Engineering Papers⌕ Search

Engineering topics

Liu, Hang

Publications and source records attributed to Liu, Hang.

At least 19 records

Demystifying the Resilience of Large Language Models: An End-to-End Perspective

Deep neural networks are known to be resilient to random bit-wise faults in their parameters. However, this resilience has primarily been established through evaluations of classification models. The extent to which this claim holds for large-language models remains underexplored. In this work, we conduct an extensive measurement study on the impact of random bitwise faults in commercial-scale language models. We perform an in-depth analysis of the resulting generation outputs. We first expose that these language models are not truly resilient to random bit-flips. While aggregate metrics such as accuracy may suggest resilience, an in-depth inspection of the generated outputs shows significant degradation in text quality. Our analysis also shows that tasks requiring more complex reasoning suffer more from performance and quality degradation. Moreover, we extend our analysis to models with augmented reasoning capabilities, such as Chain-of-Thought or Mixture of Experts architectures, and characterize their failure scenarios under random bit-flips.

Sun, Yu↗

Establish the basis for Breadth-First Search on Frontier System: XBFS on AMD GPUs

Graphics Processing Units (GPUs) offer significant potential for accelerating various computational tasks, including Breadth-First Search (BFS). Numerous efforts have been made to deploy BFS on GPUs effectively. To address the dynamic nature of BFS, XBFS, the state-of-the-art work, employs an adaptive strategy that leverages different optimized frontier queue generation designs, accommodating the varying characteristics of levels in BFS. While XBFS demonstrates excellent performance on NVIDIA Quadro P6000 GPUs, it faces challenges when deployed on AMD GPUs. In this work, we present our efforts to implement XBFS’s adaptive approach on Frontier, the most powerful supercomputer system, by porting XBFS to AMD MI250X GPUs. Through targeted optimizations tailored to the unique features of AMD GPUs, our implementation achieves an average performance of 43 Giga-Traversed Edges Per Second (GTEPS) per Graphics Compute Dies (GCD). Based on these results, we observe potential for surpassing the performance of the official Frontier results from the Graph500 benchmark released in June 2024.

Yang, Haoshen↗

Growth of Mesoscale Ordered Two-Dimensional Hydrogen-Bond Organic Framework with the Observation of Flat Band

Flat bands (FBs), presenting a strongly interacting quantum system, have drawn increasing interest recently. However, experimental growth and synthesis of FB materials have been challenging and have remained elusive for the ideal form of monolayer materials where the FB arises from destructive quantum interference as predicted in 2D lattice models. Here, we report surface growth of a self-assembled monolayer of 2D hydrogen-bond (H-bond) organic frameworks (HOFs) of 1,3,5-tris(4-hydroxyphenyl)benzene (THPB) on Au(111) substrate and the observation of FB. High-resolution scanning tunneling microscopy or spectroscopy shows mesoscale, highly ordered, and uniform THPB HOF domains, while angle-resolved photoemission spectroscopy highlights a FB over the whole Brillouin zone. Density-functional-theory calculations and analyses reveal that the observed topological FB arises from a hidden electronic breathing-kagome lattice without atomically breathing bonds. In conclusion, our findings demonstrate that self-assembly of HOFs provides a viable approach for synthesis of 2D organic topological materials, paving the way to explore many-body quantum states of topological FBs.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

A Length Adaptive Algorithm-Hardware Co-design of Transformer on FPGA Through Sparse Attention and Dynamic Pipelining

Transformers are considered one of the most important deep learning models since 2018, in part because it establishes state-of-the-art (SOTA) records and could potentially replace existing Deep Neural Networks (DNNs). Despite the remarkable triumphs, the prolonged turnaround time of Transformer models is a widely recognized roadblock. The variety of sequence lengths imposes additional computing overhead where inputs need to be zero-padded to the maximum sentence length in the batch to accommodate the parallel computing platforms. This paper targets the field-programmable gate array (FPGA) and proposes a coherent sequence length adaptive algorithm–hardware co-design for Transformer acceleration. Particularly, we develop a hardware-friendly sparse attention operator and a length-aware hardware resource scheduling algorithm. The proposed sparse attention operator brings the complexity of attention-based models down to linear complexity and alleviates the off-chip memory traffic. The proposed length-aware resource hardware scheduling algorithm dynamically allocates the hardware resources to fill up the pipeline slots and eliminates bubbles for NLP tasks. Experiments show that our design has very small accuracy loss and has 80.2 × and 2.6 × speedup compared to CPU and GPU implementation, and 4 × higher energy efficiency than state-of-the-art GPU accelerator optimized via CUBLAS GEMM.

