) materials grown or transferred on arbitrary substrates, with the primary notion that the 2-D material is the sole epitaxial seed layer in vdWE. However, the underlying substrate may still play a role in determining the orientation of the overlayers since the weak vdW potential field from 2-D materials may barely screen the stronger potential field from the substrates. Here, we reveal that the epitaxial registry of adatoms during epitaxy can be assigned by the underlying substrate remotely through 2-D materials by modulating the interaction gap between the substrate and the epilayer. Our study shows that remote epitaxial growth can be performed through a single-atom-thick gap defined by monolayer graphene at the substrate-epilayer interface. Simulations using density functional theory (DFT) prove that remote epitaxy can occur within a ~9 Å substrate-epilayer gap. We experimentally demonstrate successful remote homoepitaxy of GaAs(001) on GaAs(001) substrates through monolayer graphene (Figure 1). Characterization by high-resolution scanning transmission electron microscopy (HRSTEM) confirms single crystalline growth of GaAs film through graphene with an interaction gap of 5 Å below the critical limit outlined by the simulation. The concept of remote homoepitaxial growth is further extended to other compound semiconductors such as InP, GaP, GaN, as well as functional oxides, SrTiO3, and fluoride material systems, LiF (Figure 2). Following the growth, the single-crystalline films are rapidly released from the vdW surface of graphene to provide large-scale, single-crystalline films. This concept, here termed 2-D material based layer transfer (2-DLT), suggests a universal method to copy/paste epitaxial films of any material systems based on the underlying substrates through 2-D materials then rapidly release and transfer to substrates of interest. The potential to reuse graphene-coated substrates suggests 2-DLT will greatly advance non-Si electronics and photonics by displacing the high cost of non-Si substrates.

Controlling Concentration and Nature of Oxygen Defects in Layered Cuprate-based Materials by Electrical Bias

Both the nature and concentration of oxygen defects in oxide materials can have a significant impact on their physical and chemical properties, as well as key interfa-cial reaction kinetics such as oxygen exchange with the atmosphere. Most commonly, the desired oxygen defect concentration, or equivalently oxygen nonstoichiom-etry, is attained in a given material by controlling the oxygen partial pressure and temperature in which it is equilibrated or annealed. This approach, however, is lim-ited by the range of oxygen partial pressures readily ex-perimentally achievable and requires knowledge of the applicable defect chemical model. In this study, we fine-tune oxygen defect concentrations in promising rare earth cuprate (RE2CuO4: RE = rare earth) solid oxide fuel cell (SOFC) cathode materials by application of electrical potentials across a yttria-stabilized zirconia (YSZ) supporting electrolyte. These layered perovskites can incorporate both oxygen interstitials and vacancies, thereby broadening the range of investigations. Here, we show a strong correlation between oxygen nonstoichiometry values (which are determined by in situ measurement of chemical capacitance) and oxygen surface exchange kinetics (which is inversely proportional to the area-specific-resistance). Both types of oxygen defects – interstitials and vacancies – dramatically enhance surface kinetics. These studies are expected to provide further insight into the defect and transport mechanisms that support enhanced SOFC cathode performance.

Coherent Soft X-ray Imaging of Magnetic Nanotextures

The ability to image the nanoscale structure of materi-als with tunable magnetic textures is pivotal for the de-velopment of low-power and nonvolatile data storage technologies. Soft X-ray imaging has emerged in the last decade as a powerful and accurate methodology to resolve the bulk domain structure of several magnetic materials — magnetic multilayers, buried interfaces, or skyrmion lattices — as well as nanoelectronic devices under operating conditions. Soft X-ray imaging relies on two main requirements: (i) the ability to focus a collimated X-ray beam on a spot the size of a few tens of nm and (ii) the ability to scan the focused X-ray beam with nm precision. We have commissioned a new soft X-ray nanofocusing setup installed at beamline CSX-1 of the National Synchrotron Light Source II. The schematics of this setup are shown in Figure 1. A key element is the Fresnel zone plate (inset D), which acts as a diffractive phase mask to focus X-rays to a 70-nm spot at the sample, and is fabricated using e-beam lithographic tools. The beam spot can be moved with the aid of piezo-based nanopositioners (inset C), which translate the X-ray optics while keeping the sample in a fixed position. Diffracted X-rays are collected with a CCD camera in the far field (~30 cm from the sample). The resulting magnetic scattering intensity encodes local antiferromagnetic strength and can be acquired in less than a second. By scanning the X-ray beam across the sample, we are able to probe the spatial distribution of antiferromagnetic order.We applied this new method to the study of antiferromagnetic rare earth NdNiO3. In particular, and for the first time, we unveil the inhomogeneous nature of the spin-ordered ground state (inset B). Furthermore, we identify the spatial distribution of antiferromagnetic domains and show that it follows a scale-free distribution. Our future focus is to extend our studies to the imaging of nanoscale magnetic textures in antiferromagnetic spintronic materials and devices, including in operando studies as a function of applied current.

Electro-Chemo-Mechanical Studies of Perovskite-Structured Mixed Ionic-electronic Conducting SrSn1-xFexO3-x/2+δ

High efficiency and fuel flexibility make solid oxide fuel cells (SOFCs) attractive for conversion of fuels to electricity. Reduced operating temperatures, desirable for reduced costs and extended operation, however, result in significant losses in efficiency. This loss has been traced primarily to slow cathode surface reaction kinetics. In this work, we extend previous studies on the promising mixed ionic and electronic conducting perovskite-structured SrTi1-xFexO3-x/2+δ (STF) materials system whose exchange kinetics were correlated with the minority electron charge density by replacing Ti with Sn, due to its distinct band structure and higher electron mobility. Oxygen nonstoichiometry and the defect chemistry of the SrSn1-xFexO3-x/2+δ (SSF) system were examined by thermogravimetry as a function of oxygen partial pressure in the temperature range of 973-1273 K. Marginally higher reducibility was observed compared to corresponding compositions in the STF system. The bulk electrical conductivity was measured in parallel to examine how changes in defect chemistry and electronic band structure, associated with the substitution of Ti by Sn, impact carrier density and ultimately electrode performance. Bulk chemical expansion was measured by dilatometry as a function of oxygen partial pressure, while surface kinetics were examined using AC impedance spectroscopy. The electrochemical properties of SSF were found not to differ significantly from the corresponding composition in STF. Though slightly shifted by the larger size of Sn, the defect equilibria and the cathode area specific resistance differed only in a limited way from that in STF. This was attributed to properties being largely dominated by Fe and not by the substitution of Ti with Sn. However, due to asymmetry in the crystal structure caused by the larger size of Sn, both thermal and chemical expansion coefficients of SSF35 were found to be around 20% and 10% lower than those of STF35, thus making SSF35 much more chemo-mechanically stable in SOFC operating conditions.

Experimental Characterization and Modeling of Templated Solid-state Dewetting of Thin Single-crystal Films

Solid-state dewetting is a physical phenomenon that dis-integrates a continuous film into islands when the film is heated above a characteristic dewetting temperature but kept well below its melting temperature. It is driven by surface energy minimization and mediated via sur-face diffusion of atoms. Solid-state dewetting has been thought of as an issue in microelectronics, however, it has also demonstrated its potential as a simple patterning method that can be used to generate a complex and reg-ular array of micro- and nano-sized structures in a highly reproducible way [Figure 1a, reference 1]. It starts either from edges of the film or in the continuous flat region by forming a natural hole. Various instabilities that develop at retracting edges have been understood via modeling and experimenting over the past years, including “pinch-off,” “corner instability [reference 2],” and “Rayleigh-like in-stability’[reference 3].” The fingering instability as shown in Figure 1b, which is another instability that creates wire-like structures at retracting edges, is our current focus. Through experiments, we have found conditions that lead to the fingering instability and have learned that spacing between fingers can be controlled via templating of film edge. We have also found that controlling the period of the fingering process affects the kinetics of the fingering, and we have developed an analytical model that predicts a relationship between the retraction rate and finger period. This model agrees well with experimental results. Our increased understanding of the various instabilities at retracting edges can be used to design templates that will lead to specific complex structures during solid-state dewetting.However, before we can fully exploit our understanding of templated solid-state dewetting to make designed structures, we must understand natural hole formation in thin films. In polycrystalline films, grain boundary triple junctions facilitate hole formation in a well-understood way, but the formation of holes in single-crystal films (Figure 2) is not well understood. Studying this phenomenon is critical because holes create new edges from which the film retracts. Furthermore, thinner single-crystal films develop more natural holes per unit area, and the growth of these holes can come to dominate the overall reduction of film surface area. Unsuppressed natural hole formation interrupts edge retraction modes that were intentionally patterned to create a specific structure. If controlled, however, the formation of holes could be used to pattern periodic nanostructures that span large length scales, up to several centimeters. In parallel with studying the fingering instability, we are currently working with both Ni films on MgO substrates and Ru films on sapphire substrates to identify and understand the causes of natural hole formation in single- crystal films. By understanding these mechanisms, we aim to develop templated solid-state dewetting into a powerful and cost-effective method for producing nanostructures.

Optofluidic Real-time Cell Sorter for Longitudinal CTC Studies in Mouse Models of Cancer

Circulating tumor cells (CTCs) play a fundamental role in cancer progression. However, in mice, limited blood volume and the rarity of CTCs in the bloodstream preclude longitudinal, in-depth studies of these cells using existing liquid biopsy techniques. We developed an optofluidic system (Figures 1, 2) that continuously collects fluorescently labeled CTCs from a genetically engineered mouse model (GEMM) for several hours per day over multiple days or weeks. The system is based on a microfluidic cell-sorting chip connected serially to an unanesthetized mouse via an implanted arteriovenous shunt. Pneumatically controlled microfluidic valves capture CTCs as they flow through the device, and CTC-depleted blood is returned back to the mouse via the shunt. To demonstrate the utility of our system, we profile CTCs isolated longitudinally from animals over four days of treatment with the BET inhibitor JQ1 using single-cell RNA sequencing (scRNA-Seq) and show that our approach eliminates potential biases driven by inter-mouse heterogeneity that can occur when CTCs are collected across different mice. The CTC isolation and sorting technology presented here provides a research tool to help reveal details of how CTCs evolve over time, allowing studies to credential changes in CTCs as biomarkers of drug response and facilitating future studies to understand the role of CTCs in metastasis.

Continuous Online Monitoring of Biologics Quality during Continuous Biomanufacturing using Micro/Nanofluidic System

The growing trend in the biopharmaceutical industry is to adopt continuous biomanufacturing to reduce manufacturing cost and improve product quality. How-ever, several challenges must be solved. First, a reliable and efficient cell retention device is required. Currently, using a hollow fiber membrane is a widely adopted cell retention method in industry to maintain suspended cells in the bioreactor and remove biologics from the bioreactor. However, it suffers from membrane fouling/clogging due to cells and cell debris. Moreover, product recovery efficiency becomes significantly low over cul-tivation time, resulting in low manufacturing efficien-cy.Second, there is no robust online sensor for critical quality attributes, such as purity and binding affinity, during manufacturing to understand the real-time relationship between the critical quality attributes and bioprocesses. For example, sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), size exclusion chromatography (SEC), and capillary gel electrophoresis (CGE or CE-SDS) are commonly used to check protein purity, but they offer only at-line/offline discontinuous analysis. In this context, we developed a novel micro/nanofluidic system to demonstrate continuous online monitoring of protein size distribution of cell culture supernatant during perfusion culture (Figure 1). The system consists of perfusion culture, online sample preparation, and detection of protein size distribution. To enable long-term perfusion culture, we used a membraneless microfluidic cell retention device. The cell retention is based on size-based cell sorting. Its high cell-concentration-capacity (>40E+6 cells/mL), scalability, long-term biocompatibility, and high product recovery efficiency have already been demonstrated. The online sample preparation consists of buffer-exchange, cell clarification, protein labeling, and denaturation. At the end of the system, the nanofluidic device continuously monitors protein size distribution. It has nanofilter array and supports continuous-flow size-based protein separation and concentration.

Electrokinetic Purification for DNA Analysis

Rapid detection of ultralow-abundance pathogenic DNAs in complex clinical samples, which are often as low as <100 copies/ml (~0.1 aM), is critical for early diagnosis of infectious diseases. Despite the unprece-dented amplification capacity of the polymerase chain reaction (PCR), its detection sensitivity and specificity are limited by the ability to purify DNAs from clinical samples with minimal loss. Currently, DNA extraction relies on slow and labor-intensive spin column-based solid phase extraction, which introduces significant losses of DNAs during capture, elution, and final sam-pling, especially for ultralow-abundance samples.Ion-concentration-polarization (ICP)-based electro-kinetic trapping (ET) has attracted much attention in the past decade as a viable approach for the rapid concentration of DNAs and other biomolecules, with enrichment speeds of ten-thousand-fold in ∼10 minutes. Based on this technique, we report the direct enrichment and purification of DNAs in complex biological samples by pressure-modulated selective electrokinetic trapping (PM-SET). We showcase the utility of PM-SET in human serum that contains 60–80 mg/mL total serum protein and perhaps represents one of the most complex backgrounds for molecular detection. Through modulating the hydrostatic pressure applied to the ICP-based ET device, we demonstrate the selective trapping of DNAs (of high electrophoretic mobility) while the majority of background proteins (of low electrophoretic mobility) are simultaneously removed (Figure 1), achieving an enrichment factor of >4800 in 15 minutes for DNAs.With these advantages, we believe that PM-SET could potentially play an enabling role in developing lab-on-a-chip devices toward point-of-care diagnostics, on-site food and environment monitoring, and a variety of other applications in resource-limited settings.

Multi-parameter Cell-tracking Intrinsic Cytometry for Characterization of Single Cells

Cells possess biochemical properties that require ex-trinsic tags, e.g., fluorescent dyes, for detection and biophysical properties, e.g., morphological, mechanical, electrical, and optical properties, which are intrinsic and do not require any labels. While extrinsic labeling techniques are highly specific to cell states, analyses of label-free biophysical properties are more suitable for applications, which require quick turnaround, and for analysis where biochemical labels for targeting cell states are not known. Development of single-cell biochemical analysis techniques with high sensitivity, throughput, and multiplexing capability has advanced understanding of complex biological systems and has established their presence in biological research labs and clinical practice. In contrast, development of sin-gle-cell biophysical analysis techniques is often limited to proof-of-concept due to the low-specificity nature of the intrinsic markers and the lack of approaches to combine multiple biophysical assays for high-dimen-sional biophysical phenotyping of single cells.To address this challenge, we propose a general approach to combine multiple biophysical measurements of single cells via optical tracking. To show specific implementation of this approach, we developed a microfluidic platform that measures up to five intrinsic markers of single cells, including size, deformability, and polarizability at three frequencies (Figure 1). We chose these intrinsic markers because each has been associated with important biological functions and proven useful for cell characterization, and they are rarely studied together. Cell tracking was demonstrated on the fully integrated platform, and multiple intrinsic markers of single cells were measured from cell samples treated with varying concentrations of actin polymerization inhibitor. An unsupervised dimensionality reduction technique, viSNE, was implemented to visualize the five-dimensional intrinsic marker measurements in two-dimensional visualization (Figure 2). Our analysis showed that an increase in number of intrinsic markers measured by our intrinsic cytometry platform resulted in an increase in classification accuracy of cell states induced by drug treatment.

