Engineering topics
Cao, Yu
Publications and source records attributed to Cao, Yu.
Abisko: Deep codesign of an architecture for spiking neural networks using novel neuromorphic materials
The Abisko project aims to develop an energy-efficient spiking neural network (SNN) computing architecture and software system capable of autonomous learning and operation. The SNN architecture explores novel neuromorphic devices that are based on resistive-switching materials, such as memristors and electrochemical RAM. Equally important, Abisko uses a deep codesign approach to pursue this goal by engaging experts from across the entire range of disciplines: materials, devices and circuits, architectures and integration, software, and algorithms. Here, the key objectives of our Abisko project are threefold. First, we are designing an energy-optimized high-performance neuromorphic accelerator based on SNNs. This architecture is being designed as a chiplet that can be deployed in contemporary computer architectures and we are investigating novel neuromorphic materials to improve its design. Second, we are concurrently developing a productive software stack for the neuromorphic accelerator that will also be portable to other architectures, such as field-programmable gate arrays and GPUs. Third, we are creating a new deep codesign methodology and framework for developing clear interfaces, requirements, and metrics between each level of abstraction to enable the system design to be explored and implemented interchangeably with execution, measurement, a model, or simulation. As a motivating application for this codesign effort, we target the use of SNNs for an analog event detector for a high-energy physics sensor.
FPGA Acceleration of GCN in Light of the Symmetry of Graph Adjacency Matrix
Graph Convolutional Neural Networks (GCNs) are widely used to process large-scale graph data. Different from deep neural networks (DNNs), GCNs are sparse, irregular, and unstructured, posing unique challenges to hardware acceleration with regular processing elements (PEs). In particular, the adjacency matrix of a GCN is extremely sparse, leading to frequent but irregular memory access, low spatial/temporal data locality and poor data reuse. Furthermore, a realistic graph usually consists of unstructured data (e.g., unbalanced distributions), creating significantly different processing times and imbalanced workload for each node in GCN acceleration. To overcome these challenges, we propose an end-to-end hardware-software co-design to accelerate GCNs on resource-constrained FPGAs with the features including: (1) A custom dataflow that leverages symmetry along the diagonal of the adjacency matrix to accelerate feature aggregation for undirected graphs. We utilize either the upper or the lower triangular matrix of the adjacency matrix to perform aggregation in GCN to improve data reuse. (2) Unified compute cores for both aggregation and transform phases, with full support to the symmetry-based dataflow. These cores can be dynamically reconfigured to the systolic mode for transformation or as individual accumulators for aggregation in GCN processing. (3) Preprocessing of the graph in software to rearrange the edges and features to match the custom dataflow. This step improves the regularity in memory access and data reuse in the aggregation phase. Moreover, we quantize the GCN precision from FP32 to INT8 to reduce the memory footprint without losing the inference accuracy. We implement our accelerator design in Intel Stratix10 MX FPGA board with HBM2, and demonstrate 1.3x-110.5x improvement in end-to-end GCN latency as compared to the state-of the-art FPGA implementations, on the graph datasets of Cora, Pubmed, Citeseer and Reddit.
Allosteric autoregulation of DNA binding via a DNA-mimicking protein domain: a biophysical study of ZNF410–DNA interaction using small angle X-ray scattering
ZNF410 is a highly-conserved transcription factor, remarkable in that it recognizes a 15-base pair DNA element but has just a single responsive target gene in mammalian erythroid cells. ZNF410 includes a tandem array of five zinc-fingers (ZFs), surrounded by uncharacterized N- and C-terminal regions. Unexpectedly, full-length ZNF410 has reduced DNA binding affinity, compared to that of the isolated DNA binding ZF array, both in vitro and in cells. AlphaFold predicts a partially-folded N-terminal subdomain that includes a 30-residue long helix, preceded by a hairpin loop rich in acidic (aspartate/glutamate) and serine/threonine residues. This hairpin loop is predicted by AlphaFold to lie against the DNA binding interface of the ZF array. In solution, ZNF410 is a monomer and binds to DNA with 1:1 stoichiometry. Surprisingly, the single best-fit model for the experimental small angle X-ray scattering profile, in the absence of DNA, is the original AlphaFold model with the N-terminal long-helix and the hairpin loop occupying the ZF DNA binding surface. For DNA binding, the hairpin loop presumably must be displaced. After combining biophysical, biochemical, bioinformatic and artificial intelligence-based AlphaFold analyses, we suggest that the hairpin loop mimics the structure and electrostatics of DNA, and provides an additional mechanism, supplementary to sequence specificity, of regulating ZNF410 DNA binding.
Graph-based Compact Modeling (GCM) of CMOS transistors for efficient parameter extraction: A machine learning approach
Parameter extraction of compact transistor models is an expensive process, heavily relying on engineering knowledge and experience. To automate such a process, we propose a novel approach, Graph-based Compact Model (GCM), that integrates physical modeling and data-driven learning. GCM utilizes Graph Neural Networks (GNNs) to establish the model structure, while retaining the physicality in compact models. Here, we implement our GCM in Verilog-A to support circuit simulations. As demonstrated with an academic 7 nm FinFET PDK, the new approach automatically generates a GCM model within a minute, and achieves excellent accuracy and efficiency in SPICE.
