Binary Architecture Research Articles

NAS-BNN: Neural Architecture Search for Binary Neural Networks

Binary Neural Networks (BNNs) have gained extensive attention for their superior inferencing efficiency and compression ratio compared to traditional full-precision networks. However, due to the unique characteristics of BNNs, designing a powerful binary architecture is challenging and often requires significant manpower. A promising solution is to utilize Neural Architecture Search (NAS) to assist in designing BNNs, but current NAS methods for BNNs are relatively straightforward and leave a performance gap between the searched models and manually designed ones. To address this gap, we propose a novel neural architecture search scheme for binary neural networks, named NAS-BNN. We first carefully design a search space based on the unique characteristics of BNNs. Then, we present three training strategies, which significantly enhance the training of supernet and boost the performance of all subnets. Our discovered binary model family outperforms previous BNNs for a wide range of operations (OPs) from 20M to 200M. For instance, we achieve 68.20% top-1 accuracy on ImageNet with only 57M OPs. In addition, we validate the transferability of these searched BNNs on the object detection task, and our binary detectors with the searched BNNs achieve a novel state-of-the-art result, e.g., 31.6% mAP with 370M OPs, on MS COCO dataset. The source code and models will be released at https://github.com/VDIGPKU/NAS-BNN.

Escalation of genome defense capacity enables control of an expanding meiotic driver.

From RNA interference to chromatin silencing, diverse genome defense pathways silence selfish genetic elements to safeguard genome integrity 1,2 . Despite their diversity, different defense pathways share a modular organization, where numerous specificity factors identify diverse targets and common effectors silence them. In the PIWI-interacting RNA (piRNA) pathway, which controls selfish elements in the metazoan germline, diverse target RNAs are first identified by complementary base pairing with piRNAs and then silenced by PIWI-clade nucleases via enzymatic cleavage 1,3 . Such a binary architecture allows the defense systems to be readily adaptable, where new targets can be captured via the innovation of new specificity factors 4,5 . Thus, our current understanding of genome defense against lineage-specific selfish genes has been largely limited to the evolution of specificity factors, while it remains poorly understood whether other types of innovations are required. Here, we describe a new type of innovation, which escalates the defense capacity of the piRNA pathway to control a recently expanded selfish gene in Drosophila melanogaster . Through an in vivo RNAi screen for repressors of Stellate -a recently evolved and expanded selfish meiotic driver 6-8 -we discovered a novel defense factor, Trailblazer. Trailblazer is a transcription factor that promotes the expression of two PIWI-clade nucleases, Aub and AGO3, to match Stellate in abundance. Recent innovation in the DNA-binding domain of Trailblazer enabled it to drastically elevate Aub and AGO3 expression in the D. melanogaster lineage, thereby escalating the silencing capacity of the piRNA pathway to control expanded Stellate and safeguard fertility. As copy-number expansion is a recurrent feature of diverse selfish genes across the tree of life 9-12 , we envision that augmenting the defense capacity to quantitatively match selfish genes is likely a repeatedly employed defense strategy in evolution.

Enzyme-free intelligent films in electrochemistry of malathion with a binary architecture: Combined Cu2+ complexed carbon nitride nanosheets and poly(N,N-dimethylacrylamide) hydrogels

In the present work, surface and interfacial modifications have been used to successfully prepare a low-cost enzyme-free electrochemical sensor was constructed and assembled a smart binary hydrogel film of Cu@g-C3N4 NSs/PDEA (g-C3N4 NSs: graphitic C3N4 nanosheets; PDEA: poly(N,N- dimethylacrylamide). On account of g-C3N4 NSs have multiple uniform nitrogen coordination sites, making it easy to adsorb metal ions. Complexation of Cu2+ to g-C3N4 NSs enhances the electrical conductivity while increasing the specific surface area of the nanomaterials. They are used for trace monitoring of malathion. It showed good stability when detecting malathion in synthetic urine. A 4-Input/7-Output logic gate system including OR, AND and NOT is constructed using membrane electrode switching behavior for pH, temperature, SO42–, and methanol. And a 2-to-4 decoder was constructed. The complexity of non-traditional computer systems is enriched by this logic gate system about contaminants.