Point-of-Care Biomarker Detection through Electronic Microfluidics

Identification of protein biomarkers is a vital step for numerous biomedical applications including clinical diagnostics, monitoring, and treatment. However, tra-ditional blood analysis techniques require large sam-ple volumes and centralized laboratories with trained technicians to perform tests. This translates to long wait times (~days) for patients and healthcare provid-ers to receive testing results. Point-of-Care (PoC) de-vices have emerged as promising alternatives to these traditional blood assays as they are capable of rapid analysis (~mins) in non-laboratory settings. Thus, we are developing an integrated and electronically operat-ed PoC platform for the rapid identification of protein biomarkers. As shown in Figure 1, in principle, the PoC system is a blood-to-result platform that incorporates an interface for automated sample withdrawal from blood collection devices, an electrochemical assay, and an electronic readout. The electrochemical assay was developed as a bead-based electronic enzyme-linked immunosorbent assay (ELISA) to reduce assay time, lower required sample volume (~µL), and enable platform automation. The workflow of this bead-based electronic assay is shown in Figure 2. Following sample infusion, magnetic microbeads conjugated with antibodies and enzymes are introduced to the sample. Target biomarkers bind to antibodies on the surface of the microbeads and are then sent to electrodes for detection. Biomarker-bound beads then attach to capture antibodies coated on the electrodes. Following an incubation period, enzymes on the attached microbeads catalyze chemical reactions to generate current, which is measured electronically. Electrical readouts are then correlated with the target biomarker concentrations on the transducer. Results have indicated that this sensor has the sensitivity range required for clinically relevant concentrations of various biomarkers for a variety of biomedical applications. Furthermore, initial testing has shown that the platform produces rapid results (within 30 mins) using small volumes (~µL) of blood.

A Microfluidic System for Modeling Human Atherosclerosis and Pathophysiology

Hemodynamic flows and consequent fluid shear stress-es (FSS) directly regulate endothelial function (EF), which in turn regulates atherosclerotic disease pro-gression (atherogenesis). Laminar, helical flows with a high-magnitude pulsatile FSS waveform enhance EF (i.e., are atheroprotective), while multidirectional flows with a low-oscillatory FSS waveform impede EF (i.e., are atheroprone). Understanding atherogenesis re-quires a microenvironment with representative flows regulating EF (Figure 1). Current systems usually pro-vide a single aspect of atheroprotective or atheroprone flows: high/low shear, oscillatory/unidirectional flow, uniform/pulsatile flow, etc., but do not recreate all the spatiotemporal flow features to mimic the complexity of physiological flows.We have developed a microfluidic system that, for the first time, fully recapitulates in vivo-like spatiotemporal atheroprotective flow simultaneously with atheroprone flows, both with complex but programmable features. Applying these flows upon primary human endothelial cells (hECs), we can concurrently monitor maintenance of EF and the emergence of endothelial dysfunction in precise locations within a single cellular monolayer, as it occurs in vivo. We utilize on-chip valves to dynamically modulate flows—and hence FSS applied on cells—mimicking in vivo waveform dynamics and magnitude. Additionally, we utilize patterned grooves within the device to impart specific spatial profiles of flow, enabling us to recapitulate the complete spatiotemporal flow signatures found in vivo.Our platform allows us to monitor hECs cultured under spatiotemporal flows and execute relevant biological assays for assessing EF. As an example, we observe cell alignment exclusively under atheroprotective flows compared to atheroprone flows, matching known in vivo morphology of functional hECs (Figure 2). Overall, with this highly relevant platform, we can, for the first time, systematically and simultaneously control unexplored hemodynamic flow parameters that condition hECs to regulate human disease susceptibility.

Biochip for Drug Delivery using TERCOM

Targeted drug delivery has been an area of active investigation for several decades. Most approach target cell-borne receptors chemically or genetically. Some use external stimuli such as heat or radio waves to drive spatially-localized release. In one approach, particles estimate their own location within the body by correlating their sensed environment (e.g., temperature, pressure, salinity, sugar levels, pH, etc.) or its time history against a carried map and releases a charge of a drug based on this estimate. This eliminates external aids and is closely related to terrain contour matching (TERCOM) and scene correlation (DSMAC), techniques used in aircraft navigation. Previous work by the PI and his group focused on the development of nanoparticles capable of sensing and retaining a memory of their environment with noisy DNA. Current efforts focus on the theory of estimating location within the body from vectors of sensed variables and on development of a SiO2 MEMS biochip (microarray) that can test or screen particles and molecules for such sensitivity. Preliminarily explored particle concepts have included liposomes and proteins (bottom-up fab) and thin films (top-down fab). A chip concept that implements a microarray with a half-toned chemical library and material data drawn from conventional surgical analogs has also been considered. The objective is to demonstrate a targeted nanoparticle that implements TERCOM- or DSMAC-like navigation in the body and a biochip that can evaluate its selectivity. The concept is outlined in Figure 1.

Nanocone-arrayed SERS Substrate for Rapid Detection of Bacterial Sepsis

Rapid detection of bacteria is a very critical part of treating infectious disease. Sepsis kills more than 25 percent of its victims, resulting in as many as half of all deaths in hospitals before identification of the patho-gen for patients to get the right treatment. Raman spectroscopy is a promising candidate in pathogen di-agnosis, given its fast and label-free nature if the con-centration of the pathogen is high enough to provide reasonable sensitivity. This work develops a new kind of surface-enhanced Raman spectroscopy (SERS) sub-strate that will provide high enough sensitivity and fast and close contact of the target structure to the hot spots for an immunomagnetic-based, bacteria-concen-trating and -capturing technique. The substrate uses an inverted cone structure array made of transparent PDMS to funnel the light to the top of the cones, where plasmonic nanorods are located. A high-reflective and low-loss layer is deposited on the outer surface of the cone. Given the geometry of the cone, photons are multi-reflected by the outer layer and thus the number density of photon increases by at least an order. After the pattern and geometric shape of the cones are optimized, the hot spots of the proposed SERS substrate could have an enhancement factor of 108 or higher, which could be high enough to detect immunomagnetically densified bacteria.

Arterial Blood Pressure Estimation using Ultrasound Technology and a Transmission Line Arterial Model

This work describes the application of a transmission line model to arterial measurements in order to derive useful cardiovascular parameters. Non-invasive ultrasound tech-niques are used to make these measurements, which has several benefits over invasive methods such as arterial catheterization. However, invasive methods are seen as the “gold standard” measurements and therefore the most ac-curate. Having accurate measurements performed non-in-vasively is very desirable for cardiologists to determine their patients’ risk of developing cardiovascular disease.This work details how to obtain the “blood” flow and pulse pressure waveforms with ultrasound transducers using a flow phantom with blood mimicking fluid (BMF) shown in Figure 1. Two transducers, one for imaging and one for Doppler, are used together to derive these pulse pressure waveforms from distension and “blood” flow velocity measurements. Unfortunately, the pulse pressure waveform does not contain diastolic pressure information. By decomposing the backward and forward pulse and flow waves and using the transmission line model, the diastolic pressure can be determined, yielding a complete arterial blood pressure waveform.

Human Subject Studies of Ultrasound for Continuous and Non-invasive Arterial Blood Pressure Waveform Monitoring

Arterial blood pressure (ABP) is a key physiological parameter for evaluating the circulatory system of pa-tients. The ABP reflects the pathophysiologic states of the cardiovascular system. Currently, the ABP wave-form is usually obtained via an arterial line (A-line) in intensive care settings; while considered the gold stan-dard, the A-line is invasive. Thus, our goal is to develop a reliable, continuous, and non-invasive ABP waveform estimation system. Ultrasound is an ideal imaging mo-dality to achieve this goal due to its low cost and por-tability. Two human subject studies are in progress us-ing prototype ultrasound devices to develop this ABP waveform estimation system.The first human subject study is being done in collaboration with the Boston Medical Center to compare the measured ABP waveform on patients with A-lines with the pulse pressure waveform measured with the Flow Method we developed at the carotid artery. In the Flow Method, the blood flow is measured with pulsed Doppler using a single ultrasound transducer while the arterial area and distention are measured by using M-mode imaging with a second single ultrasound transducerA drawback of the Flow Method is that it provides only the pulse pressure waveform rather than the absolute ABP waveform. Thus, a transmission line model of the arterial blood flow system is being developed to make an estimation of the diastolic pressure, which provides the baseline for the absolute ABP waveform. The pulse pressure waveform on its own gives no information on diastolic blood pressure. However, the transmission line model suggests that the waveform may contain information regarding the patient’s vascular resistance. By decomposing the waveform into the forward and backward traveling waves, we can derive the reflection coefficient. The reflection coefficient provides an estimate of the vascular resistance, which is multiplied with the measured diastolic blood flow to yield the diastolic pressure. The second human subject study is underway in collaboration with Massachusetts General Hospital to compare the measured ABP waveform on patients with A-lines with mean arterial pressure (MAP) calculated by the transmission line model at the brachial arteries.

Measuring Saccade Latency using Smartphone Cameras

With current clinical techniques, it is difficult to ac-curately determine the condition of a patient with a neurodegenerative disease (e.g., Alzheimer’s disease). The most widely used metrics are qualitative and vari-able, exposing the need for a quantitative, accurate, and non-obtrusive metric to track disease progression. Clinical studies have shown that saccade latency--an eye movement measure of reaction time--can signifi-cantly differ between healthy subjects and patients. We propose a novel system that measures saccade latency outside the clinical environment using videos recorded with a smartphone camera. This is challenging, given the absence of infrared illumination and high-speed cameras, adverse lighting conditions, and the instabil-ity of the tracking device. To overcome these challenges and therefore enable tracking of saccade latency in large cohorts of subjects, we combined a deep convolutional neural network (CNN) for gaze estimation with a model-based approach for saccade onset determination that provides automated signal-quality quantification and artifact rejection (Figure 1). A variant of the iTracker gaze estimation CNN and a hyperbolic tangent model resulted in mean saccade latencies and associated standard deviations on iPhone recordings that were essentially the same as those obtained from recordings using a high-end, high-speed camera. With our system, we recorded over 19,000 latencies in 29 self-reported healthy subjects and observed significant intra- and inter-subject variability, which highlights the importance of individualized disease tracking (Figure 2). Our framework shows that unobtrusive, individualized tracking of neurodegenerative disease progression is possible.

A Simplified Design for Modeling Coronary Capillary Fluid Transport in a PDMS Model

Myocardial injury is the leading cause of adult mortal-ity in the United States. Despite the tremendous scien-tific interest in modeling the cardiac capillary damage that is characteristic of this event, few platforms exist to model in-vivo fluid dynamics, especially capillary interactions, accurately. Tissue-interface-mimicking microfluidic devices are the few in-vitro models for studying the critical behavior of capillaries, but fre-quently used models require single micrometer resolu-tion photolithography tools. This study examines and evaluates an accessible alternative design that employs centimeter-resolution photolithography to achieve similar flow properties. Although fundamental fluid dynamics properties of the new design are in accor-dance with expectations, some suggestions are made to improve the applicability of the new design for model-ing cross-membrane diffusion in capillaries.

A Bidirectional LLC Converter using Common Mode and Differential Mode Current Injection

Power converters are ubiquitous in today’s world of electronics, and the push for higher-power-density converters has opened new realms of applications for them. One popular converter topology for high-perfor-mance, high-power-density converters is the LLC res-onant converter, which relies on the frequency-depen-dent gain of an LLC network for voltage conversion. This LLC network consists of a capacitor, inductor, and transformer in series, with the transformer’s magnetiz-ing inductance serving as the LLC’s second inductance. This LLC network’s gain characteristic is advantageous because it allows the converter to achieve a wide range of input/output voltage gain with only a narrow range of switching frequencies. However, with a traditional LLC converter, this valuable gain characteristic is pres-ent for power conversion only in the forward direction. This trait is inopportune for bidirectional converters.In this work, we have demonstrated a converter topology that achieves the LLC gain characteristic during both forward and backward operation. This topology splits the traditional LLC topology into two equal halves, as Figure 1 illustrates. Then, we add an auxiliary inductor Lmb between the two inverter switch nodes to serve the magnetizing inductance role during reverse operation. Both halves are driven identically in parallel (the voltages at points A and B are always equal) for forward operation, resulting in common-mode current injection into the LLC resonant tank and no current through the auxiliary inductor. During reverse operation, the two halves are driven 180 degrees apart, resulting in differential-mode current injection that passes through the auxiliary inductor. As a result, the resonant tank exhibits a gain characteristic resembling that of an LLC network in both directions. This topology brings the high-performance of LLC resonant converters to a variety of new applications requiring bidirectional power flow, such as consumer electronics, electric vehicles, and grid energy storage.

Electro-Chemo-Mechanical Studies of Perovskite-structured Mixed Ionic-electronic Conducting SrSn1-xFexO3-x/2+δ

Solid oxide fuel cells (SOFCs) convert chemical ener-gy directly to electricity and thus have high potential conversion efficiency. Thermo-mechanical stability and high cathode surface reaction kinetics are two ma-jor criteria for good SOFC cathodes. In this work, we extend previous studies on the promising mixed ionic and electronic conducting perovskite-structured Sr-Ti1-xFexO3-x/2+δ (STF) materials system whose exchange kinetics were correlated with the minority electron charge density by replacing Ti with Sn, due to its dis-tinct band structure and higher electron mobility. Oxygen nonstoichiometry and the defect chemistry of the SrSn1-xFexO3-x/2+δ (SSF) system were examined by thermogravimetry as a function of oxygen partial pressure in the temperature range of 973-1273 K. Marginally higher reducibility was observed compared to corresponding compositions in the STF system. The bulk electrical conductivity was measured in parallel to examine how changes in defect chemistry and electronic band structure, associated with the substitution of Ti by Sn, impact carrier density and ultimately electrode performance. Bulk chemical expansion was measured by dilatometry as a function of oxygen partial pressure while surface kinetics were examined using AC impedance spectroscopy. The electrochemical properties of SSF were found not to differ significantly from the corresponding composition in STF. Though slightly shifted by the larger size of Sn, the defect equilibria and the cathode area specific resistance differed only in a limited way from that in STF. This small difference was attributed to properties being largely dominated by Fe and not by the substitution of Ti with Sn. However, due to asymmetry in the crystal structure caused by the larger size of Sn, both thermal and chemical expansion coefficients of SSF35 were found to be around 20% and 10% lower, respectively, than those of STF35, thus making SSF35 much more chemo-mechanically stable in SOFC operating conditions.

High Capacity CMOS-compatible Thin Film Batteries on Flexible Substrates

The miniaturization of sensors through advancements in low-powered MEMS devices in integrated circuits has opened up new opportunities for thin film micro-batteries. However, many of the available thin film battery materials require high-temperature process-es that necessitate additional packaging materials, which reduce the overall energy density of these bat-teries. Previous research with collaborators in Singa-pore demonstrated an all-solid-state materials system with high volumetric capacity that exclusively utilizes CMOS-compatible (i.e., room temperature) processes. This process allows integration of these batteries with CMOS circuits as distributed power supplies or for in-tegrated autonomous microsystems. Additionally, the ability to deposit all components of the battery at room temperature makes it possible to fabricate these bat-teries on thin, flexible substrates that can be densely stacked to achieve energy densities comparable to bulk batteries, which has been the focus of this project.We have successfully demonstrated a full thin film microbattery using germanium (Ge) and ruthenium dioxide (RuO2) as anode and cathode materials, respectively, with lithium phosphorous oxynitride (LiPON) as the solid-state electrolyte (Figure 1b). Although RuO2 has traditionally been used as an anode material, it has significantly higher volumetric capacity than typical cathode materials and sufficiently high electrochemical potential versus Ge to provide an output voltage of about 0.5V at a capacity of about 40 Ah/cm3 (Figure 1a). These materials are deposited onto a thin (~5 μm), flexible polyimide substrate with integrated interconnects and peeled off the handle substrate (Figure 2). These battery films can be stacked for higher power and energy densities and folded to fit any volume.