Peng, Hongwu↗

Excited quantum anomalous and spin Hall effect: dissociation of flat-bands-enabled excitonic insulator state

Quantum anomalous Hall effect (QAHE) and quantum spin Hall effect (QSHE) are two interesting physical manifestations of 2D materials that have an intrinsic nontrivial band topology. In principle, they are ground-state equilibrium properties characterized by Fermi level lying in a topological gap, below which all the occupied bands are summed to a non-zero topological invariant. In this work, we propose theoretical concepts and models of ‘excited’ QAHE (EQAHE) and EQSHE generated by dissociation of an excitonic insulator (EI) state with complete population inversion (CPI), a unique many-body ground state enabled by two yin-yang flat bands (FBs) of opposite chirality hosted in a diatomic Kagome lattice. The two FBs have a trivial gap in between, i.e. the system is a trivial insulator in the single-particle ground-state, but nontrivial gaps above and below, so that upon photoexcitation the quasi-Fermi levels of both electrons and holes will lie in a nontrivial gap achieved by the CPI-EI state, as demonstrated by exact diagonalization calculations. Then dissociation of singlet and triplet EI state will lead to EQAHE and EQSHE, respectively. Realizations of yin-yang FBs in real materials are also discussed.

77 NANOSCIENCE AND NANOTECHNOLOGY↗

SimNet: Accurate and High-Performance Computer Architecture Simulation using Deep Learning

While cycle-accurate simulators are essential tools for architecture research, design, and development, their practicality is limited by an extremely long time-to-solution for realistic applications under investigation. This work describes a concerted effort, where machine learning (ML) is used to accelerate microarchitecture simulation. First, an ML-based instruction latency prediction framework that accounts for both static instruction properties and dynamic processor states is constructed. Then, a GPU-accelerated parallel simulator is implemented based on the proposed instruction latency predictor, and its simulation accuracy and throughput are validated and evaluated against a state-of-the-art simulator. Leveraging modern GPUs, the ML-based simulator outperforms traditional CPU-based simulators significantly.

97 MATHEMATICS AND COMPUTING↗

High-temperature fractional quantum Hall state in the Floquet kagome flat band

A fractional quantum Hall effect (FQHE) has been predicted in a topological flat band (FB) by a single-particle band structure combined with phenomenological theory or solution of a many-body lattice Hamiltonian with fuzzy parameters. A long-standing roadblock toward the realization of a FB-FQHE is lacking the many-body solution of specific materials under realistic conditions. Here, we demonstrate a combined study of single-particle Floquet band theory with exact diagonalization (ED) of a many-body Hamiltonian. We show that a time-periodic circularly polarized laser inverts the sign of second-nearest-neighbor hopping in a kagome lattice and enhances spin-orbit coupling in one spin channel to produce a Floquet FB with a high flatness ratio of bandwidth over band gap, as exemplified in monolayer Pt 3 C 36 S 12 H 12 . The ED of the resultant Floquet-kagome lattice Hamiltonian gives a one-third-filling ground state with a laser-dependent excitation gap of a FQH state, up to an estimated temperature above 70 K. Our findings pave the way for exploring the alluding high-temperature FB-FQHE.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

GSoFa: Scalable Sparse Symbolic LU Factorization on GPUs

Decomposing a matrix $\mathbf {A}$ into a lower matrix $\mathbf {L}$ and an upper matrix $\mathbf {U}$, which is also known as LU decomposition, is an essential operation in numerical linear algebra. For a sparse matrix, LU decomposition often introduces more nonzero entries in the $\mathbf {L}$ and $\mathbf {U}$ factors than in the original matrix. A symbolic factorization step is needed to identify the nonzero structures of $\mathbf {L}$ and $\mathbf {U}$ matrices. Attracted by the enormous potentials of the Graphics Processing Units (GPUs), an array of efforts have surged to deploy various LU factorization steps except for the symbolic factorization, to the best of our knowledge, on GPUs. This article introduces gSoFa, the first GPU-based symbolic factorization design with the following three optimizations to enable scalable LU symbolic factorization for nonsymmetric pattern sparse matrices on GPUs. First, here we introduce a novel fine-grained parallel symbolic factorization algorithm that is well suited for the Single Instruction Multiple Thread (SIMT) architecture of GPUs. Second, we tailor supernode detection into a SIMT friendly process and strive to balance the workload, minimize the communication and saturate the GPU computing resources during supernode detection. Third, we introduce a three-pronged optimization to reduce the excessive space consumption problem faced by multi-source concurrent symbolic factorization. Taken together, gSoFa achieves up to 31× speedup from 1 to 44 Summit nodes (6 to 264 GPUs) and outperforms the state-of-the-art CPU project, on average, by 5×. Notably, gSoFa also achieves up to 47 percent of the peak memory throughput of a V100 GPU in the Summit Supercomputer.