Tracing the evolution of short-period binaries with super-synchronous fast rotators

Context. The initial distribution of rotational velocities of stars is still poorly known, and how the stellar spin evolves from birth to the various end points of stellar evolution is an actively debated topic. Binary interactions are often invoked to explain the existence of extremely fast-rotating stars (vsin i ≳ 200 km s−1). The primary mechanisms through which binaries can spin up stars are tidal interactions, mass transfer, and possibly mergers. However, fast rotation could also be primordial, that is, a result of the star formation process. To evaluate these scenarios, we investigated in detail the evolution of three known fast-rotating stars in short-period spectroscopic and eclipsing binaries, namely HD 25631, HD 191495, and HD 46485, with primaries of masses of 7, 15, and 24 M⊙, respectively, with companions of ∼1 M⊙ and orbital periods of less than 7 days. These systems belong to a recently identified class of binaries with extreme mass ratios, whose evolutionary origin is still poorly understood. Aims. We evaluated in detail three scenarios that could explain the fast rotation observed in these binaries: it could be primordial, a product of mass transfer, or the result of a merger within an originally triple system. We also discuss the future evolution of these systems to shed light on the impact of fast rotation on binary products. Methods. We computed grids of single and binary MESA models varying tidal forces and initial binary architectures to investigate the evolution and reproduce observational properties of these systems. When considering the triple scenario, we determined the region of parameter space compatible with the observed binaries and used a publicly available machine-learning model to determine the dynamical stability of the triple system. Results. We find that, because of the extreme mass-ratio between binary components, tides have a limited impact, regardless of the prescription used, and that the observed short orbital periods are at odds with post-mass-transfer scenarios. We also find that the overwhelming majority of triple systems compatible with the observed binaries are dynamically unstable and would be disrupted within years of formation, forcing a hypothetical merger to happen so close to a zero-age main-sequence that it could be considered part of the star formation process. Conclusions. The most likely scenario to form such young, rapidly rotating, and short-period binaries is primordial rotation, implying that the observed binaries are pre-interaction ones. Our simulations further indicate that such systems will subsequently go through a common envelope and likely merge. These binaries show that the initial spin distribution of massive stars can have a wide range of rotational velocities.

The methodology of using ontology when designing a system for evaluating the economic efficiency of geological exploration

The use of the ontological approach in the design of various information systems quite often aims at finding hidden dependencies that are not obvious at the stage of a system design. Another important aspect of the application of ontology is the need to take into account a large number of evaluation factors. Often there is a situation when the use of a standard binary architecture of connections between system elements, by analogy with databases, does not give the necessary result. This is usually due to the limitations of such relationships (one to one, one to many) when designing an information system. Accordingly, it is the ontological approach that allows us to resolve this contradiction and take into account those relationships and factors that could not be taken into account in the binary data structure. The purpose of this work is to form methodological recommendations on the use of ontology in the process of designing a system for evaluating the economic efficiency of geological exploration. As a result, we propose the structure of the ontological model, in relation to the block of parameters for assessing the economic efficiency of exploration, as well as a variant of applying this structure in practice. In the future, it is planned to develop an information system for evaluating the effectiveness of geological exploration based on the use of this ontological model.

Construction of a binary architecture of flower-like nickel phyllosilicate@zinc sulfide towards the robust, wear-resistant and thermal-stable epoxy nanocomposites

In the present study, a binary architecture of flower-like nickel phyllosilicate decorated using zinc sulfide nanoparticles (NiPS@ZnS) was constructed and introduced into epoxy resin (EP), yielding the robust, wear-resistant and thermal-stable nanocomposites. Several characterizations have been performed to confirm the synthesis of target NiPS@ZnS and explore its positive influence on the mechanical, tribological and thermal properties of EP nanocomposites. Based on the results, NiPS@ZnS could be dispersed homogeneously in the matrix without visible occurrences of debonding between interfaces, moderately improving the tensile strength and elastic modulus by 11.4% and 24.6%, respectively. The optimized tribological properties were achieved by adding 5% NiPS@ZnS and the wear rate could be reduced significantly down to 0.5×10−5 mm3/Nm, which was ca. 93% lower than pure EP. Thermal degradation analysis implied that presented NiPS@ZnS improved the thermal stability, showing pronounced suppressed mass loss and elevated char residues, while kinetic analysis demonstrated the rarely changed reaction order and steadily decreased activation energy of EP along with the increased NiPS@ZnS. This investigation provided a facile way to prepare the robust, wear-resistant and thermal-stable EP-based tribo-materials.