Kinetic Study of Lithiation-induced Crystallization in Amorphous Germanium Anodes in Thin Film Batteries

Germanium (Ge) is one of the most promising anode materials for complementary metal-oxide semiconduc-tor (CMOS)-compatible lithium-ion microbatteries. An intercalation or allowing anode is needed to avoid the presence of metallic Li for this application. Ge has a vol-umetric capacity of 7366 mAh/cm3, which is ten times as large as the graphite anodes used in commercial bulk batteries. When Ge is discharged below a threshold voltage, a crystalline phase Li15Ge4 forms. This phase is expected to affect the performance of Ge anodes. The degree of crystallinity is hugely affected by the cutoff voltage during lithiation (Figure 1), as well as other fac-tors including cycle number and initial film thickness. In addition to structural analyses and cyclic voltam-metry techniques, we have developed a potentiostatic technique to study the kinetics of crystallization at low voltage in amorphous Ge anodes.We found double peaks in the current vs. time curves under specific potentiostatic test conditions (Figure 2). The existence of double peaks indicates that two phase transitions occur under the given conditions. The appearance of peak 1 in Figure 2 exhibits clear correlations with the applied voltage, cycle number, and initial film thickness, which all indicate the formation of the c-Li15Ge4 phase. Combining kinetic studies with previously reported spectroscopic studies, we can attribute peak 1 to the amorphous-to-crystalline transition, while peak 2 corresponds to an amorphous-to-amorphous transition. The extent to which the crystalline phase forms has a dramatic effect on the delithiation behavior (Figure 1).

Mechanisms of Li Storage in RuO2 Electrodes for Thin Film Batteries

It has been demonstrated that RuO2 films can serve as high-performance electrodes for thin film lithium-ion batteries due to their large volumetric charge capacity, excellent cyclability, and rate capability. Unlike oth-er electrode materials, RuO2 films also do not require high-temperature processing, making them suitable for integration with low-power CMOS circuits and fabrica-tion on flexible membranes. However, lithiation and delithiation mechanisms for RuO2 are poorly under-stood, and an improved understanding is required for optimization of battery performance and yield.Lithium is stored in RuO2 films through a complex sequence of phase transformations. We have carried out detailed electrochemical studies coupled with the physical characterization of sputtered RuO2 thin films. The sequence of phase transformations during lithiation and delithiation was electrochemically characterized using galvanostatic intermittent titration technique (GITT) and cyclic voltammetry (CV) measurements (Figure 1). These characterizations were correlated with ex-situ selected area electron diffraction (SAED), X-ray photoelectron spectroscopy (XPS), Raman spectroscopy, optical microscopy (OM), scanning electron microscopy (SEM), energy-dispersive X-ray spectroscopy (EDS), and in-situ electrochemical impedance spectroscopy (EIS) results. This allows identification of phase transformations α, β, γ, and δ as reactions of Li storing between the grain boundaries between nanosized grains, formation of a reversible SEI layer, main conversion reaction, and formation of alloy LixRuO2, respectively, as Figure 2 shows. Current studies are focused on application of these insights to optimization of the performance of RuO2 electrodes in full thin film batteries. The methodology developed in this study can also be applied to other candidate thin film electrode materials. In addition, lessons from studying thin films can be applied to more complex powder-based electrodes used in bulk batteries.

Crystal Engineering of Mixed Cation Perovskite for Fabrication of Highly Efficient Solar Cells

Inorganic-organic perovskite solar cells (PSCs) have caught tremendous interest from many research groups in the field of photovoltaic devices due to their low cost, ease of fabrication, and excellent optical and electrical properties, which resulted in a record certi-fied personal consumption expenditure (PCE) of 23.3%. The presence of surface and grain boundary defects in organic–inorganic halide perovskite films is detrimen-tal to both the performance and operational stability of PSCs. Here, we study the effect of chloride (Cl) addi-tives on the bulk and surface defects of mixed-cation and halide PSCs. We found that using an anti-solvent technique divides the perovskite film into two separate layers, i.e., a bottom layer with large grains and a thin capping layer with small grains. Moreover, we demonstrate that the addition of formamidinium chloride (FACl) into the precursor solution removes the small grain perovskite capping layer and suppresses the formation of bulk and surface defects (Figure 1). This modification by FACl provides the perovskite film with remarkably improved orientation, crystallinity, and large grain size up to over 1 μm (Figure 2a). Time-resolved photoluminescence measurements show longer lifetimes for perovskite films modified by FACl and subsequently passivated by 1-adamantylamine hydrochloride (ADAHCl) than for the reference sample. Based on these treatments, we improve the quality of perovskite film and increase the power conversion efficien cy (PCE) from 19.43% for a reference sample to 21.2% for the modified device by Cl additives. This efficiency is among the highest reported values for a planar perovskite solar cell. This PCE enhancement is mostly due to the improvement of open circuit voltage (Voc) from 1110 mV to 1152 mV (Figure 2b). Moreover, the device modified by Cl additives shows a lower hysteresis effect than the reference sample. Importantly, the molecular engineering created by applying Cl additives greatly enhances the stability of the PSCs, which show only 5% degradation after aging for 90 days, which is higher than the 16% PCE loss of the reference device (Figure 2c). Additionally, we found that the modified device with Cl additives shows a smaller ideality factor of 1.8 than 2.1 for the reference device, due to the lower recombination. Our proposed approach opens up a new direction for the commercialization of efficient and stable solar cell devices.

Buckled MEMS Beams for Energy Harvesting from Low-frequency Vibrations

Vibration energy harvesters based on the resonance of the beam structure work effectively only when the operating frequency window of the beam resonance matches that of the available vibration source. None of the resonating micro-electro-mechanical system (MEMS) structures can operate with low-frequency, low-amplitude, and unpredictable ambient vibrations since the resonant frequency rises as the structure gets smaller. A bi-stable buckled beam energy harvester is has been developed to lower the operating frequency window below 100Hz for the first time at the MEMS scale. This design does not rely on the resonance of the MEMS structure but operates with the large snapping motion of the beam at very low frequencies when in-put energy overcomes an energy threshold. A fully functional piezoelectric MEMS energy harvester was designed, monolithically fabricated, and tested. An electromechanical lumped parameter model was developed to analyze the nonlinear dynamics and to guide the design of the nonlinear oscillator-based energy harvester. Multi-layer beam structure with residual stress-induced buckling was achieved through the progressive residual stress control of the deposition processes along with fabrication steps. The surface profile of the released device shows bi-stable buckling of 200𝜇𝑚, which matches well with the amount of buckling designed. Dynamic testing demonstrates that the energy harvester operates with 50% bandwidth under 70Hz at 0.5g input, operating conditions that have not been demonstrated by MEMS vibration energy harvesters before.

A Robust Electromagnetic MEMS Vibration Energy Harvester

Modern production plants lack an effective way to autonomously monitor equipment health. It is uneco-nomical to engage personnel solely to monitor ma-chines that function normally most of the time and impractical to wire plant-wide arrays of sensors for power and communication. As an alternative, vibration energy harvesters could power autonomous sensor networks that communicate wirelessly. Further, vibra-tion-based machine health monitoring could be an ef-fective method of assessing real-time machine perfor-mance. Such monitoring could become preventive by prompting maintenance prior to unrecoverable plant failures. To this end, this project seeks to advance the state of vibration energy harvesting.Our previous work yielded silicon-micro-electro-mechanical systems (MEMS) electromagnetic vibration energy harvesters suitable for powering machine health sensors. To further improve robustness and increase electrical power output, a new harvester is designed, fabricated, and demonstrated using the MP35N alloy. Its design and optimization follow that developed for earlier silicon harvesters. The new material has a mechanical modulus close to that of the silicon while not being brittle. Thus, with similar material thickness, we maintain the harvester footprint while improving robustness . The MP35N alloy allows for less stressful full stroke operation, enabling improved output power while being much more tolerant of external shock.Fabrication of the new harvester combines electric discharge machining and water-jet cutting for prototype production. The Lorentz-force harvester, with its folded-spring- suspended magnets, is packaged between two coupling coils using 3D-printed plastic package parts. The new harvester can survive large transient accelerations, common in an industrial setting; such accelerations are unsustainable by a comparable silicon harvester. This added durability brings the harvester much closer to practical application. The improved robustness enables the installation of back-irons, further improving the output power. The power output and power density (1.47 mW/cm3) are comparable to that of the previous record-setting silicon device.

Foulant-agnostic Coatings for Extreme Environments

Fouling is ubiquitous in large-scale energy production, decreasing efficiency and increasing cost due to foulant buildup. Fouling degrades systems that rely on fluid flow and heat transfer by increasing system pressure drops, impeding heat transfer, and accelerating corro-sion by fostering oxidation or concentrating chemical species within the foulant itself. This leads directly to system derating and early failure. To restore these functions, the deposits must be removed by techniques such as ultrasonic cleaning or manual removal, or the affected part must be replaced. However, these actions are often impractical, prolonging system outages and incurring significant costs due to downtime and com-ponent replacement. Therefore, it is crucial to prevent foulant deposition in the first place. The adhesion of foulant particles is due to their interaction with materi-al surfaces, which can comprise many different types of surface forces. This attraction is dominated by van der Waals (vdW) forces in extreme environments of interest to large-scale energy production, where temperatures and pressures are too high to support electrochemical double layers, and in the absence of other forces like magnetism, static charge, or steric bonding. Therefore, minimizing vdW forces should create an atomistically slick surface, preventing foulant deposition.Here, we hypothesize and experimentally demonstrate a design principle for anti-fouling coatings that exploits the relation between vdW forces and the refractive index of the coating, when vdW forces are dominant. These coatings can be made foulant-agnostic. Both experimental results and first-principles calculations support our hypothesis. As can be seen in Figure 1, the findings show that the closer the refractive index spectrum of a coating to the surrounding fluid, the better it resists the deposition of all foulants. Immediate implications include improving the efficiency of both geothermal reservoirs and nuclear power plants, which are two of the largest sources of carbon-free electricity.

All-solid-state Glucose Fuel Cell for Energy Harvesting in the Human Body

Efficiently powering sensors, pacemakers, and bio-elec-tronic devices for the human body defines a new era of medicine to track, support, and operate bodily func-tions. Glucose fuel cells have seen a renaissance in re-cent years as an implantable power source harvesting energy from readily available fuels in the human body. Compared to existing implantable batteries, glucose fuel cells require much less frequent replacement sur-gery. However, state-of-the-art glucose fuel cells are based primarily on relatively bulky polymer electro-lytes , suffer from long-term stability issues, and exhib-it low power densities. Here, we innovate a miniaturized glucose fuel cell that is fully composed of solid-state materials based on thin film processing. This all-solid-state glucose fuel cell can be scaled down to the sub-micrometer range for unprecedented miniaturization and is built on a Si chip using semiconductor fabrication methods suitable for integrated and direct powering of bio-electronic implants. Through the use of abiotic catalysts instead of conventional biological catalysts such as enzymes and microbes, long-term stability and increased power density are in perspective. Free-standing fuel cell membranes based on a proton conducting oxide on Si chips were assembled using a microfabrication route with standard semiconductor processing techniques. Oxide thin films were prepared via pulsed laser deposition. The anode is in contact with glucose in phosphate-buffered saline solution to mimic blood, whereas the cathode is in contact with oxygen. Performance characterizations were carried out via electrochemical impedance spectroscopy and galvanostatic polarization curve measurements. We report that the proposed cell is electrochemically active and shows promise in functioning as the first all-solid-state glucose fuel cell with a roughly 100-fold lowered thickness of the device (only 250 nm) compared to polymer-based glucose fuel cells.

Hybrid Intelligence in Design

One of the greatest challenges facing society is address-ing the complexities of the big-picture, system-level, interdisciplinary problems in a holistic way. Human designers, architects, and engineers have come to rely on steadily improving computational tools to design, model, and analyze their systems of interest. The de-sign of real-world systems (engineering, architecture, software, industrial, financial, and social systems) is, however, often a tumultuous endeavor fraught with great triumphs and, at times, significant regrets. Many believe that only human experts can conceptualize and orchestrate big projects upstream of designing sys-tems. There are two challenging issues in the current practice of a heuristic way of systems design. Firstly, it takes too long (decades) to become area experts through accumulating experience in many successes and some failures. Secondly, human experts also fail sometimes, especially at critical times. The questions one might ask at this stage are, “How could we teach junior engineers, architects, and scientists to design complex systems successfully without spending years of effort training on the job? Could we also assist human experts to minimize the probability of failure by leveraging recent developments in AI and big data?” While the resurgence of artificial intelligence and machine learning suggests ways to even more fully automate downstream tasks in the design process, we propose to go upstream of design, where all the key concepts are determined. Could machine intelligence help this early stage of designing beyond routine design and the optimization of pre-specified goals toward the generation of good, novel designs? Our solution to the question above will be the use of Hybrid Intelligence: combining human intelligence, which grows through experience, and machine intelligence, which can learn from all the past successes and failures and does not forget them at all. Early-stage design across disciplines requires high-level intelligence based on one’s intuition and experiential perceptions to understand challenges, constraints, and requirements in achieving the goals set. Instead of replacing humans with computational systems such as machine intelligence, we see humans and computers as working together within an ecosystem where each must bring their strengths to bear. We propose in the long run a fundamentally broad investigation of this likely convergence across the disciplines of Architecture, Structural Engineering, System Engineering, Mechanical Engineering, and Product Design. We call this approach Hybrid Intelligence because our concern is not with the intelligence of artifice, or the constraining of human designers, but rather with the effectiveness of their hybridized combination. Hybrid Intelligence for design is an umbrella term in which humans and computers collaborate from their strengths to find new processes for thinking, working, and designing.

HAQ: Hardware-aware Automated Quantization

Model quantization is a widely used technique to compress and accelerate deep neural network (DNN) inference. Emergent DNN hardware accelerators be-gin to support mixed precision (1-8 bits) to improve the computation efficiency further. This goal raises a great challenge to find the optimal bitwidth for each layer: it requires domain experts to explore the vast design space, trading off among accuracy, latency, ener-gy, and model size, which is both time-consuming and sub-optimal. The conventional quantization algorithm ignores the different hardware architectures and quan-tizes all the layers uniformly. In this paper, we introduce the Hardware-aware Automated Quantization (HAQ) framework, which leverages the reinforcement learning to determine the quantization policy automatically, and we take the hardware accelerator’s feedback in the design loop. Rather than relying on proxy signals such as FLOPs and model size, we employ a hardware simulator to generate direct feedback signals (latency and energy) to the RL agent. Compared with conventional methods, our framework is fully automated and can specialize the quantization policy for different neural network architectures and hardware architectures. Our framework effectively reduced the latency by 1.4-1.95x and the energy consumption by 1.9x with negligible loss of accuracy compared with the fixed bit width (8 bits) quantization. Our framework reveals that the optimal policies on different hardware architectures (i.e., edge and cloud architectures) under different resource constraints (i.e., latency, energy, and model size) are drastically different. We interpreted the implications of different quantization policies, which offer insights for both neural network architecture design and hardware architecture design.

AMC: AutoML for Model Compression and Acceleration on Mobile Devices

Model compression is a critical technique to efficiently deploy neural network models on mobile devices, which have limited computation resources. Conventional model compression techniques rely on hand-crafted heuristics and rule-based policies that require domain experts to explore the large design space, which is usu-ally sub-optimal and time-consuming. In this paper, we propose AutoML for Model Compression (AMC), which leverages reinforcement learning to provide the model compression policy. This learning-based policy outperforms conventional rule-based policy by having a higher compression ratio, better preserving the accu-racy, and freeing human labor. Under 4x floating point operations per second (FLOPs) reduction, we achieved 2.7% better accuracy than the hand-crafted compres-sion policy for VGG-16 on ImageNet. We applied this automated compression pipeline to MobileNet and achieved a 1.81x speedup of measured inference latency on an Android phone and 1.43x speedup on the Titan XP GPU, with only 0.1% loss of ImageNet accuracy.