97 MATHEMATICS AND COMPUTING↗

Orbital design of flat bands in non-line-graph lattices via line-graph wave functions

Line-graph (LG) lattices are known for having flat bands (FBs) from the destructive interference of Bloch wave functions encoded in only lattice symmetry. In this work, we develop a generic atomic/molecular orbital design principle for FBs in non-LG lattices. Based on linear combination of atomic orbital theory, we demonstrate that the underlying wave-function symmetry of FBs in a LG lattice can be transformed into the atomic/molecular orbital symmetry in a non-LG lattice. We illustrate such orbital-designed topological FBs in three 2D non-LG, square, trigonal, and hexagonal lattices, where the designed orbitals faithfully reproduce the corresponding lattice symmetries of checkerboard, kagome, and diatomic-kagome lattices, respectively. Interestingly, systematic design of FBs with a high Chern number is also achieved based on the same principle. Fundamentally our theory enriches the FB physics; practically, it significantly expands the scope of FB materials, since most materials have multiple atomic/molecular orbitals at each lattice site, rather than a single s orbital mandated in graph theory and generic lattice models.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

A General Strategy to Immobilize Single‐Atom Catalysts in Metal–Organic Frameworks for Enhanced Photocatalysis

Abstract Single‐atom catalysts (SACs) are witnessing rapid development due to their high activity and selectivity toward diverse reactions. However, it remains a grand challenge in the general synthesis of SACs, particularly featuring an identical chemical microenvironment and on the same support. Herein, a universal synthetic protocol is developed to immobilize SACs in metal–organic frameworks (MOFs). Significantly, by means of SnO 2 as a mediator or adaptor, not only different single‐atom metal sites, such as Pt, Cu, and Ni, etc., can be installed, but also the MOF supports can be changed (for example, UiO‐66‐NH 2 , PCN‐222, and DUT‐67) to afford M 1 /SnO 2 /MOF architecture. Taking UiO‐66‐NH 2 as a representative, the Pt 1 /SnO 2 /MOF exhibits approximately five times higher activity toward photocatalytic H 2 production than the corresponding Pt nanoparticles (≈2.5 nm) stabilized by SnO 2 /UiO‐66‐NH 2 . Remarkably, despite featuring identical parameters in the chemical microenvironment and support in M 1 /SnO 2 /UiO‐66‐NH 2 , the Pt 1 catalyst possesses a hydrogen evolution rate of 2167 µmol g –1 h –1 , superior to the Cu 1 and Ni 1 counterparts, which is attributed to the differentiated hydrogen binding free energies, as supported by density‐functional theory (DFT) calculations. This is thought to be the first report on a universal approach toward the stabilization of SACs with identical chemical microenvironment on an identical support.

Sui, Jianfei↗

Optimizing FPGA-based Accelerator Design for Large-Scale Molecular Similarity Search (Special Session Paper)

Molecular similarity search has been widely used in drug discovery to rapidly identify structurally similar compounds from large molecular databases. With the increasing size of chemical libraries, there is growing interest in the efficient ac- celeration of large-scale similarity search. Existing works mainly focus on CPU and GPU to accelerate the computation of Tatimoto coefficient in measuring the pairwise similarity between different molecular fingerprints. In this paper, we propose and optimize an FPGA-based accelerator design on exhaustive and approximate search algorithms. On exhaustive search using BitBound & fold- ing, we analyze the similarity cutoff and folding level relationship with search speedup and accuracy, and propose a scalable on- the-fly query engine on FPGAs to reduce the resource utilization and pipeline interval. We achieve a 450 million compounds-per- second processing throughput for a single query engine. On approximate search using hierarchical navigable small world (HNSW), a popular algorithm with high recall and query speed, we propose an FPGA-based graph traversal engine to utilize high throughput register array based priority queue and fine- grained distance calculation engine to increase the processing capability. Experimental results show that the proposed FPGA- based HNSW implementation achieves a 35× speedup than existing works on CPU. To the best of our knowledge, our FPGA- based implementation is the first attempt to accelerate molecular similarity search on FPGA and has the highest performance among existing approaches.

