arXiv subjects

Sergi Abadal

Publications and source records attributed to Sergi Abadal.

At least 91 records · Page 5Linked to original sources

arXiv2022

ProGNNosis: A Data-driven Model to Predict GNN Computation Time Using Graph Metrics

Graph Neural Networks (GNN) show great promise in problems dealing with graph-structured data. One of the unique points of GNNs is their flexibility to adapt to multiple problems, which not only leads to wide applicability, but also poses important challenges when finding the best model or acceleration technique for a particular problem. An example of such challenges resides in the fact that the accuracy or effectiveness of a GNN model or acceleration technique generally depends on the structure of the underlying graph. In this paper, in an attempt to address the problem of graph-dependent acceleration, we propose ProGNNosis, a data-driven model that can predict the GNN training time of a given GNN model running over a graph of arbitrary characteristics by inspecting the input graph metrics. Such prediction is made based on a regression that was previously trained offline using a diverse synthetic graph dataset. In practice, our method allows making informed decisions on which design to use for a specific problem. In the paper, the methodology to build ProGNNosis is defined and applied for a specific use case, where it helps to decide which graph representation is better. Our results show that ProGNNosis helps achieve an average speedup of 1.22X over randomly selecting a graph representation in multiple widely used GNN models such as GCN, GIN, GAT, or GraphSAGE.

cs.LG↗

arXiv2022

Scale up your In-Memory Accelerator: Leveraging Wireless-on-Chip Communication for AIMC-based CNN Inference

Analog In-Memory Computing (AIMC) is emerging as a disruptive paradigm for heterogeneous computing, potentially delivering orders of magnitude better peak performance and efficiency over traditional digital signal processing architectures on Matrix-Vector multiplication. However, to sustain this throughput in real-world applications, AIMC tiles must be supplied with data at very high bandwidth and low latency; this poses an unprecedented pressure on the on-chip communication infrastructure, which becomes the system's performance and efficiency bottleneck. In this context, the performance and plasticity of emerging on-chip wireless communication paradigms provide the required breakthrough to up-scale on-chip communication in large AIMC devices. This work presents a many-tile AIMC architecture with inter-tile wireless communication that integrates multiple heterogeneous computing clusters, embedding a mix of parallel RISC-V cores and AIMC tiles. We perform an extensive design space exploration of the proposed architecture and discuss the benefits of exploiting emerging on-chip communication technologies such as wireless transceivers in the millimeter-wave and terahertz bands.

cs.AR↗

arXiv2022

Bias Reduction via Cooperative Bargaining in Synthetic Graph Dataset Generation

In general, to draw robust conclusions from a dataset, all the analyzed population must be represented on said dataset. Having a dataset that does not fulfill this condition normally leads to selection bias. Additionally, graphs have been used to model a wide variety of problems. Although synthetic graphs can be used to augment available real graph datasets to overcome selection bias, the generation of unbiased synthetic datasets is complex with current tools. In this work, we propose a method to find a synthetic graph dataset that has an even representation of graphs with different metrics. The resulting dataset can then be used, among others, for benchmarking graph processing techniques as the accuracy of different Graph Neural Network (GNN) models or the speedups obtained by different graph processing acceleration frameworks.

cs.LG↗

arXiv2022

Wireless On-Chip Communications for Scalable In-memory Hyperdimensional Computing

Hyperdimensional computing (HDC) is an emerging computing paradigm that represents, manipulates, and communicates data using very long random vectors (aka hypervectors). Among different hardware platforms capable of executing HDC algorithms, in-memory computing (IMC) systems have been recently proved to be one of the most energy-efficient options, due to hypervector manipulations in the memory itself that reduces data movement. Although implementations of HDC on single IMC cores have been made, their parallelization is still unresolved due to the communication challenges that these novel architectures impose and that traditional Networks-on-Chip and Networks-in-Package were not designed for. To cope with this difficulty, we propose the use of wireless on-chip communication technology in unique ways. We are particularly interested in physically distributing a large number of IMC cores performing similarity search across a chip, and maintaining the classification accuracy when each of which is queried with a slightly different version of a bundled hypervector. To achieve it, we introduce a novel over-the-air computing that consists of defining different binary decision regions in the receivers so as to compute the logical majority operation (i.e., bundling, or superposition) required in HDC. It introduces moderate overheads of a single antenna and receiver per IMC core. By doing so, we achieve a joint broadcast distribution and computation with a performance and efficiency unattainable with wired interconnects, which in turn enables massive parallelization of the architecture. It is demonstrated that the proposed approach allows to both bundle at least three hypervectors and scale similarity search to 64 IMC cores seamlessly, while incurring an average bit error ratio of 0.01 without any impact in the accuracy of a generic HDC-based classifier working with 512-bit vectors.