Transferable Automatic Transistor Sizing with Graph Neural Networks and Reinforcement Learning

Automatic transistor sizing is challenging due to the large design space, complex performance trade-offs, and fast technology advancement. Although much work has focused on transistor sizing targeting one circuit, limited research has explored transferring knowledge from one circuit to another to reduce re-de-sign overhead. We propose leveraging a Reinforcement Learning (RL) algorithm to conduct knowledge trans-fer between different technology nodes and schemat-ics. Inspired by the fact that circuits are graphs, we also propose to learn on the schematic graph with Graph Convolutional Neural Networks (GCN). The GCN-RL agent extracts features on the schematic graph, whose vertices are transistors and edges are wires. By learning the schematic information, our method consistently achieves higher Figures of Merit (FoMs) on four differ-ent circuits than conventional black box optimization methods (Bayesian Optimization, Evolutionary Algo-rithms). Experiments on transfer learning between five technology nodes and two circuit schematics demon-strate that with the same number of simulations, RL with transfer learning can achieve much higher FoMs than agents without knowledge transfer.To the best of our knowledge, we are the first to leverage RL to transfer knowledge between technology nodes and schematics and to leverage GCN to learn on the schematic graph. Our work makes three main contributions. First, we leverage the schematic graph information in the optimization loop (open-box optimization) to build a GCN based on the circuits schematic graph to open the optimization black box effectively and embed the domain knowledge of circuits to improve performance. We use RL as an optimization algorithm; it consistently achieves better performance than a human expert, random search, Evolution Strategy, Bayesian Optimization, and MACE. Third, we use knowledge transfer with GCN-RL between technology nodes and circuit schematics to reduce the required number of simulations and shorten the design cycle.

ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware

Neural architecture search (NAS) has a great impact by automatically designing effective neural network architectures. However, the prohibitive computational demand of conventional NAS algorithms (e.g., 104 GPU hours) makes it difficult to directly search the archi-tectures on large-scale tasks (e.g., ImageNet). Differ-entiable NAS can reduce the cost of GPU hours via a continuous representation of network architecture but suffers from the high GPU memory consumption issue (grow linearly w.r.t. candidate set size). As a result, they need to utilize proxy tasks, such as training on a smaller dataset, or learning with only a few blocks, or training just for a few epochs. These architectures optimized on proxy tasks are not guaranteed to be optimal on the target task. In this paper, we present ProxylessNAS that can directly learn the architectures for large-scale target tasks and target hardware platforms. We address the high memory consumption issue of differentiable NAS and reduce the computational cost (GPU hours and GPU memory) to the same level of regular training while still allowing a large candidate set. Experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of directness and specialization. On CIFAR-10, our model achieves 2.08% test error with only 5.7M parameters, better than the previous state-of-the-art architecture AmoebaNet-B, while using 6X fewer parameters. On ImageNet, our model achieves 3.1% better top-1 accuracy than MobileNetV2, while being 1.2X faster with measured GPU latency. We also apply ProxylessNAS to specialize neural architectures for hardware with direct hardware metrics (e.g., latency) and provide insights for efficient CNN architecture design.

Defensive Quantization: When Efficiency Meets Robustness

Neural network quantization is becoming an industry standard to efficiently deploy deep learning models on hardware platforms such as CPU, GPU, TPU, and FPGAs. However, we observe that the conventional quantization approaches are vulnerable to adversarial attacks. This paper aims to raise awareness about the security of the quantized models, and we designed a novel quantization methodology to optimize the effi-ciency and robustness of deep learning models jointly. We first conduct an empirical study to show that vanilla quantization suffers more from adversarial attacks. We observe that the inferior robustness comes from the error amplification effect, where the quantization operation further enlarges the distance caused by amplified noise. Then we propose a novel defensive quantization (DQ) method by controlling the Lipschitz constant of the network during quantization, such that the magnitude of the adversarial noise remains non-expansive during inference. Extensive experiments on CIFAR-10 and Street View House Number datasets demonstrate that our new quantization method can defend neural networks against adversarial examples and even achieves superior robustness to their full- precision counterparts while maintaining the same hardware efficiency as vanilla quantization approaches. As a by-product, DQ can also improve the accuracy of quantized models without adversarial attack.

Scalable Free-space Optical Neural Networks

The transformative impact of deep neural networks (DNNs) in many fields has motivated the develop-ment of hardware accelerators to improve speed and power consumption. We present a novel photonic ap-proach based on homodyne detection where inputs and weights are encoded optically and can be repro-grammed and trained on the fly. This architecture is naturally adapted to free-space optics where both ful-ly-connected and convolutional networks can be im-plemented and scaled to millions of neurons. By utiliz-ing passive optical fan-out and performing arithmetic coherently with optical interference, this scheme cir-cumvents fundamental limits of irreversible electronic processing. We study the effect of detector shot noise on neural-network accuracy to establish a “standard quantum limit” for this system. This bound, which can be as low as 50 zJ/FLOP, suggests performance below the Landauer (thermodynamic) limit is theoretically possible with photonics.

DeeperLab: Single-shot Image Parser

Image parsing is the process of partitioning an image into multiple semantically meaningful regions (called semantic segmentation), such as car and road, and tell-ing different countable instances apart (called instance segmentation), such as car A and car B. It is a long-last-ing unsolved problem in computer vision and a basic component of many applications, such as autonomous driving. Recent approaches to image parsing typically employ separate standalone neural networks for the semantic and instance segmentation tasks and require multiple passes of inference.Instead, the proposed DeeperLab image parser performs image parsing with a significantly simpler, more fully convolutional approach that jointly addresses the semantic and instance segmentation tasks and requires only one pass of inference (i.e., one-shot), resulting in a streamlined system that better lends itself to fast processing. For quantitative evaluation, we use both the instance-based panoptic quality (PQ) metric and the proposed region-based parsing covering (PC) metric, which better captures the image parsing quality on non-countable classes and larger object instances. We report experimental results on the challenging Mapillary Vistas dataset, in which our single model achieves 31.95% (val) / 31.6% PQ (test) and 55.26% PC (val) with 3 frames per second (fps) on a graphics processing unit (GPU) or near real-time speed (22.6 fps on GPU) with reduced accuracy.

Architecture-level Energy Estimation of Accelerator Designs

With Moore’s law slowing down and Dennard scaling ending, energy-efficient domain-specific accelerators, such as deep neural network (DNN) processors for machine learning and programmable network switch-es for cloud applications, have become a promising direction for hardware designers to continue bringing energy-efficiency improvements to data and computa-tion intensive applications. To ensure fast exploration of accelerator design space, architecture-level energy estimators, which perform energy estimations without requiring complete hardware description of the de-signs, are critical to designers. However, using existing architecture-level energy estimators to obtain accurate estimates for accelerator designs is hard, as accelerator designs are diverse and sensitive to data patterns (e.g., sparsity in DNNs).To solve this problem, we present Accelergy (Figure 1), an architecture-level energy estimation methodology for accelerator designs. Accelergy interprets a design in terms of its components (e.g., an arithmetic logic unit (ALU) design consists of multipliers and adders). Since accelerator design space is very diverse, Accelergy allows users to define their own components to describe the designs. At the same time, to reflect the energy differences brought by special data processing (e.g., zero-gating in DNN accelerators), Accelergy allows users to define special actions types related to the components (e.g., read and gated read actions for SRAM). To illustrate the usage of Accelergy methodology, we implemented a sample framework for energy estimations of DNN accelerators. The framework provides a set of primitive components for users to describe the design or construct their new components. To further enhance flexibility, Accelergy provides an interface to communicate with different primitive component estimators for system-level estimations of designs that involve emerging technologies (e.g., optical DNN). Accelergy achieves 95% accuracy on total energy estimation with a well-known accelerator design – Eyeriss. Accelergy can also produce accurate energy breakdown across components estimations comparing to other estimation methodologies (Figure 2).

FastDepth: Fast Monocular Depth Estimation on Embedded Systems

Depth sensing is a critical function for many robotic tasks such as localization, mapping, and obstacle de-tection. There has been significant and growing inter-est in performing depth estimation from a single red-green-blue image, due to the relatively low cost and size of monocular cameras. However, state-of-the-art single-view depth estimation algorithms are based on fairly large deep neural networks that have high com-putational complexity and slow runtimes on embed-ded platforms. This poses a significant challenge when cameras perform real-time depth estimation on an embedded platform, for instance, mounted on a micro aerial vehicle. Our work addresses this problem of fast depth estimation on embedded systems. We investigate efficient and lightweight encoder-decoder network architectures. To further improve their computational efficiency in terms of real metrics (e.g., latency), we apply resource-aware network adaptation (NetAdapt) to automatically simplify proposed architectures. In addition to reducing encoder complexity, our proposed optimizations significantly reduce the cost of the decoder network (Figure 1). We perform hardware-specific compilation targeting deployment on the NVIDIA Jetson TX2 platform. Our methodology demonstrates that it is possible to achieve accuracy similar to that of prior work on depth estimation, but at inference speeds that are an order of magnitude faster (Figure 2). Our proposed network, FastDepth, runs at 178 fps on a TX2 GPU and at 27 fps when using only the TX2 CPU, with active power consumption under 10 W.

Low-power Adaptive Time-of-Flight Imaging for Multiple Rigid Objects

Time-of-Flight (ToF) cameras are becoming increasing-ly popular for many mobile applications. To obtain ac-curate depth maps, ToF cameras must emit many puls-es of light, which consumes a lot of power and lowers the battery life of mobile devices. However, lowering the number of emitted pulses results in noisy depth maps. To obtain accurate depth maps while reducing the overall number of emitted pulses, we propose an al-gorithm that adaptively varies the number of pulses to infrequently obtain high-power depth maps and uses them to help estimate subsequent low- power ones as shown in Figure 1. To estimate these depth maps, our technique uses the previous frame by accounting for the 3D motion in the scene. We assume that the scene contains independently moving rigid objects and show that we can efficiently estimate the motions. In contrast to our previous work, this approach uses only the data from the ToF camera and does not need RGB images to estimate the 3D motion in the scene. The resulting algorithm estimates 640 × 480 depth maps at 30 frames per second on an embedded processor. We evaluate our approach on data collected with a pulsed ToF camera and show that we can reduce the mean relative error of the low-power depth maps by up to 65% (see Figure 2) and the number of emitted pulses by up to 80%.

Fast Shannon Mutual Information Accelerator for Autonomous Robotics Exploration

Robotic exploration problems arise in various contexts, ranging from search and rescue missions to underwater and space exploration. In these domains and beyond, exploration algorithms that can rapidly reduce uncer-tainty can provide significant benefits, for instance, by shortening time and reducing resources required for exploration. Unfortunately, principled algorithms based on rigorous information-theoretic metrics, such as maximizing Shannon mutual information (MI) along the exploration path, are computationally extremely demanding.We propose a novel computing hardware architecture to efficiently compute Shannon MI on an occupancy grid map, which is the standard probabilistic representation for a 2D environment. The proposed architecture consists of multiple MI computation cores, each evaluating the MI between a single sensor beam and the occupancy grid map. We find that parallelization alone is not sufficient for high-throughput computation due to the limited bandwidth of the memory. In fact, it is critical to consider 1) memory management of the occupancy grid map storage and 2) data delivery from the occupancy grid map to MI cores. Thus, our key contributions consist of 1) a novel memory architecture that diagonally partitions the occupancy grid map into multiple banks to minimize the memory access conflicts among multiple cores (Figure 1); 2) a fast and fair memory request arbiter that ensures effective utilization of all MI computation cores; and 3) an energy-efficient, high-throughput MI computation core.This architecture (Figure 2) was optimized for 16 MI computation cores and was implemented on a field-programmable gate array. We show that it computes the MI metric for an entire map of 20m × 20m at 0.1m resolution in near real time, at 2 frames per second, which is approximately two orders of magnitude faster, while consuming an order of magnitude less power than an equivalent implementation on a Xeon CPU.

An Energy-efficient Configurable Lattice Cryptography Processor for the Quantum-secure Internet of Things

Modern public-key cryptography protocols, such as Rivest-Shamir-Adleman and elliptic-curve cryptog-raphy (ECC) will be rendered insecure by Shor’s algo-rithm when large-scale quantum computers are built. Therefore, cryptographers are working on quantum-re-sistant algorithms, and lattice-based cryptography has emerged as a prime candidate. However, the high computational complexity of these algorithms makes it challenging to implement lattice-based protocols on resource-constrained Internet of things (IoT) devices, which need to secure data against both present and future adversaries. To address this challenge, we pres-ent a lattice cryptography processor with configurable parameters that enables energy savings of up to two orders of magnitude and 124k-gate reduction in system area through architectural optimizations. This is also the first ASIC implementation that demonstrates mul-tiple lattice-based protocols proposed in the National Institute of Standards and Technology’s post-quantum standardization process.Figure 1 shows a block diagram of our system along with the chip micrograph. The chip was fabricated in a 40-nm low-power CMOS process and supported voltage scaling from 1.1V down to 0.68V. Our lattice cryptography processor occupies 106k NAND Gate Equivalents and uses 40.25KB of SRAM. When executing the Kyber-768 and NewHope-1024 key exchange schemes, our design is 28x and 37x more energy-efficient, respectively, than Cortex-M4 software, after accounting for voltage scaling. Moreover, post-quantum key exchange using our processor is 30x more energy-efficient than state-of-the-art pre-quantum ECC-based key exchange at the same pre-quantum security level. Through architectural and algorithmic optimizations, this work demonstrates practical hardware-accelerated quantum-resistant lattice-based cryptographic protocols that can be used to secure resource-constrained IoT devices of the near future.

Power Side-channel Attack on Successive Approximation Register Analog-to-digital Converters

When sensing hardware is used to acquire a private signal, there must be no information leakage through-out the entire signal chain. Applications that require such security include biomedical and military sensor platforms. Industrial and infrastructure monitoring sensing hardware must also be secure to prevent po-tentially harmful activities of adversaries. By using well-established cryptographic primitives, communica-tion links for sensing hardware can be protected from hackers. However, once hackers physically access the sensing hardware, the sensor-interface circuit can leak critical information via its power side-channel.Both analog and digital circuit blocks of the sensor-interface circuit can leak through a power side-channel as their operations depend on the sensor output value. Since the first discovery of the encryption engine’s power side-channel leakage, countermeasures against digital circuit’s power side-channel attacks have been researched in the cryptographic hardware community. However, unlike digital circuit blocks that can be protected by countermeasures, analog/mixed-signal circuit blocks are now vulnerable to side-channel attacks as their exploitations have not been recognized yet. In this work, we have developed practical power side-channel attack scenarios that make analog/mixed-signal circuit blocks become the security loophole of the entire system. We chose analog-to-digital converters (ADCs) as our target block of study. We focused our research on successive approximation register (SAR) ADCs because they are more power-efficient than other ADC types in the performance range (resolution, sampling rate) that is suitable for most sensor platforms. To experiment with power side-channel attack on SAR ADCs, we devised an attack method and mounted it on two SAR ADC products from different manufacturers. The experimental results show that SAR ADCs’ input waveforms could be faithfully reconstructed from their current traces.

Energy-efficient SAR ADC with Background Calibration and Resolution Enhancement

Many signals, for example, medical signals, do not change much from sample to sample most of the time. Conventional switching schemes for SAR ADCs do not exploit this signal characteristic and test each bit start-ing with the MSB. Previous work called least-signifi-cant-bit (LSB)-first saves energy and bit-cycles by start-ing with a previous sample code and searching for the remainder by testing bits from the LSB end. However, certain code transitions consume unnecessary energy, even when the code change over the previous code is small.This work addresses it with a new algorithm called Recode then LSB-first (RLSB-first) that reduces the switching energy and bit-cycles required for all cases of small code change across the full range of possible previous sample codes. RLSB-first uses split-DAC to systematically encode the previous code before LSB-first. RLSB-first lowers switching energy by up to 2.5 times and uses up to 3 times fewer bit-cycles than LSB-first. In addition to an energy-efficient SAR ADC, this work aims to use the savings for background calibration and resolution enhancement.