Self-assembly of Ag photosensitized SrTiO3 3D binary architectures for highly efficient visible light-driven dyeing wastewater splitting

Semiconductor/metal hybrid nanostructures-based finely tuning the charge transfer as a promising approach has been identified for photocatalytic activity recently towards renewable hydrogen (H 2 ) evolution reaction (HER) and environmental remediation. Engineering surface-active facets of the metal cocatalyst on the surface of 3-dimensional (3D) structured photocatalysts not only efficiently promotes charge separation and transfer but also serves as reaction active sites for the further development of advanced photocatalysts. Herein, taking 3D SrTiO 3 (STO) as a model, we introduce a plasmonic photocatalyst involving the adjusted loading of well-defined Ag active components as a cocatalyst deposited on the surface of 3D STO nanocubes to form the binary Ag/STO nanostructures for photocatalytic behavior driven by light at wavelengths of 365 nm and 420 nm. After the incorporation of the functional cocatalyst, the photocurrent response of STO was substantially increased, while its photoluminescence emission significantly declined, which could be due to the Schottky channel and enrichment of surface reactants: (i) the enhancement of the interfacial separation and inhibition of the recombination process of the photogenerated electron-hole pairs; and (ii) the Ag arrangement on the 3D structure could not only promote the enhanced light absorption but also provide a large surface area with highly active sites for the photocatalytic reaction. The as-synthesized hybrid photocatalyst with a loading content of 1 wt% Ag displays a high photocatalytic H 2 evolution and RhB degradation reaching 30 μmol g −1 h −1 (H 2 gas) and a rate constant (k) of 0.027 min −1 (RhB dye), which was 2.5, and 3.4 times higher than that of the pristine STO specimen (12.5 μmol g −1 h −1 and k = 0.008 min −1 , respectively, at 385 nm). Under a wavelength of 420 nm, the binary Ag/STO structures exhibited excellent performance for the H 2 production (17.5 μmol g −1 h −1 ) and RhB degradation (k = 0.0134 min −1 ) in comparison with the pristine STO specimen. They also demonstrated remarkable stability for H 2 evolution and dye degradation performance in continuous cycles. Based on the experimental outcomes, an energy band outline and a plausible mechanism have been proposed to describe the charge transfer and separation process for enhanced HER activity and RhB degradation. • 3D SrTiO 3 with well-defined SrTiO 3 nanocubes was fabricated via facile hydrothermal method. • AgNPs decorated 3D SrTiO 3 nanocubes were easily prepared by a one-step process. • The as-synthesized hybrid photocatalyst exhibits an outstanding photocatalytic dyeing wastewater splitting. • The properties of photocatalyst before and after the photocatalytic performance were further assessed.

Sequential stacking of a thin BHJ layer acting as a morphology regulator for efficiency enhancement in non-fullerene ternary solar cells

Although ternary-blend solar cells (TSCs) featuring alloyed systems with two acceptors can display high power conversion efficiencies (PCEs) while minimizing energy loss, it can be difficult to control the morphology of the alloyed ternary blend. In this study we developed a gradient ternary (Gradient-Ternary) device architecture by transferring a bulk-heterojunction (BHJ) layer over the spin-coated ternary blend to enhance the PCEs of alloyed TSCs. The Gradient-Ternary architecture, which was similar to that of a compositional-gradient BHJ blend, featured a vertically stratified morphology that established efficient charge extraction pathways, thereby mitigating the recombination loss faced by the corresponding Ternary blend, while concurrently improving the photocurrent, relative to that of the Ternary blend, without suffering losses in any of the other photovoltaic parameters. A champion efficiency of 16.5% was obtained using the Gradient-Ternary architecture (PM6:Y6:ITIC/PM6:ITIC) was higher than those of the Ternary (PM6:Y6:ITIC) and Binary (PM6:Y6) architectures exhibiting an efficiency of 15.7% and 15.0% respectively. Applying such gradient architectures might also accelerate the efficiency of other organic solar cells.

Self-Assembly of Ag Photosensitized Srtio3 3d Binary Architectures for Highly Efficient Visible Light-Driven Dyeing Wastewater Splitting

Self-Assembly of Ag Photosensitized Srtio3 3d Binary Architectures for Highly Efficient Visible Light-Driven Dyeing Wastewater Splitting

Approximate Constant-Coefficient Multiplication Using Hybrid Binary-Unary Computing for FPGAs