cs.AR↗

arXiv2022

Multi-Wideband Terahertz Communications via Tunable Graphene-based Metasurfaces in 6G Networks

The next generation of wireless networks is expected to tap into the terahertz (0.1--10 THz) band to satisfy the extreme latency and bandwidth density requirements of future applications. However, the development of systems in this band is challenging as THz waves confront severe spreading and penetration losses, as well as molecular absorption, which leads to strong line-of-sight requirements through highly directive antennas. Recently, reconfigurable intelligent surfaces (RISs) have been proposed to address issues derived from non-line-of-sight propagation, among other impairments, by redirecting the incident wave toward the receiver and implementing virtual-line-of-sight communications. However, the benefits provided by a RIS may be lost if the network operates at multiple bands. In this position paper, the suitability of the RIS paradigm in indoor THz scenarios for 6G is assessed grounded on the analysis of a tunable graphene-based RIS that can operate in multiple wideband transparency windows. A possible implementation of such a RIS is provided and numerically evaluated at 0.65/0.85/1.05 THz separately, demonstrating that beam steering and other relevant functionalities are realizable with excellent performance. Finally, the challenges associated with the design and fabrication of multi-wideband graphene-based RISs are discussed, paving the way to the concurrent control of multiple THz bands in the context of 6G networks.

physics.app-ph↗

arXiv2022

Metasurface-Programmable Wireless Network-on-Chip

We introduce the concept of smart radio environments, currently intensely studied for wireless communication in metasurface-programmable meter-scaled environments (e.g., inside rooms), on the chip scale. Wireless networks-on-chips (WNoCs) are a candidate technology to improve inter-core communication on chips but current proposals are plagued by a dilemma: either the received signal is weak, or it is significantly reverberated such that the on-off-keying modulation speed must be throttled. Here, we overcome this vexing problem by endowing the wireless on-chip environment with in situ programmability which allows us to shape the channel impulse response (CIR); thereby, we can impose a pulse-like CIR shape despite strong multipath propagation and without entailing a reduced received signal strength. First, we design and characterize a programmable metasurface suitable for integration in the on-chip environment ("on-chip reconfigurable intelligent surface"). Second, we optimize its configuration to equalize selected wireless on-chip channels "over the air". Third, by conducting a rigorous communication analysis, we evidence the feasibility of significantly higher modulation speeds with shaped CIRs. Our results introduce a programmability paradigm to WNoCs which boosts their competitiveness as complementary on-chip interconnect solution.

physics.app-ph↗

arXiv2022

Understanding the Design-Space of Sparse/Dense Multiphase GNN dataflows on Spatial Accelerators

Graph Neural Networks (GNNs) have garnered a lot of recent interest because of their success in learning representations from graph-structured data across several critical applications in cloud and HPC. Owing to their unique compute and memory characteristics that come from an interplay between dense and sparse phases of computations, the emergence of reconfigurable dataflow (aka spatial) accelerators offers promise for acceleration by mapping optimized dataflows (i.e., computation order and parallelism) for both phases. The goal of this work is to characterize and understand the design-space of dataflow choices for running GNNs on spatial accelerators in order for mappers or design-space exploration tools to optimize the dataflow based on the workload. Specifically, we propose a taxonomy to describe all possible choices for mapping the dense and sparse phases of GNN inference, spatially and temporally over a spatial accelerator, capturing both the intra-phase dataflow and the inter-phase (pipelined) dataflow. Using this taxonomy, we do deep-dives into the cost and benefits of several dataflows and perform case studies on implications of hardware parameters for dataflows and value of flexibility to support pipelined execution.

cs.DC↗

arXiv2021

Toward Autonomous Reconfigurable Intelligent Surfaces Through Wireless Energy Harvesting