An 8-bit Multi-GHz Flash ADC with Time-based Techniques

High-speed and medium-to-low-resolution flash an-alog-to-digital converters (ADCs) are widely used in applications such as 60-GHz receivers, serial links, and high-density disk drive systems. In this project, we propose an 8-bit, multi-gigahertz flash ADC with two major innovations: the time-based comparator offset calibration and the time-based 4x interpolation.A high-speed, low-power comparator with low noise and offset requirements is a key building block. Figure 1 shows the two-stage dynamic comparator used in our design. With the scaling of CMOS technology, the offset voltage of the comparator keeps increasing due to greater transistor mismatches, making offset calibration a necessity. Traditional offset calibration methods that use digitally-controlled capacitor banks or extra input transistor pairs add extra parasitics to the comparators and slow down the operation. In this work, the proposed time-based comparator offset calibration put no additional load on the comparators and avoids the speed penalty of traditional methods.The number of comparators in a flash ADC grows exponentially with resolution. This is a major drawback of flash ADCs. Time-domain interpolation is a popular technique that utilizes the timing information from adjacent comparators to resolve extra bits of resolution without adding comparators. Figure 2 shows the pro-posed flash ADC. Sixty-five comparators are used to achieve the six most significant bits (MSBs). Sixty-four interpolators are inserted between the comparators to obtain two extra bits by comparing the delay from neighboring comparators. The input capacitance of this design is ¼ of the conventional 8-bit flash ADC. Therefore, a higher operating speed can be achieved. We introduce gating logic so that only one interpolator is enabled during operation, which reduces power consumption significantly.The prototype ADC is realized in 65-nm CMOS technology. At 2.8 GS/s, the prototype measures an SNDR of 43.3 dB at Nyquist input frequency and achieves a state-of-the-art figure-of-merit.

A Sampling Jitter-tolerant Continuous-time Pipelined ADC in 16-nm FinFET

Analog-to-digital converters (ADCs) interface re-al-world analog signals with digital systems, and hence they are an essential part of any electronic system. Al-though there have been steady improvements in the performance of ADCs, the improvements in conversion speed have been less significant because the speed-res-olution product is limited by the sampling clock jitter. The effect of sampling clock jitter has been considered fundamental. However, it has been shown that con-tinuous-time delta-sigma modulators may reduce the effect of sampling jitter. Since delta-sigma modulators rely on relatively high oversampling, they are unsuit-able for high-frequency applications such as 5G base-band processors. Therefore, ADCs with low oversam-pling ratio are desirable for high-speed data conversion.In conventional Nyquist-rate ADCs, the input is sampled upfront (Figure 1). Any jitter in the sampling clock directly affects the sampled input and degrades the signal-to-noise ratio (SNR). For fast varying input signals, the sampling jitter severely limits the maximum attainable SNR. It is well known that for a known rms sampling jitter σt, the maximum achievable SNR is limited to 1/(2πfinσt), where fin is the input signal frequency. Typically, reducing the rms jitter below 100 fs is difficult. This challenge limits the maximum SNR to just 44 dB (which is equivalent to 7 bits) for a 10-GHz input signal. Therefore, unless the effect of sampling jitter is reduced, the performance of an ADC would be greatly limited for high-frequency input signals.In this project, we propose a hybrid ADC with reduced sensitivity to sampling jitter. We are designing this ADC in 16-nm FinFET technology to give a proof-of-concept for improved sensitivity to the sampling clock jitter.

Studies on Long-term Frequency Stability of OCS Molecular Clock

Miniature clocks with high long-term stability are critical to navigation, sensing, and communication networks. Crystal/micro-electro-mechanical sys-tems (MEMS) oscillators with typical stability of 10-4 to 10-8 are not well suited for high-precision systems. Small-volume atomic clocks improved the stability to 10-11 to 10-12 by probing hyperfine transitions of Cs and Rb atoms at microwave frequencies, but their compli-cated electro-optical implementation leads to exceed-ingly high cost. Recently, complementary metal-oxide semiconductor (CMOS) molecular clocks emerged as a promising alternative to miniature clocks with high long-term stability. By probing the rotational lines of gaseous carbonyl sulfide (OCS) molecules at 267.530 GHz and then calibrating the clock’s 10 MHz output fre-quency according to the measured terahertz transition frequency of OCS, the molecular clock achieved Allan deviation of 1×10−11 with fully electronic operations. To verify the clock’s robustness to external environmental variations, two critical metrics related to the long-term stability of THz OCS clocks were studied: temperature and magnetic field. The intrinsic frequency OCS transition line is very robust to the temperature change having the temperature coefficient of a few parts per trillion per kelvin. However, the clock’s sensitivity to temperature is increased by the baseline tilting, which is mainly caused by the reflection of the THz wave at the waveguide vacuum sealing window. Also, with the presence of a magnetic field, the rotational energy levels associated with different magnetic quantum numbers deviate from their degenerate value at zero field due to the Zeeman effects. While first-order Zeeman effects of all transition sub-levels maintain the symmetry of the transition line and introduce no shift, the clock shift caused by the second-order Zeeman effects is, by theory, 4×10−13. In our preliminary testing, the temperature coefficient of the clock is ∼1.3×10−10/◦C without ovenized temperature stabilization and temperature compensation, and the upper limit of the magnetic-induced shift in response to a 75-Gauss external magnetic field is 4×10−11. This study verifies the molecular clock’s high robustness under temperature variations and strong-magnetic conditions.

Miniaturized, Ultra-stable Chip-scale Molecular Clock

Mobile electronic devices require stable, portable, and energy-efficient frequency references (or clocks). However, current approaches using quartz-crystal and micro-electro-mechanical systems (MEMS) oscil-lators suffer from frequency drift. Recent advances in chip-scale atomic clocks, which probe the hyperfine transitions of evaporated alkali atoms, have led to de-vices that can overcome this issue, but their complex construction, cost, and power consumption limit their broader deployment. Here, we show that sub-terahertz rotational transitions of polar gaseous molecules can be used as frequency bases to create low-cost, low-pow-er miniaturized clocks. A molecular clock probing 231.061 GHz (J=19←18) spectral line of carbonyl sulfide (16O12C32S) is shown in Figure 1. Based on complementary metal–oxide semiconductor (CMOS) technology, a terahertz phase-locked loop with built-in frequency-shifting-keying (FSK), referenced to an 80-MHz crystal oscillator, and generates the probing signal. The OCS molecules are accessed within a compact WR4.3 waveguide gas cell. The relative frequency error through comparing the probing frequency and selected spectral line center is detected by envelope rectification and phase-sensitive detection in a CMOS receiver. Finally, a type-I frequency locking feedback loop is established to stabilize the crystal frequency. Figure 2 shows the photograph of the CMOS molecular clock chipset. Figure 3 shows that with an averaging time of 103 s, the clock stability (defined by Allan deviation) achieves 3.8×10-10. Compared with chip-scale atomic clocks, our approach is less sensitive to external influences (temperature variation, electromagnetic field fluctuation, and mechanical vibration); offers faster frequency error compensation; and, by eliminating the need for alkali metal evaporation, offers faster start-up time and lower power consumption. Our work demonstrates the feasibility of monolithic integration of atomic-clock-grade frequency references in mainstream silicon-chip systems.

A Dense 240-GHz 4×8 Heterodyne Receiving Array on 65-nm CMOS Featuring Decentralized Generation of Coherent Local Oscillation Signals

There is a growing interest in pushing the frequency of beam-steering systems towards the terahertz range, in which case narrow beams can be formed at chip scale. However, this calls for disruptive changes to traditional terahertz receiver architectures, e.g., square-law direct detector arrays (with low sensitivity and no phase in-formation preserved) and small heterodyne mixer ar-rays (bulky and not scalable). Specifically, for the latter case, corporate feed (for generating and distributing the local oscillation (LO) signals), typically a necessary component, can be very lossy at large scale. Here, we report a highly scalable 240-GHz 4×8 heterodyne array achieved by replacing the LO corporate feed with a net-work that couples LOs generated locally at each unit. A major challenge for this architecture is that each unit should fit into a tight λ/2×λ/2 area to suppress side lobes in beamforming--it makes the integration of mixer, lo-cal oscillator, and antenna in a unit extremely difficult. This challenge is well addressed in our design, where highly compact units enable the implementation of two interleaved 4×4 phase-locked sub-arrays in an area of 1.2 mm2.The architecture of the entire array is shown in Figure 1(a). Its core component is a self-oscillating harmonic mixer (SOHM), which simultaneously (1) generates high-power LO signal and (2) down-mixes the radio frequency (RF) signal. Owing to the coupling, LOs generated in each unit are all locked to an external reference signal, so that the array is coherent. Die photo showing the placement of the array and the phase-locked loop (PLL) is given in Figure 1(b). A measured spectrum at 475-MHz (beyond the noise corner frequency) baseband signal is shown in Figure 2. The measured sensitivity (required incident RF power to achieve SNR=1 at baseband) over 1-kHz detection bandwidth is 58fW–a more than 4000× improvement over prior state-of-the-art large-scale square-law detector arrays in silicon.

A PLL-free Molecular Clock based on Second-order Dispersion Curve Interrogation of a Carbonyl Sulfide Transition at 231 GHz

Miniature clocks with high long-term stability are critical to navigation, sensing, and communication networks. Crystal/MEMS oscillators with a typical stability of 10-4 to 10-8 are not well suited for high-pre-cision systems. Small-volume atomic clocks improved the stability to 10-11 to 10-12 by probing hyperfine tran-sitions of Cs and Rb atoms at microwave frequencies, but their complicated electro-optical implementation leads to exceedingly high cost. Recently, CMOS molec-ular clocks that use a sub-THz spectrometer to probe the absorption lines of carbonyl sulfide molecules have emerged to achieve a low-cost miniature clock with high long-term stability.To generate the sub-THz probing signal within the lock-range, molecular clocks require a voltage-controlled crystal oscillator (VCXO) and a fractional-N phase-locked loop (PLL) as a frequency multiplier. However, eliminating the VCXO and PLL is necessary to further reduce the power consumption and form factor. In addition, using PLL leads to degraded in-band noise because of the high-frequency multiplication factor of the PLL. This work proposes a molecular clock without a VCXO and a PLL. A sub-THz voltage-controlled oscillator (VCO) is directly controlled by a negative feedback loop and then locked to the center of the absorption line. For frequency initialization and coarse frequency tuning, the second-harmonic dispersion curve of the absorption line profile was utilized instead of a PLL. Since the polarity of the second-harmonic dispersion curve is positive only when the frequency of the probing signal is very close to the absorption line, detection of the absorption line does not depend on the signal strength. Also, the second harmonic signal is robust against spectral baseline variations. By eliminating the VCXO and PLL from the loop and using the proposed coarse frequency tuning method, the noise performance of the proposed molecular clock is expected to improve, and further miniaturization of an ultra-stable clock can be achieved.

Chip-scale Scalable Ambient Quantum Vector Magnetometer in 65-nm CMOS

Room-temperature coherent spin state control and de-tection of nitrogen-vacancy (NV) centers in diamond have enabled magnetic field sensing with high sensitiv-ity and spatial resolution. However, current NV sens-ing apparatuses use bulky off-the-shelf discrete com-ponents, which increases the system scale and limits practical applications. To address this challenge, we de-veloped a hybrid complementary metal-oxide semicon-ductor (CMOS)-NV platform to shrink this spin-based magnetometer to chip scale. In this work, we present a fully integrated CMOS-NV quantum sensor fabricated using a 65-nm CMOS process. Magnetic field sensing is accomplished by the excitation and detection of the spin states of the NV. The frequency of the spin states is determined through optically detected magnetic resonance (ODMR). The magnetic field is proportional to the frequency splitting of the spin states (2.8 MHz/Gauss). Our CMOS-NV magnetometer system is composed of (i) a microwave generation and delivery system to control the NV’s spin states and (ii) an optical system for the readout of spin states. We implement a highly scalable microwave delivery structure, which consists of an array of current-carrying conductors. We control the current flowing in each conductor to achieve a uniform magnetic-field profile. This uniform field enables coherent driving of the NV centers, which enhances the sensitivity. The on-chip optical readout follows the microwave manipulation of the NV spin ensembles. We implemented a CMOS-compatible, three-layer grating structure to filter out the green excitation. The filter reduces the shot noise of the photo-detector caused by the input green laser. The Talbot effect is used in the filter, where we place layers of gratings with positions aligned with the maxima and minima of the green and the red diffraction patterns generated from the preceding grating layer. We detect the spin-dependent red fluorescence of the NV centers using on-chip N-Well/P-sub photodiode. This work presents a hybrid NV-CMOS platform that can perform coherent spin control and readout of the NV ensemble’s spin state: a highly advanced, scalable, and compact platform for quantum sensing.

Broadband Inter-chip Link using a Terahertz Wave on a Dielectric Waveguide

The development of data links between different mi-crochips of an onboard system has encountered a speed bottleneck due to the excessive transmission loss and dispersion of the traditional inter-chip elec-trical interconnects. Although high-order modulation schemes and sophisticated equalization techniques are normally used to enhance the speed, they also lead to significant power consumption. Silicon photonics pro-vides an alternative path to solve the problem, thanks to the excellent transmission properties of optical fi-bers; however, the existing solutions are still not fully integrated (e.g., off-chip laser source) and normally re-quire process modification to the mainstream comple-mentary metal-oxide semiconductor (CMOS) technolo-gies. Here, we aim to utilize a modulated THz wave to transmit broadband data. Similar to the optical link, the wave is confined in dielectric waveguides, with suf-ficiently low loss (~0.1dB/cm) and bandwidth (>100GHz) for board-level signal transmission (Figure 1). In com-mercial CMOS/BiCMOS platforms, we have previously demonstrated high-power THz generation with modu-lation, frequency conversion, and phase-locking capa-bilities. In addition, a room-temperature Schottky-barrier diode detector (in 130-nm CMOS) with <10pW/Hz1/2 sensitivity (antenna loss excluded) is also reported. The prototype data link will leverage these techniques to achieve a ~100Gbps/channel transmission rate with <1pJ/bit energy efficiency. As the first step of this project, we have designed a new broadband chip-to-fiber THz wave coupler, passive channelizers, broadband THz modulators, and sub-harmonic carrier generation. In contrast to previous couplers using off-chip antennas, our THz coupler is entirely implemented using the metal backend of a CMOS process and requires no post-processing (e.g., wafer thinning). The structure is also fully shielded, which prevents THz power leakage into the silicon substrate. Conventional on-chip radiators using ground shield work are the resonance type (e.g., patch antenna) and have only <5% bandwidth. In comparison, our design is based on a traveling-wave, tapered structure, which supports broadband transmission. A proof-of-concept is shown in Figure 1: two on-chip couplers are connected with a 2-cm waveguide using Rogers 3006 dielectric material. The entire back-to-back setup exhibits only ~11dB insertion loss across over 60-GHz bandwidth (Figure 2). Additionally, our on-chip and on-interposer channelizers provide a compact and efficient means of reducing ISI while combining incoherent parallel data streams.

Low-energy Current Sensing with Integrated Fluxgate Magnetometers

The ability to sense current is crucial to many industri-al applications including power line monitoring, motor controllers, battery fuel gauges, etc. We are developing smart connectors with current sensing abilities for use in the industrial Internet of things (IoT). These connec-tors can be used for 1) power quality management to measure real power, reactive power, and distortion and 2) machine health monitoring applications for continu-ous monitoring, control, prevention, and diagnosis. At the system level, the smart connectors need to 1) measure AC, DC, and multiphase currents; 2) reject stray magnetic fields; and 3) detect impending connector failure. On the sensor level, they need high accuracy and performance and a small area to fit inside the outer plastic encasing of the connectors. Therefore, the sensors must not use large external magnetic cores as field concentrators.A good system solution is to use an array of integrated fluxgate (FG) sensors (Figure 1), which offer a better alternative than Hall/magneto-resistive sensors and shunt-sensing in terms of dynamic range (~10^5), sensitivity (200 V/T), linearity (0.1%), low temperature drift, and inherent isolation. But high power consumption is a drawback for FG sensors. FG sensors work by driving magnetic cores in and out of saturation and sensing the resulting voltage difference (Figure 2). They achieve high linearity by balancing the external magnetic fields within the core with an equivalent compensation current, which can be quite power-hungry. We need to reduce the energy needs of the FG sensors so they can be used in an array, especially in energy-constrained environments. We propose a low-energy front-end design with bandwidth scalability and lower energy per measurement for FG sensors. We use a mixed-signal architecture with quick convergence techniques to enable duty cycling from >50 kHz bandwidth for machine health monitoring to <1 kHz for power quality management.