DeePKS + ABACUS as a Bridge between Expensive Quantum Mechanical Models and Machine Learning Potentials
Recently, the development of machine learning (ML) potentials has made it possible to perform large-scale and long-time molecular simulations with the accuracy of quantum mechanical (QM) models. However, for different levels of QM methods, such as density functional theory (DFT) at the meta-GGA level and/or with exact exchange, quantum Monte Carlo, etc., generating a sufficient amount of data for training an ML potential has remained computationally challenging due to their high cost. In this work, we demonstrate that this issue can be largely alleviated with Deep Kohn–Sham (DeePKS), an ML-based DFT model. DeePKS employs a computationally efficient neural network-based functional model to construct a correction term added upon a cheap DFT model. Upon training, DeePKS offers closely matched energies and forces compared with high-level QM method, but the number of training data required is orders of magnitude less than that required for training a reliable ML potential. As such, DeePKS can serve as a bridge between expensive QM models and ML potentials: one can generate a decent amount of high-accuracy QM data to train a DeePKS model and then use the DeePKS model to label a much larger amount of configurations to train an ML potential. Further, this scheme for periodic systems is implemented in a DFT package ABACUS, which is open source and ready for use in various applications.
Doped gate dielectrics materials
A field effect transistor having at least a gate, source, and drain electrodes and a semiconductor channel for controlling transport of charge carriers between the source and drain electrodes, the gate being insulated from the channel by an dielectric, at least a portion of the dielectric disposed between the gate electrode and the semiconductor channel being doped or imbued with the an element which if doped or imbued into a semiconductor material would cause the semiconductor to be p-type. The p-type element used to dope or imbue the gate dielectric is preferably Mg.
Gradient-Based Novelty Detection Boosted by Self-Supervised Binary Classification
Novelty detection aims to automatically identify out-of-distribution (OOD) data, without any prior knowledge of them. It is a critical step in data monitoring, behavior analysis and other applications, helping enable continual learning in the field. Conventional methods of OOD detection perform multi-variate analysis on an ensemble of data or features, and usually resort to the supervision with OOD data to improve the accuracy. In reality, such supervision is impractical as one cannot anticipate the anomalous data. In this paper, we propose a novel, self-supervised approach that does not rely on any pre-defined OOD data: (1) The new method evaluates the Mahalanobis distance of the gradients between the in-distribution and OOD data. (2) It is assisted by a self-supervised binary classifier to guide the label selection to generate the gradients, and maximize the Mahalanobis distance. In the evaluation with multiple datasets, such as CIFAR-10, CIFAR-100, SVHN and TinyImageNet, the proposed approach consistently outperforms state-of-the-art supervised and unsupervised methods in the area under the receiver operating characteristic (AUROC) and area under the precision-recall curve (AUPR) metrics. We further demonstrate that this detector is able to accurately learn one OOD class in continual learning.
Understanding and Manipulating Helical Nanofilaments in Binary Systems with Achiral Dopants
Here, in this paper, we report the relationship between helical pitch of the helical nanofilament (HNF) phase formed by bent-core molecule NOBOW and the concentration of achiral dopants 5CB and octane, using linearly polarized resonant soft X-ray scattering (RSoXS). Utilizing theory-based simulation, which fits well with the experiments, the molecular helices in the filament were probed and the superstructure of helical 5CB directed by groove of HNFs was observed. Quantitative pitch determination with RSoXS reveals that helical pitch variation is related to 5CB concentration with no temperature dependence. Doping rodlike immiscible 5CB led to a pitch shortening of up to 30%, which was attributed to a change in interfacial tension. By shedding light not only on phase behavior of binary systems but also enabling control over pitch length, our work may benefit various applications of HNF-containing binary systems, including optical rotation devices, circularly polarized light emitters, and chirality transfer agents.
Supramolecular meso -Trick: Ambidextrous Mirror Symmetry Breaking in a Liquid Crystalline Network with Tetragonal Symmetry
Bicontinuous and multicontinuous network phases are among nature's most complex structures in soft matter systems. Here, a chiral bicontinuous tetragonal phase is reported as a new stable liquid crystalline intermediate phase at the transition between two cubic phases, the achiral double gyroid and the chiral triple network cubic phase with an I 23 space group, both formed by dynamic networks of helices. The mirror symmetry of the double gyroid, representing a meso -structure of two enantiomorphic networks, is broken at the transition to this tetragonal phase by retaining uniform helicity only along one network while losing it along the other one. This leads to a conglomerate of enantiomorphic tetragonal space groups, P 4 1 2 1 2 and P 4 3 2 1 2. Phase structures and chirality were analyzed by small-angle X-ray scattering (SAXS), grazing-incidence small-angle X-ray scattering (GISAXS), resonant soft X-ray scattering (RSoXS) at the carbon K-edge, and model-dependent SAXS/RSoXS simulation. In conclusion, our findings not only lead to a new bicontinuous network-type three-dimensional mesophase but also reveal a mechanism of mirror symmetry breaking in soft matter by partial meso -structure racemization at the transition from enantiophilic to enantiophobic interhelical self-assembly.