Peng, Hongwu↗

Non-Hermitian topological states in 2D line-graph lattices: evolving triple exceptional points on reciprocal line graphs

Non-Hermitian (NH) topological states, such as the doubly-degenerate nodes dubbed as exceptional points (EPs) in Bloch band structure of 2D lattices driven by gain and loss, have attracted much recent interest. We demonstrate theoretically that in the three-site edge-centered lattices, i.e. the so-called line-graph lattices, such as kagome lattice which is a line graph of hexagonal lattice, there exist three types of triply-degenerate EPs evolving intriguingly on another set of line graphs in the reciprocal space. A single TEP (STEP) with ±1/3 topological charge moves faithfully along the edges of reciprocal line graphs with varying gain and loss, while two STEPs merge distinctively into one unconventional orthogonal double TEP (DTEP) with ±2/3 charge at the vertices, which is characterized with two ordinary self-orthogonal eigenfunctions but one surprising 'orthogonal' eigenfunction. Differently, in a modified line-graph lattice with an off-edge-center site, the ordinary coalesced state of DTEPs emerges with three identical self-orthogonal eigenfunctions. Such NH states and their evolution can be generally realized in various artificial systems, such as photonic and sonic crystals, where light and sonic vortex beams with different fractional twisting can be found. Our findings shed new light on fundamental understanding of gapless topological states in NH systems in terms of creation and evolution of high-order EPs, and open up new research directions to further link line graph and flow network theory coupled with topological physics, especially under non-equilibrium gain/loss conditions.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Binary Complex Neural Network Acceleration on FPGA

Being able to learn from complex data with phase information is imperative for many signal processing applications. Today’s real-valued deep neural networks (DNNs) have shown efficiency in latent information analysis but fall short when applied to the complex domain. Deep complex networks (DCN) , in contrast, can learn from complex data, but have high computational costs; therefore, they cannot satisfy the instant decision making requirements of many deployable systems dealing with short observations or short signal bursts. Recent, Binarized Complex Neural Network (BCNN), which integrates DCNs with binarized neural networks (BNN), shows great potential in classifying complex data in real-time. In this paper, we propose a structural pruning based accelerator of BCNN, which is able to provide more than 5000 frames/s inference throughput on edge devices. The high performance comes from both the algorithm and hardware sides. On the algorithm side, we conduct structural pruning to the original BCNN models and obtain 20 × pruning rates with negligible accuracy loss; on the hardware side, we propose a novel 2D convolution operation accelerator for the binary complex neural network. Experimental results show that the proposed design works with over 90% utilization and is able to achieve the inference throughput of 5882 frames/s and 4938 frames/s for complex NIN-Net and ResNet-18 using CIFAR-10 dataset and Alveo U280 Board.

Peng, Hongwu↗

Screening two-dimensional materials with topological flat bands

The topological flat band (TFB) has been proposed theoretically in various lattice models, to exhibit a rich spectrum of intriguing physical behaviors. However, the experimental demonstration of flat band (FB) properties has been severely hindered by the lack of materials realization. In this study, by screening materials from a first-principles materials database, we identify a group of two-dimensional materials with TFBs near the Fermi level, covering some simple line-graph and generalized line-graph FB lattice models. These include the kagome sublattice of O in Ti O 2 yielding a spin-unpolarized TFB, and that of V in ferromagnetic V 3 F 8 yielding a spin-polarized TFB. The monolayer Nb 3 Te Cl 7 and its counterparts from element substitution are found to be breathing-kagome-lattice crystals. The family of monolayer II I 2 V I 3 compounds exhibit a TFB representing the coloring-triangle lattice model. Re F 3 , Mn F 3 , and Mn Br 3 are all predicted to be diatomic-kagome-lattice crystals, with TFB transitions induced by atomic substitution. Finally, Hg F 2 , Cd F 2 , and Zn F 2 are discovered to host dual TFBs in the diamond-octagon lattice. Our findings pave the way to further experimental exploration of eluding FB materials and properties.

36 MATERIALS SCIENCE↗

Viral Load Kinetics of Severe Acute Respiratory Syndrome Coronavirus 2 in Hospitalized Individuals With Coronavirus Disease 2019

Abstract Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) kinetics remain understudied, including the impact of remdesivir. In hospitalized individuals, peak sputum viral load occurred in week 2 of symptoms, whereas viremia peaked within 1 week of symptom-onset, suggesting early systemic seeding of SARS-CoV-2. Remdesivir treatment was associated with faster viral decay.

60 APPLIED LIFE SCIENCES↗