Contactless Current and Voltage Detection using Signal Processing and Machine Learning

Measuring current and voltage in electrical systems is a critical task in industrial environments and can be used to monitor power quality and machine and pro-cess performance. Easily retrofitted contactless mea-surements are preferred, but they can require difficult installations and bulky hardware. In contrast, we are developing a contactless clip-on sensor that will esti-mate voltage and current in three-phase power cables. Our goal is to create a measurement system that uses less hardware than present state-of-the-art solutions while maintaining a high level of accuracy. Current is estimated using an array of magnetic field sensors embedded in a yoke that fits around the cables, as shown in Figure 1. The measurements are filtered to remove magnetic fields from external sources, such as adjacent cables or eddy currents. This filtering employs a Best Linear Unbiased Estimate of cable currents that is based on a covariance matrix calculated from a probabilistic model of external magnetic fields detected by the sensor array. Additionally, we are using collected data to train neural networks and explore whether machine learning can generate a better estimate. To estimate voltage, we employ guarded electrodes in the yoke that fit snugly against the cables. We then sense cable voltage capacitively coupled to the electrodes and use a physical model of the electrode system to estimate the voltage differences between cables. A voltage estimate example is shown in Figure 2.At present, our system can estimate voltage with an error of less than 1% and current with an error of less than 2%, even in the presence of electric and magnetic field interference. This performance is comparable to currently used contactless detection systems but uses significantly less hardware and should thus be less costly to manufacture. Furthermore, since our estimates produce full current and voltage waveforms, we can calculate quantities such as instantaneous power and power quality.

SHARC: Self-healing Analog Circuits with RRAM and CNFETs

Next-generation applications require processing of a massive amount of data in real time, exceeding the ca-pabilities of electronic systems today. This has spurred research in a wide range of areas: from new devices to replace silicon field-effect transistors (FETs) to im-proved circuit implementations to new system archi-tectures with dense integration of logic and memory. However, isolated improvements in any one area are insufficient. Rather, enabling these next-generation applications will require combining benefits across all levels of the computing stack: leveraging new devices to realize new circuits and architectures. For instance, carbon nanotube (CNT) field-effect transistors (CNFETs) for logic and resistive random-access memory (RRAM) for memory are two promising emerging nanotechnologies for energy-efficient electronics. However, CNFETs suffer from inherent imperfections (such as of metallic CNTs (m-CNTs)), which have prohibited realizing large-scale CNFET circuits in the past. M-CNTs create shorts between the CNFET source and drain, which translates into (1) a 100x intrinsic gain reduction for analog circuits causing the failure of the whole system and (2) high power consumption and degraded noise margin for digital circuits. This work proposes a circuit design technique (called self-healing analog circuitry with RRAM correction (SHARC)) that integrates and combines the benefits of both CNFETs and RRAM to realize three-dimensional circuits that are immune to m-CNTs. Non-volatile RRAMs are 3D-integrated with CNFETs, whereas each CNFET is split into multiple minimum-width FETs (i.e., “sub-CNFETs”), with a RRAM cell in series fabricated directly under (or over) the source or drain contact of each sub-CNFET.SHARC is a non-volatile technique that self-reconfigures the circuit by programming RRAMs. The sub-FETs including m-CNTs become connected in series to reset high-resistance RRAM that effectively removes those sub-CNFETs from the circuit, while CNFETs containing only semiconducting CNTs are connected in series with set low-resistance RRAM. Leveraging this technique, we experimentally demonstrate the first and largest CMOS CNFET mixed-signal systems robust to m-CNTs (by implementing SHARC in amplifiers and switches) such as a 4b-DAC and 4b-SAR ADC. SHARC can also be combined with additional existing circuit techniques to further improve performance for very-large-scale integrated circuits.

DISC-FETs: Dual Independent Stacked Channel Field-effect Transistors

We experimentally demonstrate a three-dimensional (3D) field-effect transistor (FET) architecture leverag-ing emerging nanomaterials: dual independent stacked channel FET (DISC-FET) (Fig. 1). DISC-FET is composed of two FET channels vertically integrated on separate circuit layers separated by a shared gate. This gate mod-ulates the conductance of both FET channels simulta-neously. This 3D FET architecture enables new oppor-tunities for area-efficient 3D circuit layouts. The key to enabling DISC-FET is low-temperature processing to avoid damaging lower-layer circuits. As a case study, we use carbon nanotube (CNT) FETs (CNFETs) since they can be fabricated at low temperature (e.g., <250 ºC). We demonstrate wafer-scale CMOS CNFET-based digital logic circuits: 2-input “not-or” (NOR2) logic gates designed using DISC-FETs with independent NMOS CNT channels below and PMOS CNT channels above a shared gate (Fig. 2 and 3). This work highlights the potential of 3D integration for enabling not only new 3D system architectures, but also new 3D FET architectures and 3D circuit layouts.

Modeling and Optimizing Process Uniformity using Gaussian Process Methods

Modeling process uniformity is critical for achieving the required specifications in many advanced process technologies. For example, sputter deposition systems are prone to significant wafer-scale deposition rate variations due to the complex dynamics of the cham-ber plasma. Our work focuses on developing and apply-ing machine learning methods for modeling and mini-mizing these non-uniformities. Traditionally, modeling this process using first physics principles has been par-ticularly difficult due to the chaotic nature of plasma physics. Instead, we model this process using a Gauss-ian process (GP) framework, which uses historical data to model the deposition rate across the wafer as a func-tion of both process parameters, such as power and chamber pressure, and as a function of the equipment configuration (Figure 1). Recent work focused on creating a method for optimizing the process parameters and the equipment configuration once a predictive GP model has been fit. As the input space to our process is extremely high-dimensional, many sets of process parameters and equipment configurations may lead to our desired response. For this reason, it is neither possible to explore and model the whole space, nor required to find a configuration that meets our specifications. Therefore, when optimizing a specific process, we search not only for process inputs that lead to our desired response, but also ones that lead to tight confidence intervals (Figure 2), allowing us to accurately model only a portion of the input space and converge on a solution that meets our requirements with relatively few deposition runs. Future work will focus on additional data collection and comparing the convergence rates of our Bayesian optimization method to standard process optimization methods.

Magic-angle Graphene Superlattices: A New Platform for Strongly Correlated Physics

Understanding strongly correlated quantum matter has challenged physicists for decades. Such difficul-ties have stimulated new research paradigms, such as ultra-cold atom lattices for simulating quantum ma-terials. Here, we present a new platform to investigate strongly correlated physics, based on graphene moiré superlattices. In particular, when two graphene sheets are twisted by an angle close to the theoretically pre-dicted “magic angle,” the resulting flat band structure near the Dirac point gives rise to a strongly correlated electronic system. These flat bands exhibit half-fill-ing insulating phases at zero magnetic field, which we show to be a correlated insulator arising from electrons localized in the moiré superlattice. Moreover, upon doping this system, we find electrically tunable superconductivity in it, with many characteristics similar to the superconductivity of high-temperature cuprates. These unique properties of magic-angle twisted bilayer graphene open a new playground for exotic many-body quantum phases in a 2D platform made of pure carbon and without a magnetic field. We also present data demonstrating nematicity in the superconducting state, strange metal behavior at correlated fillings with near Planckian dissipation, and correlated states in other types of graphene superlattices. This novel platform may pave the way towards more exotic correlated systems.

Giant Enhancement of Interlayer Exchange in an Ultrathin 2D Magnet

A primary question in the emerging field of two-di-mensional van der Waals magnetic materials is how exfoliating crystals to the few-layer limit influences their magnetism. Studies on CrI3 have shown a dif-ferent magnetic ground state for ultrathin exfoliated films, but the origin is not yet understood. Here, we use electron tunneling through few-layer crystals of the layered antiferromagnetic insulator CrCl3 to probe its magnetic order (Figure 1), finding a ten-fold enhance-ment in the antiferromagnetic interlayer exchange compared to bulk crystals.Moreover, polarization-dependent Raman spectroscopy (Figure 2) reveals that exfoliated thin films of CrCl3 possess a different low-temperature stacking order than bulk crystals. Temperature-dependent Raman spectra further attribute this difference in stacking to the absence of a stacking phase transition in these thin films, even though it is well established in bulk CrCl3. We hypothesize that this difference in stacking is the origin of the unexpected magnetic ground states in the ultrathin chromium trihalides. Our study provides new insight into the connection between stacking order and interlayer interactions in novel two-dimensional magnets, which may be relevant for correlating stacking faults and mechanical deformations with the magnetic ground states of other more exotic layered magnets, such as RuCl3.

Splitting of 2D Materials with Monolayer Precision

Traditionally, two-dimensional (2D) heterostructures at the micrometer-scale level are formed by using adhesive tape, which requires isolating 2D flakes in monolayers from bulk material. However, this is a very time-consuming and random process. Moreover, al-though flakes have been isolated into a nominal mono-layer, the lateral dimensions (hundreds of micrometers) are not sufficient to guarantee the fabrication of large-scale 2D heterostructures.We introduce a layer-resolved splitting (LRS) technique that can be applied universally to harvest multiple 2D material monolayers at the wafer scale (5-centimeter diameter) by splitting single stacks of thick 2D materials grown on a single wafer. Figure 1 shows a schematic of the LRS process. The LRS process is initiated by depositing a Ni film and exfoliating the entire WS2 stack from the sapphire wafer. A Ni layer is deposited on the bottom of the WS2 film while retaining the top tape/Ni/WS2 stack as-exfoliated to harvest the a continuous WS2 monolayer. The Ni/WS2 stack is separated upon peeling while the bottom Ni strongly adheres to the WS2 monolayer, leaving a monolayer of WS2 on the bottom Ni layer. We transferred this monolayer film onto an 8-inch (20.3 cm) Si wafer coated with 90 nm of SiO2 (Figure 2A). Figure 2B shows the wafer-scale photoluminescence mapping image, which indicates that the 2D monolayer isolation was uniform across the entire 2-inch wafer area.We then fabricated arrays of 2D heterostructure devices at the wafer scale (10x10 arrays of MoS2 transistors on a 1-cm2 wafer) (Figure 2C). The transistors without h-BN exhibited very large hysteresis in their drain current-gate voltage sweep, which is detrimental to a transistor’s operation. However, substantial suppression of hysteresis has been observed in transistors with h-BN (Figure 2D).

Investigation of Atomic Interaction through Graphene via Remote Epitaxy

Remote epitaxy opens the possibility of growing epi-taxial films that “copy” the substrate crystal structure through a 2D material interlayer, enabling the produc-tion of ultrathin components for device integration. We report advances in understanding the physics of the in-teraction between the substrate and the epitaxial film. Remote atomic interaction through 2D materials is governed by the binding nature, that is, the polarity of atomic bonds, both in the underlying substrates and in 2D material interlayers. Although the potential field from covalent-bonded materials is screened by a monolayer of graphene, that from ionic-bonded materials is strong enough to penetrate through a few layers of graphene. The ionicity of the substrate material determines the distance at which its potential field is still effective for epitaxy (Figure 1). However, such field penetration can be substantially attenuated by hexagonal boron nitride (hBN), which itself has polarization in its atomic bonds. A transition from remote epitaxy to van der Waals epitaxy can be seen with an increasing number of hBN layers (Figure 2). Based on the control of transparency, modulated by the nature of materials as well as interlayer thickness, various types of single-crystalline materials across the periodic table can be epitaxially grown on 2D material-coated substrates. The epitaxial films can subsequently be released as free-standing membranes (Figure 3), which provides unique opportunities for the heterointegration of arbitrary single-crystalline thin films in functional applications.

Experimental Characterization and Modeling of Templated Solid-state Dewetting of Thin Single Crystal Films

Templated solid-state dewetting of thin single crystal films has shown potential for use as a self-assembly method for fabrication of regular, complex structures with sub-lithographic length scales (Figure 1 and first Reading below). This potential can be realized by un-derstanding and controlling dewetting instabilities and mechanisms that lead to different dewetting morphol-ogies. Since dewetting instabilities, and hence the re-sulting morphologies , depend on a number of param-eters, including crystal structure of the film (fcc, hcp, etc.), texture of the film, initial film thickness, annealing ambient, temperature, and geometry of the initial tem-plate for the film before subject to dewetting, there is a great opportunity/challenge that we are addressing through both experiments and computationally.During the past several years, we have used pre-patterned single-crystal Ni(110) or (100) epitaxial films grown on MgO as a model system and have identified and studied individual dewetting instabilities, including corner-induced instability and Rayleigh-like instability. We are currently focusing on a fingering instability that can occur during edge retraction and results in the formation of a parallel array of wire-like features. An observation motivated our current study that rough edges produced by poor lithographic edge definition led to fingering instabilities. To better understand the effect of edge roughness on the fingering instability and to control the instability, we patterned edges of large Ni(110) lithographically defined patches with a wide range of periodic perturbations. The edges of patches with the same periodic perturbation were also aligned along different crystallographic in-plane orientations to studying anisotropic effects on a templated fingering instability. We have found that fingering can be induced from those film edges with periodic perturbations (Figure 2), demonstrating that development of the fingering instability has a strong correlation with edge roughness. Furthermore, the template not only induced fingering instabilities but also provided control of the period of the fingers and the corresponding parallel wire-like structures. We have developed a kinetic model that predicts the relationship between the retraction rate of fingers and the templated finger period and are testing this model through additional experiments.

Toward Robust, Condensation-resistant, Omniphobic Surfaces

Surfaces that are repellent to liquids have broad appli-cations in anti-fouling, chemical shielding, heat trans-fer enhancement, drag reduction, self-cleaning, water purification, and icephobic surfaces. State-of-the-art omniphobic surfaces based on reentrant surface struc-tures repel all liquids, regardless of the surface mate-rial, without requiring low-surface-energy coatings. While omniphobic surfaces have been designed and demonstrated, they fail catastrophically during con-densation, a phenomenon ubiquitous in both nature and industrial applications. Specifically, as condensate nucleates within the reentrant geometry, omniphobic-ity is destroyed. Here, we show a nanostructured surface that can repel liquids even during condensation (Figure 1). This surface consists of isolated reentrant cavities with a pitch on the order of 100 nanometers to prevent droplets from nucleating and spreading within all structures. We developed a model to guide surface design and subsequently fabricated and tested these surfaces with various liquids. We demonstrated repellency to various liquids up to 10 °C below the dew point and showed durability over three weeks. Furthermore, the design is robust to defects or damage to the surface. This work provides important insights for achieving robust, omniphobic surfaces.

Observation of Second Sound in Graphite above 100K

Second sound is an unusual phenomenon in which heat transports in a wave-like manner, rather than by the more usual diffusive motion. This wave-like motion is a result of the dominance of normal phonon-phonon scattering, which conserves the total phonon momen-tum over any other phonon resistive scatterings. Simi-lar to a gas system, where particles scatter without los-ing total momentum, phonons gain an average velocity under a temperature gradient when normal scattering dominates, and their transport is said to be in the hy-drodynamic regime. In this regime, a heat pulse prop-agates as a wave similar to the way a pressure impulse generates sound waves, and this wave is called second sound. Previously, second sound has been observed in only a few materials at very low temperature (<20 K). Recently, we successfully predicted and observed second sound in graphite up to 150 K using the transient thermal grating (TTG) technique. In TTG, two transient pump laser beams create interference at the sample surface and generate a thermal grating. A probe beam detects the transient decay of the thermal grating. When the phonon system is diffusive, the thermal grating will decay diffusively with fixed peak and valley positions, corresponding to an exponential decay signal, as shown by the curves at 200 K and 300 K in Figure 1A. However, between 85-150K, the heat wave motion leads to an oscillating exponential signal (hallmark of second sound in TTG), as shown by the curves at 85-150 K. The experimental result is well supported by an ab initio simulation (Figure 1B).