In this work, we examine the potential of autonomous operation of a reconfigurable intelligent surface (RIS) using wireless energy harvesting from information signals. To this end, we first identify the main RIS power-consuming components and introduce a suitable power-consumption model. Subsequently, we introduce a novel RIS power-splitting architecture that enables simultaneous energy harvesting and beamsteering. Specifically, a subset of the RIS unit cells (UCs) is used for beamsteering while the remaining ones absorb energy. For the subset allocation, we propose policies obtained as solutions to two optimization problems. The first problem aims at maximizing the signal-to-noise ratio (SNR) at the receiver without violating the RIS's energy harvesting demands. Additionally, the objective of the second problem is to maximize the RIS harvested power, while ensuring an acceptable SNR at the receiver. We prove that under particular propagation conditions, some of the proposed policies deliver the optimal solution of the two problems. Furthermore, we report numerical results that reveal the efficiency of the policies with respect to the optimal and very high-complexity brute-force design approach. Finally, through a case study of user tracking, we showcase that the RIS power-consumption demands can be secured by harvesting energy from information signals.

cs.IT↗

arXiv2021

Computing Graph Neural Networks: A Survey from Algorithms to Accelerators

Graph Neural Networks (GNNs) have exploded onto the machine learning scene in recent years owing to their capability to model and learn from graph-structured data. Such an ability has strong implications in a wide variety of fields whose data is inherently relational, for which conventional neural networks do not perform well. Indeed, as recent reviews can attest, research in the area of GNNs has grown rapidly and has lead to the development of a variety of GNN algorithm variants as well as to the exploration of groundbreaking applications in chemistry, neurology, electronics, or communication networks, among others. At the current stage of research, however, the efficient processing of GNNs is still an open challenge for several reasons. Besides of their novelty, GNNs are hard to compute due to their dependence on the input graph, their combination of dense and very sparse operations, or the need to scale to huge graphs in some applications. In this context, this paper aims to make two main contributions. On the one hand, a review of the field of GNNs is presented from the perspective of computing. This includes a brief tutorial on the GNN fundamentals, an overview of the evolution of the field in the last decade, and a summary of operations carried out in the multiple phases of different GNN algorithm variants. On the other hand, an in-depth analysis of current software and hardware acceleration schemes is provided, from which a hardware-software, graph-aware, and communication-centric vision for GNN accelerators is distilled.

cs.LG↗

arXiv2021

Localization in Power-Constrained Terahertz-Operating Software-Defined Metamaterials

Software-Defined Metamaterials (SDMs) show a strong potential for advancing the engineered control of electromagnetic waves. As such, they are envisioned to enable a variety of exciting applications, among others in the domains of smart textiles, high-resolution structural monitoring, and sensing in challenging environments. Many of the applications envisage deformations of the SDM structures, such as their bending, stretching or rolling, which implies that the locations of metamaterial elements will be changing relative to one another. In this paper, we argue that if the metamaterial elements would be accurately localizable, this location information could potentially be utilized for enabling novel SDM applications, as well as for optimizing the control of the elements themselves. To enable their localization, we assume that these elements are controlled wirelessly through a Terahertz (THz)-operating nanonetwork. We consider the elements to be power-constrained, with their sole powering option being to harvest energy from different environmental sources. By means of simulation, we demonstrate sub-millimeter accuracy of the two-way Time of Flight (ToF)-based localization, as well as high availability of the service (i.e., consistently more than 80% of the time), which is a result of the low energy consumed in the localization process. Finally, we qualitatively characterize the latency of the proposed localization service, as well as outline several challenges and future research directions.

eess.SP↗

arXiv2021

Toward Location-aware In-body Terahertz Nanonetworks with Energy Harvesting

Nanoscale wireless networks are expected to revolutionize a variety of domains, with significant advances conceivable in in-body healthcare. In healthcare, these nanonetworks will consist of energy-harvesting nanodevices passively flowing through the bloodstream, taking actions at certain locations, and communicating results to more powerful Body Area Network (BAN) nodes. Assuming such a setup and electromagnetic nanocommunication in the Terahertz (THz) frequencies, we propose a network architecture that can support fine-grained localization of the energy-harvesting in-body nanonodes, as well as their two-way communication with the outside world. The main novelties of our proposal lie in the introduction of location-aware and Wake-up Radio (WuR)-based wireless nanocommunication paradigms, as well as Software-Defined Metamaterials (SDMs), to THz-operating energy-harvesting in-body nanonetworks. We argue that, on a high level, the proposed architecture can handle (and actually benefits from) a large number of nanonodes, while simultaneously dealing with a short range of THz in-body propagation and highly constrained nanonodes.