Multipliers are used in virtually all Digital Signal Processing (DSP) applications such as image and video processing. Multiplier efficiency has a direct impact on the overall performance of such applications, especially when real-time processing is needed, as in 4K video processing, or where hardware resources are limited, as in mobile and IoT devices. We propose a novel, low-cost, low energy, and high-speed approximate constant coefficient multiplier (CCM) using a hybrid binary-unary encoding method. The proposed method implements a CCM using simple routing networks with no logic gates in the unary domain, which results in more efficient multipliers compared to Xilinx LogiCORE IP CCMs and table-based KCM CCMs (Flopoco) on average. We evaluate the proposed multipliers on 2-D discrete cosine transform algorithm as a common DSP module. Post-routing FPGA results show that the proposed multipliers can improve the {area, area × delay, power consumption, and energy-delay product} of a 2-D discrete cosine transform on average by {30%, 33%, 30%, 31%}. Moreover, the throughput of the proposed 2-D discrete cosine transform is on average 5% more than that of the binary architecture implemented using table-based KCM CCMs. We will show that our method has fewer routability issues compared to binary implementations when implementing a DCT core.

SANT: Static Analysis of Native Threads for Security Vetting of Android Applications

Most of the current research on static analysis of Android applications for security vetting either work on Java source code or the Dalvik bytecode. Nevertheless, Android allows developers to use C or C++ code in their programs that is compiled into various binary architectures. Moreover, Java and the native code components (C or C++) can collaborate with each other using Java Native Interface. Recent research shows that native codes are frequently used in both benign and malicious Android applications. Most of the present Android static analysis tools avert considering native codes in their analysis and applied trivial models for their data-flow analysis. As we know only the open source JN-SAF tool has tried to solve this issue statically. However, there are still challenges like libC functions and multi-threading in native codes that we want to address in this work. We presented SANT as an extension of JN-SAF for supporting Static Analysis of Native Threads. We considered modeling libC functions in our data-flow analysis to have a more precise analysis when dealing with security vetting of native codes. We also used control flow and data dependence graphs in SANT to handle multiple concurrent threads and find implicit data-flow between them. Our experiments show that the conducted improvements outperforms JN-SAF in real-world benchmark applications.

UGEMM: Unary Computing for GEMM Applications

General matrix multiplication (GEMM) is pervasive in various domains, such as signal processing, computer vision, and machine learning. Conventional binary architectures for GEMM exhibit poor scalability in area and energy efficiency, due to the spatial nature of number representation and computing. On the contrary, unary computing processes data in temporal domain with extremely simple logic. However, to date, there rarely exist efficient architectures for unary GEMM. In this work, we first present uGEMM, a hardware-efficient unary GEMM architecture enabled by universally compatible arithmetic units, which simultaneously achieves input-insensitivity and high output accuracy. Next, we demonstrate that the proposed uGEMM can reliably early terminate the computation and offers dynamic energy-accuracy scaling for real-world applications via an accuracy-aware metric. Finally, to propel the future research for unary computing, we open source our unary computing simulator, UnarySim.

Spatially Controlled Lithium Deposition on Silver‐Nanocrystals‐Decorated TiO2 Nanotube Arrays Enabling Ultrastable Lithium Metal Anode

Abstract3D scaffolds and heterogeneous seeds are two effective ways to guide Li deposition and suppress Li dendrite growth. Herein, 3D TiO2 nanotube (TNT) arrays decorated using ultrafine silver nanocrystals (7–10 nm) through cathodic reduction deposition are first demonstrated as a confined space host for lithium metal deposition. First, TiO2 possesses intrinsic lithium affinity with large Li absorption energy, which facilitates Li capture. Then, ultrafine silver nanocrystals decoration allows the uniform and selective nucleation in nanoscale without a nucleation barrier, leading to the extraordinary formation of lithium metal importing into 3D nanotube arrays. As a result, Li metal anode deposited on such a binary architecture (TNT‐Ag‐Li) delivers a high Coulomb efficiency at around 99.4% even after 300 cycles with a capacity of 2 mA h cm−2. Remarkably, TNT‐Ag‐Li exhibits ultralow overpotential of 4 mV and long‐term cycling life over 2500 h with a capacity of 2 mAh cm–2 in Li symmetric cells. Moreover, the full battery with 3D spaced Li nanotubes anode and LiFeO4 cathode exhibits a stable and high capacity of 115 mA h g–1 at 5 C and an excellent Coulombic efficiency of ≈100% over 500 cycles.