Evolutionary NAS in Light of Model Stability for Accurate Continual Learning
Continual learning, the capability to learn new knowledge from streaming data without forgetting the previous knowledge, is a critical requirement for dynamic learning systems, especially for emerging edge devices such as self-driving cars and drones. However, continual learning is still facing the catastrophic forgetting problem. Previous work illustrate that model performance on continual learning is not only related to the learning algorithms but also strongly dependent on the inherited model, i.e., the model where continual learning starts. The better stability of the inherited model, the less catastrophic forgetting and thus, the inherited model should be elaborately selected. Inspired by this finding, we develop an evolutionary neural architecture search (ENAS) algorithm that emphasizes the Stability of the inherited model, namely ENAS-S. ENAS-S aims to find optimal architectures for accurate continual learning on edge devices. On CIFAR-10 and CIFAR-100, we present that ENAS-S achieves competitive architectures with lower catastrophic forgetting and smaller model size when learning from a data stream, as compared with handcrafted DNNs.
Deciphering helix assembly in the heliconical nematic phase via tender resonant X-ray scattering
Being a link between a uniaxial nematic and a chiral nematic, the twist bend nematic (N TB ) has been an intriguing topic for research over the last decade as a key to understanding chirality generation. Accurate description and understanding of resonance effects in helical structures provide crucial knowledge on the phase behavior beyond positional ordering. We examined the manifestation of resonance effects in N TB via tender resonant X-ray scattering (TReXS) at the sulfur K-edge. For the first time we demonstrate quantitatively that the energy dependence of the scattering peak in the N TB phase follows the energy dependence of the complex refractive indices measured by X-ray absorption. Furthermore, taking advantage of molecular energy calculation, we decipher the distinct helical pitch variation trend for symmetric/asymmetric dimers from the perspective of hybridization and reveal the strong effect of substituting oxygen for sulfur atoms on the helical pitch and the thermal helix extension.
Development of Gallium Nitride Monolithic Microwave Integrated Circuits for Ka-Band Remote Sensing
We have been developing gallium nitride (GaN)monolithic microwave integrated circuits (MMICs) towards thegoal of implementing a single chip transceiver frontend forcloud radar arrays at Ka-band. We present our first iterationdesign, fabrication and test cycle implementation of GaN poweramplifiers (PAs), low noise amplifiers (LNAs), driver amplifier,single-pole- double and triple throw switches, and voltagecontrolled attenuator. These components are necessary forimplementing a transceiver unit cell concept with digitalcalibration capabilities to enable scalable array sizes for radarand radiometry.
Doped gate dielectric materials
A field effect transistor having at least a gate, source, and drain electrodes and a semiconductor channel for controlling transport of charge carriers between the source and drain electrodes, the gate being insulated from the channel by an dielectric, at least a portion of the dielectric disposed between the gate electrode and the semiconductor channel being doped or imbued with the an element which if doped or imbued into a semiconductor material would cause the semiconductor to be p-type. The p-type element used to dope or imbue the gate dielectric is preferably Mg.
A Redox-Active 2D Metal-Organic Framework for Efficient Lithium Storage with Extraordinary High Capacity
Explore the source record for details and available documents.
Efficient Perovskite Solar Cells Fabricated by Co Partially Substituted Hybrid Perovskite
Abstract not provided
Y-shaped tricatenar azobenzenes – functional liquid crystals with synclinic–anticlinic transitions and spontaneous helix formation
A series of achiral tricatenar rod-like molecules with a 3,5-disubstitution pattern at one end and a single alkyl chain at the other end of a rod-like azobenzene derived core is reported. Depending on temperature and alkyl chain length, these Y-shaped compounds self-assemble into different types of liquid crystalline (LC) phases, ranging from non-tilted and synclinic tilted hexatic, via non-tilted and anticlinic tilted smectic and bicontinuous cubic LC phases, to a spontaneous mirror symmetry broken isotropic liquid (Iso 1 [*]) or a related achiral liquid network phase (Iso 1 ). An additional tilted, but uniaxial smectic phase was observed at the transition between anticlinic and synclinic tilt correlation and was investigated by soft resonant X-ray scattering with respect to possible helix formation. This work provides a new concept for the design of technological interesting azobenzene based LC materials with anticlinic tilted smectic C phases (SmC a ) and with azobenzene units organized in the long range or short range helical network structures of bicontinuous cubic and chiral isotropic liquid phases, respectively. Core fluorination removes all lamellar phases, leaving only the cubic phase over wide temperature ranges, even at ambient temperature.