Ultrahigh Thermal Conductivity and Mobility in c-BAs

As the transistor density gets larger and larger in to-day’s central processing unit, thermal management becomes necessary to improve reliability and prevent overheating failure. Utilizing ultrahigh thermal con-ductivity materials that can help efficiently dissipate the generated heat from the chips is one of the passive cooling strategies in electronics. In this way, diamond, as the highest thermally conducting material, has been used as the heat spreader. However, diamond is limited by its high cost and interface issues like poor thermal and mechanical coupling to common semiconductors. Therefore, finding other ultrahigh thermal conductiv-ity materials that can totally or partially overcome the limitation of diamond can be significantly beneficial. Recently, our group with collaborators has predicted, synthesized, and measured ultrahigh thermal conductivity in cubic barium arsenides (c-BAs). First, c-BAs samples of mm-size were successfully synthesized by the chemical vapor transport technique at the University of Houston. With a metal layer coated on top of the sample surface as the transducer, we carried out thermal transport measurements on the samples using time-domain thermoreflectance and frequency-domain thermoreflectance, and the measured thermal conductivity is as high as ~1200 W/mK at room temperature (Figure 1). This places c-BAs as the second most heat-conducting cubic material. In addition, c-BAs is a semiconductor with an indirect bandgap around 1.7 eV. We predict that they have comparably high mobility for both electrons and holes (Figure 2). The high thermal conductivity and high mobility of c-BN promise interesting applications in microelectronics.

Morphological Stability of Single Crystal Co and Ru Nanowires

High-performance integrated circuits contain tens of kilometers of metal interconnects, the cross-sectional area of which must shrink in lockstep with shrinking transistors. The reliability of integrated circuits is con-tingent upon morphologically stable interconnects. At the tiny length scales of next-generation intercon-nects, the electrical resistance of Ru and Co nanowires is expected to be lower than that of nanowires based on current copper technology; thus the morphological stability of these two materials is of particular practical interest. Solid-state dewetting by surface self-diffusion is often the dominant mechanism by which the mor-phology of micro- and nano-scale features evolve at el-evated temperatures. As feature dimensions decrease, the temperature at which dewetting also occurs drops, which can lead to significant morphological degrada-tion at surprisingly low homologous temperatures. Al-though solid-state dewetting is fairly well understood in isotropic systems, the dewetting behaviors of aniso-tropic, crystalline solids are far more complicated and more experimental, and modeling work is required to identify crystallographic characteristics that will opti-mize morphological stability.Previous work on single-crystal Ni films has demonstrated that crystalline anisotropy gives rise to special crystallographic orientations along which single-crystal wires are kinetically resistant to morphological instabilities. The strongly faceted surfaces of these wires are also predicted to reduce electron scattering and decrease interconnect resistance. For Ru nanowires, exploratory work with single-crystal (0001) films suggests that wires oriented along <1-210> directions will be particularly stable. Work on patterning and testing of such wires is currently underway. We have also begun similar experiments on single-crystal Co films and will compare our results across the Co, Ni, and Ru systems to construct a more fundamental understanding of dewetting behavior in crystalline nano-scale structures such as interconnects.

Field Controlled Defects in Layered Cuprate-based Materials

Both the nature and concentration of oxygen defects in oxide materials can have a significant impact on their physical and chemical properties, as well as on key interfacial reaction kinetics such as oxygen exchange with the atmosphere. Most commonly, the desired oxy-gen defect concentration, or equivalently oxygen non-stoichiometry, is attained by doping with aliovalent cations and/or controlling the oxygen partial pressure and temperature in which the materials are equilibrat-ed or annealed. These approaches, however, are limit-ed by dopant solubility limits and the range of oxygen partial pressures readily experimentally achievable, and they require knowledge of the applicable defect chemical model. In this study, we fine-tune oxygen defect concentrations in rare earth cuprate (RE2CuO4: RE = rare earth) solid oxide fuel cell (SOFC) cathode materials by application of electrical potentials across an yttria-stabilized zirconia (YSZ) supporting electrolyte. These layered perovskites can incorporate both oxygen interstitials, and vacancies, thereby broadening the range of investigations. Here, we show a strong correlation between oxygen nonstoichiometry values (which are determined by in-situ measurement of chemical capacitance) and oxygen surface exchange kinetics (which are inversely proportional to the area-specific-resistance) without changing cation chemistry. Both types of oxygen defects, interstitials and vacancies, dramatically enhance surface kinetics. These studies are expected to provide further insights into the defect and transport mechanisms that support enhanced SOFC cathode performance.

Mixed Electron-proton Conductor Membrane Mediates H2 Oxidation

Electrochemical transformations are key to the inter-conversion of electrical and chemical energy and ubiq-uitous in the formation of commodity chemicals. Elec-trocatalysts which enable these transformations must serve to both activate chemical bonds and facilitate electron-proton transfer. In conventional electrocatal-ysis, these two functions occur at a singular catalyst electrolyte interface that prevents independent opti-mization of either process; changes to the interface will inherently affect both functions. Critically, the optimal interface for one function often does not coincide with the optimal structure for the other. We have shown that for hydrogen oxidation reaction (HOR), these two functions can be segregated by interposing a mixed electron-proton conductor (MEPC) membrane between the catalyst and electrolyte.We have designed a device that enables concurrent electrochemical proton-electron extraction at an MEPC electrolyte interface and H2 activation at a gas catalyst interface. A reduced WO3 (WOx) membrane supported on a porous support is decorated with a platinum catalyst on one side (Figure 1). At the gas Pt interface, H2 is dissociatively activated at Pt surfaces to generate H-atoms. The resulting H-atoms migrate across the Pt WOx boundary to intercalate into the WOx via H-spillover and diffuse through the WOx membrane. At the MEPC electrolyte interface, the applied electrochemical potential drives the separation of protons and electrons with protons entering the solution and electrons passing current through the external circuit. This work represents the first demonstration of employing an MEPC membrane to segregate the bond activation and charge transfer functions in electrocatalysis.These devices exhibit respectable current densities that exceed 20 mA cm–2 at 0.5 V vs. RHE. We found that the thickness of the membrane does not limit the rate of H2 oxidation catalysis, suggesting that H-diffusion within the WOx membrane is relatively rapid. Instead, the condensed MEPC membrane serves as a barrier to prevent impurities and poisoning species dissolved in the electrolyte to degrade HOR catalysis. On the other hand, the rate of HOR depends on the Pt sputtering time (Figure 2). An increasing rate was found, up to 35 s of Pt deposition, which decreased upon continued sputtering. This suggests that H-spillover across the Pt WOx boundary limits the overall rate of HOR and a 35 s deposition of Pt maximizes the Pt WOx boundary line density. Future work focuses on a selection of materials for these devices to enable a library of diverse reactivity.

Dynamic Approach of Quantifying Strain Effects on Ionic and Electronic Defects in Functional Oxides

The search for novel electronic and magnetic proper-ties in functional oxides has generated a growing in-terest in understanding the mobility and stability of ionic and electronic defects in these materials. Instead of altering material content, most research views me-chanical strain as a lever for modulating defect concen-tration and mobility more finely and continuously in both semiconductors and function-al oxides. Previous studies also proposed that strain may increase ionic mobility by orders of magni-tude, which is crucial for lowering the operation temperature of solid oxide fuel cells.However, experimental and computational results differ significantly among research groups due to the convoluted effect of mechanical strain and film/substrate interface on defect content and mobility. Such reliance on substrate selection to induce strain in the oxide thin film also limits the range of strain accessible, with limited data available to date.We have developed an experimental technique that facilitates application of in-plane strain to functional oxide thin films continuously on the same substrate. First, we combine photolithography and metal sputtering to deposit an interdigitated Pt electrode on our sample (Figure 1). Next, we conduct 3- or 4-point bending and concurrent conductivity measurement of the thin film-on-substrate device (Figure 2). This approach is accessible to a wide temperature range and precise gas control relevant to mixed ionic-electronic conducting oxides. We can strain and measure the transport properties of the same functional oxide thin film at high temperature in situ, over a range of strains applied to a single system. Combining these experiments with our ab initio computational simulations and predic-tions of carrier dominance over a range of strains and temperatures, we also aim to measure the carri-er mobility in Nb-doped SrTiO3 as a function of applied strain, to observe the sudden change of carrier mobility and temperature dependency. We believe this will also be a powerful technique for studying the strain effect on surface reactions like exsolution or catalytic reaction.

3-D Printed Microarchitected Ceramics for Low-heat Capacity Reactors

Efficient heat and mass transfer for catalytic reactors are desirable for a broad array of biological and envi-ronmental applications and are of great import to the automotive and power plant industry. The conversion efficiency of catalytic reactors relies on the tempera-ture of the constituent substrate and its thermal re-sponse. Although porous substrates with thinner cell/pore walls and higher cell/pore density enable faster catalyst activation due to low thermal mass and, larger surface area, the manufacturing of well-engineered structures with thin walls and higher cell/pore density remains a challenge. For example, there is a practical limit to the maximum cell density and the minimum wall thickness of the honeycomb substrate caused by difficulties in the extrusion-based process, such as larger flow drag force and inhomogeneity. Another promising candidate, the open-cell foams, also suffer manufacturing and assembly difficulties due to their low mechanical strength, high flow resistance, and high heat capacity caused by random-pore architectures. To overcome these limitations, we proposed manufacturing-friendly structural design and additive manufacturing for microarchitected ceramic substrates having both a large catalytic surface area and low thermal mass. Our idea for achieving efficient catalytic substrates is leveraging 3D micro-lattices of thin-walled tubular networks instead of conventional honeycomb monoliths (Figure 1). We measured surface temperature by thermal IR camera (Figure 1) and investigated the thermal response of each architecture (Figure 2). The proposed 3D hollow micro-lattice was heated up and cooled down faster than the monoliths’ structure. This result verifies the low thermal mass of the proposed 3D micro-lattice ceramic to enable a faster thermal response for the faster catalytic activation. Therefore, we expect that the proposed 3D ceramic micro-lattice structure will have a high catalytic conversion efficiency and accelerate the development of an efficient gas purification system for automotive and environmental applications.

Additively Manufactured Externally-fed Electrospray Sources

Additive manufacturing (AM) is a layer-by-layer fabri-cation technique that creates solid objects by putting material where needed, instead of removing material from stock. Recent advances in AM have made possi-ble the implementation of microsystems that surpass the performance of state-of-the-art counterparts made in a clean room, as well as the demonstration of devic-es that are challenging or unfeasible to create using standard microfabrication–particularly in the area of microfluidics. In addition, AM is inherently compatible with implementing, with great precision, hierarchical structures with features spanning orders of magnitude in size to accomplish multiple tasks efficiently.In this project, we are exploring AM to develop, at a low-cost, massively multiplexed externally-fed electrohydrodynamic liquid ionizers (Figure 1) for a wide range of applications such as mass spectrometry, nanosatellite propulsion, species transport, and agile manufacturing. These devices are mesoscaled arrays of high-aspect-ratio, hundreds-of-microns tall, micron-sharp tips that are conformally covered with a nanostructured layer that transports and regulates the flow of liquid from the reservoir to the emission sites. Manufacturing issues such as inter-process compatibility and tip array uniformity need to be addressed to implement devices that operate efficiently successfully. Current work focuses on exploring and optimizing various manufacturing techniques to monolithically create the electrospray source out of different structures made of different materials; future work includes assessment of device performance, e.g., emission characteristics and uniformity.

Additive Manufacturing of Microfluidics via Extrusion of Metal Clay

Most microfluidics uses closed microchannels to effi-ciently accomplish tasks such as species mixing, heat transfer, and particle sorting by increasing the sur-face-to-volume ratio of the fluid(s) involved in the pro-cess. However, the current manufacturing techniques for microfluidics present disadvantages such as high-cost, long production time, no device customization, elaborated design iteration, restriction in the kinds of structures that can be made, and low fabrication yield.Recent research results demonstrate that additive manufacturing can readily address the shortcomings outlined, often yielding devices that surpass the state of the art or for which traditional microfabrication creates no counterpart. However, most 3D-printed microfluidics are made of polymeric feedstock, which is not compatible with high-pressure and/or high-temperature applications. Mainstream 3D printing methods for metal include lost-wax micro molding, inkjet binder, and direct metal laser sintering; these processes are unideal to produce monolithic closed-channel microfluidics because they either require internal dummy structures or create internal voids filled in with unprocessed printable material, both of which are challenging to remove from the printed part.In this project, we are exploring the use of extrusion of metal clay to implement closed-channel microfluidics; the technique is arguably similar to fused filament fabrication and can readily create voids without spurious infill or post-processing required. Via the extrusion of metal clay, leak-tight metal microchannel with monolithic, working ports have been created (Figure 1). A cross-section of the microfluidic shows an unclogged microchannel, evidencing the feasibility of the technique to create closed channels with hydraulic diameters of relevance to microfluidics (Figure 2). Current work focuses on exploring the design space of the technology and demonstrating an application of relevance.

3D-Printed, Low-cost, Miniature Liquid Pump

Many compact systems use pumps to precisely set flow rates of liquid or, in general, to manipulate small liquid volumes for effective mass transport, cooling, or momentum transfer. Numerous microfabricated posi-tive displacement pumps for liquids with chamber vol-umes that are cycled using valves have been proposed. Pumps made via standard (i.e., cleanroom) microfabri-cation typically cannot deliver large flow rates with-out integrating hydraulic amplification or operating at high frequency due to their small pump chambers.Additive manufacturing, i.e., the layer-by-layer fabrication of objects using as template a computer-aided design model, has recently been explored as a processing arena for microsystems. In particular, researchers have reported 3-D printed pumps for liquids and gases with performance on par or better than counterparts made with standard microfabrication. Building upon earlier work on printed MEMS magnetic actuators, we recently developed miniature liquid pumps printed in pure nylon 12 via fused filament fabrication (FFF) whereby a thermoplastic filament is extruded from a hot nozzle to create a solid object layer by layer.Our low-cost, leak-tight, miniature devices are microfabricated using 150- to 300-µm layers with a multi-step printing process (Figure 1) that monolithically creates all key features with <13- µm in-plane misalignment. Each pump has a rigid frame, a 21-mm-diameter, 150-µm-thick membrane connected at its center to a piston with an embedded magnet, chamber, passive ball valves, and two barbed fluidic connectors (Figure 2). Pump fabrication under 2 hours and costs less than $4.65 are achieved. Finite element analysis of the actuator predicts a maximum stress of 18.7 MPa @ 2-mm deflection, about the fatigue limit of nylon 12 (i.e., 19 MPa). A maximum water flow rate of 1.37 ml/min at 15.1 Hz actuation frequency is calculated, comparable to reported values of miniature liquid pumps with up to 200X higher actuation frequency.

3D-Printed Microfluidics to Evaluate Immunotherapy Efficacy

Microfluidic devices are conceptually an ideal platform for the provision of personalized medical evaluations as they require small analyte volumes and facilitate rapid and sensitive investigations. However, inher-ent challenges in device fabrication have impeded the widespread adoption of microfluidic technologies in the clinical setting. Additive manufacturing could ad-dress the constraints associated with traditional mi-crofabrication, enabling greater microfluidic design complexity, fabrication simplification (e.g., removal of alignment and bonding process steps), manufacturing scalability, and rapid and inexpensive design iterations. We have developed an entirely 3D-printed microfluidic platform that enables modeling of interactions between tumors and immune cells, providing a microenvironment for testing the efficacy of immunotherapy treatment. The monolithic platform allows for real-time analysis of interactions between a resected tumor fragment and resident or circulating lymphocytes in the presence of immunotherapy agents. Our high-resolution, non-cytotoxic, transparent device monolithically integrates a variety of microfluidic components into a single chip, greatly simplifying device operation vs. traditionally-fabricated microfluidic systems. The 3D-printed device sustains viability of biopsied tissue fragments under dynamic perfusion for at least 72 hours while enabling simultaneous administration of drug treatments, illustrating a useful tool for drug development and precision medicine for immunotherapy. Confocal microscopy of the tumor tissue and resident lymphocytes in the presence of fluorescent tracers provides real-time monitoring of tumor response to various immunotherapy. The platform and accompanying analysis methods distinguish between a positive immune response and a lack of tissue response in the presence of immunotherapeutic agents.This platform introduces novel methodologies in modeling and analyzing tumor response to improve prediction of patient-specific immunotherapy efficacy. To the best of our knowledge, this is the first report of human tumor fragments cultured in a dynamic perfusion system capable of testing the effect of circulating immune checkpoint inhibitors on resident tumor-infiltrating lymphocytes.