Short-Term Long-Term Compute-in-Memory Architecture: A Hybrid Spin/CMOS Approach Supporting Intrinsic Consolidation

Biological memory structures impart enormous retention capacity while automatically providing vital functions for chronological information management and update the resolution of the domain and episodic knowledge. A crucial requirement for hardware realization of such cortical operations found in biology is to first design both short-term memory (STM) and long-term memory (LTM). Herein, these memory features are realized via a beyond-CMOS-based learning approach derived from the repeated input information and retrieval of the encoded data. We first propose a new binary STM-LTM architecture with composite synapse of the spin Hall effect-driven magnetic tunnel junction (SHE-MTJ) and capacitive memory bit cell to mimic the behavior of biological synapses. This STM-LTM platform realizes the memory potentiation through a continual update process using STM-to-LTM transfer, which is applied to neural networks based on the established capacitive crossbar. We then propose a hardware-enabled and customized STM-LTM transition algorithm for the platform considering the real hardware parameters. We validate the functionality of the design using SPICE simulations that show the proposed synapse has the potential of reaching ~30.2-pJ energy consumption for STM-to-LTM transfer and 65 pJ during STM programming. We further analyze the correlation between energy, array size, and STM-to-LTM threshold utilizing the MNIST data set.

Controllable synthesis of gossamer-like Nb2O5-RGO nanocomposite and its application to supercapacitor

The gossamer-like Nb2O5-RGO nanocomposite was synthesized through a solvothermal treatment, and the Nb2O5 nanoparticles were uniformly coated on the RGO sheets. The size of the Nb2O5 nanoparticles is about 200–700 nm. The applications of Nb2O5-RGO as supercapacitor were then explored systematically. The morphology, textural structure, and physicochemical property of Nb2O5-RGO were investigated by using SEM, TEM, XRD, FT-IR, TG, and Raman. The electrochemical performance of RGO, Nb2O5, and Nb2O5-RGO was assessed through CV and galvanostatic charge-discharge test. The experimental results show that gossamer-like Nb2O5-RGO exhibits partial faradic pseudocapacitance and good electrical double layer capacitance property. The specific capacitance value 299.23 Fg−1 of Nb2O5-RGO nanocomposite is about 2.7 and 1.7 times higher than that of Nb2O5 and RGO, retaining 87.72% of its initial capacitance after 2000 cycles under the optimized conditions, which demonstrate that more ordered binary architecture Nb2O5-RGO nanocomposite along with easy synthesis possesses faster charge-discharge kinetics and excellent supercapacitive performance.

Ultralight Programmable Bioinspired Aerogels with an Integrated Multifunctional Surface for Self-Cleaning, Oil Absorption, and Thermal Insulation via Coassembly.

Creating a configurable and controllable surface for structure-integrated multifunctionality of ultralight aerogels is of significance but remains a huge challenge because of the critical limitations of mechanical vulnerability and structural processability. Herein, inspired by Salvinia minima, the facile and one-step coassembly approach is developed to allow the structured aerogels to spontaneously replicate Salvinia-like textures for function-adaptable surfaces morphologically. The in situ superimposed construction of bioinspired topography and intrinsic topology is for the first time performed for programmable binary architectures with multifunctionality without engendering structural vulnerability and functional disruption. By introducing the binding groups for hydrophobicity tailoring, functionalized nanocellulose (f-NC) is prepared via mechanochemistry as a structural, functional, and topographical modifier for a multitasking role. The self-generated bioinspired surface with f-NC greatly maintains the structural unity and mechanical robustness, which enable self-adaptability and self-supporting of surface configurations. With fine-tuning of nucleation-driving, the binary microstructures can be controllably diversified for structure-adaptable multifunctionalities. The resulting ultralight S. minima-inspired aerogels (e.g., 0.054 g cm-3) presented outstanding temperature-endured elasticity (e.g., 90.7% high-temperature compress-recovery after multiple cycles), durable superhydrophobicity, anti-icing properties, oil absorbency efficiency (e.g., 60.2 g g-1), and thermal insulating (e.g., 0.075 W mK-1), which are superior to these reported on the overall performance. This coassembly strategy offers the opportunities for the design of ultralight materials with topography- and function-tailorable features to meet the increasing demands in many fields such as smart surfaces and self-cleaning coatings.

Au130-x Agx Nanoclusters with Non-Metallicity: A Drum of Silver-Rich Sites Enclosed in a Marks-Decahedral Cage of Gold-Rich Sites.