Electrohydrodynamic Printing of Ceramic Piezoelectric Films for High-frequency Applications

The high operating frequencies that ceramic piezoelec-tric ultra-thin films attain have made possible exciting applications such as energy harvesting, telecommuni-cations’ filters, high sensitivity biosensors, and acous-tofluidic devices; however, the inherently high cost and complexity of current manufacturing methods limit, in general, their widespread use. Additive manufacturing (AM), which has proven successful in creating complex devices and components of relevance to micro and nanosystems, could overcome these disadvantages; nevertheless, AM of piezoelectrics has been achieved only with polymer-based materials–unsuitable for said applications.We report the first additively manufactured ceramic ultra-thin piezoelectric films compatible with high-frequency applications using electrohydrodynamic deposition (EHD) at room temperature. The films were made by electrospraying a zinc oxide (ZnO) nanoparticle liquid feedstock, directly writing line imprints as thin as 213 nm and as narrow as 198 µm. We harness a previously unreported effect to align the polar axis of the imprint and obtain overall piezoelectricity. As Figure 1a shows, the (100) orientation monotonically increases as the linear density of the deposition is reduced by increasing the raster speed or reducing the feedstock flow rate (Q)–provided two conditions are met: the feedstock is ionized (via EHD), and a small separation between emitter and substrate is used. Notably, the orienting effect directly acts on the direction of the polar axis by means of the rastering direction (Figure 2a), allowing for vibration modes and resonator configurations that were previously unfeasible. The macroscopic piezoelectric behavior is shown through piezoforce response microscopy (PFM) (Figure 2b) and the suitability for high-frequency applications was demonstrated by testing thin-film bulk acoustic resonators (FBAR) on a flexible polymer substrate, where the resonant frequency of ~5 GHz was used to calculate the acoustic speed of the films (~2,000 m/s), which is close to the transversal wave speed of ZnO.

3D-Printed, Monolithic, Multi-tip MEMS Corona Discharge Ionizers

A corona discharge is a high-electric field ionization phenomenon caused by the development of a self-sus-tained electron avalanche between a sharp electrode (i.e., corona electrode) and a blunt electrode; the ions create a plasma region around the corona electrode and in their travel to the opposite electrode transfer momentum to the surrounding fluid. In this project, we are harnessing advanced metal inkjet printing technol-ogy to demonstrate massively multiplexed MEMS coro-na discharge ionizers (Figure 1), with the aim to increase greatly their ionization throughput and optimize their transduction mechanism to be able to implement ex-citing applications such as no-moving-parts pumps for gases and compact ion mobility spectrometers.A 1D electrohydrodynamic coaxial cylinder model was implemented in COMSOL Multiphysics to study the ionization and collision processes in air at atmospheric pressure and room temperature of a 1-tip device, predicting a 400-µm-thick corona region surrounding the corona tip. The onset voltage estimated from the simulation is 5.849 kV, which is close to the theoretical value from Peek’s formula of 6.416 kV. In addition, current over voltage (I/V) versus bias voltage minus the onset voltage (V-V0) characteristics were collected for different ionizer array designs while varying the separation between the corona electrode and the collector electrode; the data follow the Townsend current-voltage model (Figure 2). Moreover, the data show that the corona current decreases with increased spacing of the corona electrode-to-collector electrode due to the decrease of the electric field on the tips; however, a smaller separation between the corona electrode and the collector electrode results in larger fluctuations in the corona discharge current. Devices with different numbers of tips tend to generate the same total corona current at the same bias voltage although more tips are set to discharge as the number of tips increases; this increase can be ascribed to the stronger interference between adjacent tips when the tip-to-tip spacing decreases. Current research efforts focus on optimizing the array design to minimize electric field shadowing and sharpening the tips to achieve operation at a lower bias voltage.

3D-Printed Gas Ionizer with CNT Cathode for Compact Mass Spectrometry

Mass spectrometers are powerful chemical analytical instruments used to quantitatively characterize the composition of unknown samples via ionization and mass-to-charge ratio species sorting. However, main-stream mass spectrometers are large, heavy, power hungry, and expensive, limiting their applicability in real-time and in-situ applications. Gas molecules can be ionized via electron impact ionization (EII), for which a source of electrons, i.e., a cathode, is required. Cold cathodes emit electrons into a vacuum via quantum tunneling due to high surface electric fields that low-er and narrow the barrier that traps electrons within the material; typically, high-aspect-ratio, nano-sharp tips are used to produce such fields with moderate bias voltages. Compared to thermionic cathodes, field emission electron sources have faster response and less power consumption. Compared to other field emitters, carbon nanotubes (CNTs) are less affected by back-ion bombardment and chemical degradation. There are nu-merous reports of gas ionizers with CNT cathodes EIIs; however, these devices are microfabricated using clean-room technology and/or use ion-generating structures machined with standard technologies, affecting their cost and size.In this project, we are harnessing additive manufacturing (AM) to develop novel electron impact ionizers that circumvent these challenges. AM has unique advantages over traditional manufacturing methods including compatibility with creating complex 3D geometries, print customization, and waste reduction. Our design (Figure 1) uses inkjet binder printing of SS 316L to create electrodes to efficiently generate ions and steer charged species, stereolithography of polymer resin to define the dielectric structures that electrically isolate the different electrodes, and an additively manufactured CNT electron source. We have successfully characterized the ionizers at pressures as high as 5 mTorr while achieving ionization efficiencies as high as 8.5% (Figure 2).

Printed CNT Field Emission Sources with Integrated Extractor Electrode

Field emission cathodes are promising electron sourc-es for exciting applications such as flat-panel displays, free-electron lasers, and portable mass spectrometry where fast switching, low-pressure operation, and low power consumption are favored metrics. A field emitter quantum tunnels electrons to a vacuum due to the high electrostatic fields at its surface; this tunneling is typi-cally done at low voltage using a whisker-like structure. Carbon nanotubes (CNTs) are attractive structures to produce electron field emission due to their ultrasharp tip diameter, high aspect ratio, high electrical conduc-tivity, and excellent mechanical and chemical stability. Although CNT-based cold cathodes have been widely reported in the literature, their manufacture could be quite expensive (e.g., devices partially or fully made in a semiconductor cleanroom), or the extractor electrode of the cathode is an external mesh, causing high-beam interception (e.g., in screen-printed devices) or requir-ing an advanced method of assembly to the emitting component to achieve high transmission.In this project, we are developing novel field emission sources that are fully additively manufactured to circumvent the aforementioned challenges. The devices are made via direct ink write (DIW) printing, which is one of the least expensive and most versatile additive manufacturing methods as is capable of monolithic multi-material printing. Compared to screen printing, DIW does not involve static masks to transfer patterns and produces significantly less waste. The fully-printed field emission electron source is composed of two continuous imprints: a spiral trace made of a CNT compound, acting as an emitting electrode, symmetrically surrounded on both sides by a spiral trace made of silver nanoparticles, acting as in-plane extractor electrode (Figure 1). After printing, the CNT spiral receives a mechanical treatment that releases the CNT tips from the bulk of the imprint (Figure 2), enabling field emission from the CNT imprint. Characterization of the printed CNT field emission sources in triode configuration (i.e., using an external anode) shows low turn-on voltage and low interception of the emitted current by the extractor electrode. Current work focuses on design optimization and experimental characterization of the devices.

Controlling the Nanostructure in Room-temperature-microsputtered Metal

Sputter deposition involves the ejection of atoms from a target and the atoms’ subsequent deposition on a nearby substrate. Because the deposition is done on the atomic level, the nanostructure of the deposit can vary significantly. This variance is of concern, as it can great-ly affect material strength and conductivity. Tradition-al sputtering relies on vacuum and thermal annealing to ensure dense, highly conductive deposits. However, agile manufacturing on temperature-sensitive sub-strates is not compatible with these two solutions.To enable high-quality material without heating the material or requiring a vacuum, we performed a statistically-motivated set of experiments to determine what deposition parameters improve the material quality. We developed an empirical model and found that an appropriate electrical bias voltage, applied either to the substrate or to a conductive plate under the substrate, has the greatest impact on the material quality. This is due to the presence of charged nanoparticles, formed by collisions between sputtered atoms in the dense plasma around the sputter target. The applied electric field attracts positively charged nanoparticles, allowing the nanoparticles to strike the substrate with more energy than their temperature alone would dictate. This extra energy enhances the mobility of the deposited metal, allowing it to form denser, more energetically favorable coatings (Figure 1) without significant substrate heating. With this technique, we have improved the conductivity of the sputter coating to 5x bulk metal (15 µΩ·cm) at room temperature.Applied electric fields also improve the coating’s thickness. In the absence of electric fields, the sputtering process is self-limiting. As the positively charged sputtered material reaches the substrate, charge builds upon the substrate, repelling charged sputtered material and preventing the deposit from thickening. However, biasing the substrate with a negative voltage prevents this charge from accumulating, allowing for thicker (> 200 nm) films.

Gated Silicon Field Ionization Arrays for Compact Neutron Sources

Neutron radiation is widely used in various applications, ranging from the analysis of the composition and structure of materials and cancer therapy to neutron imaging for security. However, most applications require a large neutron flux that is often achieved only in large infrastructures such as nuclear reactors and accelerators. Neutrons are generated by ionizing deuterium (D2) to produce deuterium ions (D+) that can be accelerated towards a target loaded with either D or tritium (T). The reaction generates neutrons and isotopes of He, with the D-T reaction producing the higher neutron yield. Classic ion sources require extremely high positive electric fields, on the order of 108 volts per centimeter (10 V/nm). Such a field is achievable only in the vicinity of sharp electrodes under a large bias, and consequently, ion sources for neutron generation are bulky. This work explores, as an alternative, highly scalable and compact Si field ionization arrays (FIAs) with a unique device architecture that uses self-aligned gates and a high-aspect-ratio (~40:1) silicon nanowire current limiter to regulate electron flow to each field emitter tip in the array (Figure 1). The tip radius has a log-normal distribution with a mean of 5 nm and a standard deviation of 1.5 nm, while the gate aperture is ~350 nm in diameter and is within 200 nm of the tip. Field factors, β, > 1 × 106 cm-1 can be achieved with these Si FIAs, implying that gate-emitter voltages of 250-300 V (if not less) can produce D+ based on the tip field of 25-30 V/nm. In this work, our devices achieve ionization current of up to 5 nA at ~140 V for D2 at pressures of 10 mTorr. Gases such as He and Ar can also be ionized at voltages (<100 V) with these compact Si FIAs (Figure 2).

Silicon Field Emitter Arrays (FEAs) with Focusing Gate and Integrated Nanowire Current Limiter

The advent of microfabrication has enabled scalable and high-density Si field emitter arrays (FEAs). These are advantageous due to compatibility with comple-mentary metal-oxide-semiconductor (CMOS) process-es, the maturity of the technology, and the ease in fabri-cating sharp tips using oxidation. The use of a current limiter is necessary to avoid burn-out of the sharper tips. Active methods using integrated MOS field-effect transistors and passive methods using a nano-pillar (~200 nm wide, 8 µm tall) in conjunction with the tip have been demonstrated. Si FEAs with single gates re-ported in our previous works have current densities >100 A/cm2 and operate with lifetimes of over 100 hours. The need for another gate (Figure 1) becomes essential to control the focal spot size of the electron beam as electrons leaving the tip have an emission angle of ≈ 12.5°. The focus electrode provides a radial electric field that reduces the lateral velocity of stray electrons and narrows the cone angle of the beam reaching the anode. Varying the voltage on the focus gate reduces the focal spot size or achieves an electron beam modulator for radio frequency applications. In this work, we fabricate dense (1-μm pitch) double-gated Si with an integrated nanowire current limiter (Figure 2). The apertures are ~350 nm and ~550 nm for the extractor and focus gates, respectively, with a 350-nm-thick oxide insulator separating the two gates. Electrical characterization of the fabricated devices shows that the focus-to-gate ratio (VFE/VGE) can be used to control the anode current (Figure 2). When the focus voltage exceeds the gate voltage, the field superposition increases the extracted current, and vice versa. These devices can potentially find applications as high-current focused electron sources in flat panel displays, nano-focused X-ray generation, and microwave tubes.

Highly Uniform Silicon Field Emitter Arrays

Cold cathodes based on silicon field emitter arrays (FEAs) have shown promise in a variety of applica-tions requiring high-current-density electron sources. However, FEAs face a number of challenges that have prevented them from achieving widespread use in commercial and military applications. One problem limiting the reliability of FEAs is emitter tip burnout due to Joule heating. The current fabrication process for FEAs results in a non-uniform distribution of emit-ter tip radii. At a fixed voltage, emitters with a small ra-dius emit a higher current while emitters with a large radius emit a lower current. Therefore, emitters with a small radius reach their thermal limit due to Joule heating at lower voltages and consequently burn out. Previous solutions to tip burnout have focused on lim-iting the emitter current with resistors, transistors, or nanowires to obtain more uniform emission current.In this project, we focused on increasing the uniformity of emitter tip radii as a means to reduce tip burnout. Figure 1 shows a typical distribution of emitter tip radii for FEAs. The non-uniform distribution of emitter tip radii first forms during the photolithography step that defines the array of “dots” that become the etching mask for the silicon tips. In our FEA fabrication process, we used a trilevel resist process that nearly eliminated the light wave reflected at the photoresist/silicon interface and hence improved the uniformity of the dot diameter. Furthermore, we integrated the emitter tips with silicon nanowires to improve their reliability. Figure 2 shows a diagram of the fabricated structure. Our fabrication process resulted in FEAs with more uniform emission current and potentially a longer lifetime.

Development of a Subnanometer-Precision Scanning Anode Field Emission Microscope

Field emitter arrays (FEA) have not found widespread adoption in demanding applications such as THz, RF, Deep UV, X-ray, electron, ion, and neutron sources, where high current (>1 mA) and long operating lifetime (>10,000 hrs.) are required. These limitations arise as a result of the sensitivity of emitted electron and ion cur-rent to the spatial non-uniformity of emitter tip radi-us and the temporal non-uniformity of work function due to adsorption and desorption of gas molecules at the surface of the emitter tips. These non-uniformi-ties result in the variation of the field factor (β), a key performance parameter for field emitters and ioniz-ers. Variations in β result in severe underutilization of emitter tips as only few tips with large β contributes to emission or ionization current. These tips, if not pro-tected, burn out, leading to low emission current and very short operation lifetime. Emission current and operational lifetime of a FEA could thus be improved by making emitters with more uniform characteristics. We are developing a subnanometer-precision scanning anode field-emission microscope (SAFEM) that could be used to probe the fundamental processes in the operation of emitter tips of FEA. The SAFEM is designed to precisely and accurately position, with subnanometer resolution, a probing anode over the tips of an FEA and in scanning mode directly acquire the spatial map of emission tip current. From the map, other characteristics of the emitter tip such as anode voltage, tip radius, density, and field factor can be extracted. The map of extracted parameters could yield insight into the operation of the FEAs. Also, a statistical distribution of the field factor will enable study of the dependence of tip characteristics on the fabrication process and thereby enable exploration of novel process for engineering high-performance FEAs with high current densities and long operational lifetimes.