The synthesis and structure of atomically precise Au130-x Agx (average x=98) alloy nanoclusters protected by 55 ligands of 4-tert-butylbenzenethiolate are reported. This large alloy structure has a decahedral M54 (M=Au/Ag) core. The Au atoms are localized in the truncated Marks decahedron. In the core, a drum of Ag-rich sites is found, which is enclosed by a Marks decahedral cage of Au-rich sites. The surface is exclusively Ag-SR; X-ray absorption fine structure analysis supports the absence of Au-S bonds. The optical absorption spectrum shows a strong peak at 523 nm, seemingly a plasmon peak, but fs spectroscopic analysis indicates its non-plasmon nature. The non-metallicity of the Au130-x Agx nanocluster has set up a benchmark to study the transition to metallic state in the size evolution of bimetallic nanoclusters. The localized Au/Ag binary architecture in such a large alloy nanocluster provides atomic-level insights into the Au-Ag bonds in bimetallic nanoclusters.

Au130−xAgx Nanoclusters with Non‐Metallicity: A Drum of Silver‐Rich Sites Enclosed in a Marks‐Decahedral Cage of Gold‐Rich Sites

AbstractThe synthesis and structure of atomically precise Au130−xAgx (average x=98) alloy nanoclusters protected by 55 ligands of 4‐tert‐butylbenzenethiolate are reported. This large alloy structure has a decahedral M54 (M=Au/Ag) core. The Au atoms are localized in the truncated Marks decahedron. In the core, a drum of Ag‐rich sites is found, which is enclosed by a Marks decahedral cage of Au‐rich sites. The surface is exclusively Ag−SR; X‐ray absorption fine structure analysis supports the absence of Au−S bonds. The optical absorption spectrum shows a strong peak at 523 nm, seemingly a plasmon peak, but fs spectroscopic analysis indicates its non‐plasmon nature. The non‐metallicity of the Au130−xAgx nanocluster has set up a benchmark to study the transition to metallic state in the size evolution of bimetallic nanoclusters. The localized Au/Ag binary architecture in such a large alloy nanocluster provides atomic‐level insights into the Au−Ag bonds in bimetallic nanoclusters.

Two-Sided Topological Architecture on a Monolithic Flexible Substrate for Ultrasensitive Strain Sensors.

Flexible ultrasensitive strain sensors are highly desirable in view of their widespread applications in wearable electronics, health monitoring systems, and smart robots, where subtle strain detection is required. However, traditional fabrication of such sensors was done to prepare sensitive layers on bare or single-sided structural substrates, leading to limited sensitivity. Herein, a stretchable resistive-type strain sensor was demonstrated by self-assembling conductive networks onto a monolithic polydimethylsiloxane substrate with a two-sided topological design, for example, a sinusoid/auxetic binary architecture. The sensitivity of the obtained sensor was greatly improved by 22-fold as compared to the traditional counterpart with a bare substrate. The remarkably good agreement between the experimental results and finite element analysis predictions confirmed that the superior sensitivity is a synergistic effect of local strain enhancement derived from the topological structure on the foreside and an additional strain concentration and a reduced Poisson's ratio from the auxetic arrays on the backside. Furthermore, this sensor can withstand an extreme mechanical force (>750 N) because of the shear stiffening characteristic of the auxetic structure. Benefiting from the characteristics of ultrahigh sensitivity (gauge factor ∼1744 at 5%), low detection limit (<0.05%), and long-term durability (>500 loading cycles), this as-prepared sensor shows promise in practical applications of high-performance wearable electronics.

Constructing a “pea-pod-like” alumina-graphene binary architecture for enhancing thermal conductivity of epoxy composite

Constructing a three-dimensional thermal transport framework is an effective strategy for enhancing thermal conductivity of polymer composites. Vacuum-assisted filtration has been utilized as an effective method to prepare heat spreader films, which always exhibit poor heat dissipation performance in the axial direction. Herein, with the incorporation of spherical alumina, a pea-pod-like binary alumina-graphene architecture has been fabricated via vacuum filtration and impregnated with epoxy to obtain alumina-graphene/epoxy composite. The thermal conductivity of epoxy composite in axial direction has risen to 13.3 W m−1 K−1 with the assistance of pea-pod-like binary alumina-graphene foam with 12.1 wt% graphene and 42.4 wt% alumina loading. Thermal conductivity of the composite in radial direction reached 33.4 W m−1 K−1, which exhibits about 166 times enhancement than that of pure epoxy. The test result of the thermal conductivity of the composite under various conditions and heat transport applications confirmed the excellent thermal transport performance of the composite. Our work provides a new idea to significantly enhance the thermal conductivity of thermal management materials.