cs.ET↗

arXiv2021

Autonomous Reconfigurable Intelligent Surfaces Through Wireless Energy Harvesting

In this paper, we examine the potential for a reconfigurable intelligent surface (RIS) to be powered by energy harvested from information signals. This feature might be key to reap the benefits of RIS technology's lower power consumption compared to active relays. We first identify the main RIS power-consuming components and then propose an energy harvesting and power consumption model. Furthermore, we formulate and solve the problem of the optimal RIS placement together with the amplitude and phase response adjustment of its elements in order to maximize the signal-to-noise ratio (SNR) while harvesting sufficient energy for its operation. Finally, numerical results validate the autonomous operation potential and reveal the range of power consumption values that enables it.

cs.IT↗

arXiv2021

Material-Dependencies of the THz Emission from Plasmonic Graphene-Based Photoconductive Antenna Structures

Graphene supports surface plasmon polaritons with comparatively slow propagation velocities in the THz region, which becomes increasingly interesting for future communication technologies. This ability can be used to realize compact antennas, which are up to two orders of magnitude smaller than their metallic counterparts. For a proper functionality of these antennas some minimum material requirements have to be fulfilled, which are presently difficult to achieve, since the fabrication and transfer technologies for graphene are still evolving. In this work we analyze available graphene materials experimentally and extract intrinsic characteristics at THz frequencies, in order to predict the dependency of the THz signal emission threshold as a function of the graphene relaxation time tau_r and the chemical potential mu_c.

cond-mat.mes-hall↗

arXiv2021

Characterizing the Communication Requirements of GNN Accelerators: A Model-Based Approach

Relational data present in real world graph representations demands for tools capable to study it accurately. In this regard Graph Neural Network (GNN) is a powerful tool, wherein various models for it have also been developed over the past decade. Recently, there has been a significant push towards creating accelerators that speed up the inference and training process of GNNs. These accelerators, however, do not delve into the impact of their dataflows on the overall data movement and, hence, on the communication requirements. In this paper, we formulate analytical models that capture the amount of data movement in the most recent GNN accelerator frameworks. Specifically, the proposed models capture the dataflows and hardware setup of these accelerator designs and expose their scalability characteristics for a set of hardware, GNN model and input graph parameters. Additionally, the proposed approach provides means for the comparative analysis of the vastly different GNN accelerators.

cs.AR↗

arXiv2021

Survey on Terahertz Nanocommunication and Networking: A Top-Down Perspective

Recent developments in nanotechnology herald nanometer-sized devices expected to bring light to a number of groundbreaking applications. Communication with and among nanodevices will be needed for unlocking the full potential of such applications. As the traditional communication approaches cannot be directly applied in nanocommunication, several alternative paradigms have emerged. Among them, electromagnetic nanocommunication in the terahertz (THz) frequency band is particularly promising, mainly due to the breakthrough of novel materials such as graphene. For this reason, numerous research efforts are nowadays targeting THz band nanocommunication and consequently nanonetworking. As it is expected that these trends will continue in the future, we see it beneficial to summarize the current status in these research domains. In this survey, we therefore aim to provide an overview of the current THz nanocommunication and nanonetworking research. Specifically, we discuss the applications envisioned to be supported by nanonetworks operating in the THz band, together with the requirements such applications pose on the underlying nanonetworks. Subsequently, we provide an overview of the current contributions on the different layers of the protocol stack, as well as the available channel models and experimentation tools. As the final contribution, we identify a number of open research challenges and outline several potential future research directions.

cs.NI↗

arXiv2020

Digital Metasurface Based on Graphene: An Application to Beam Steering in Terahertz Plasmonic Antennas

Metasurfaces, the two-dimensional counterpart of metamaterials, have caught great attention thanks to their powerful capabilities on manipulation of electromagnetic waves. Recent times have seen the emergence of a variety of metasurfaces exhibiting not only countless functionalities, but also a reconfigurable response. Additionally, digital or coding metasurfaces have revolutionized the field by describing the device as a matrix of discrete building block states, thus drawing clear parallelisms with information theory and opening new ways to model, compose, and (re)program advanced metasurfaces. This paper joins the reconfigurable and digital approaches, and presents a metasurface that leverages the tunability of graphene to perform beam steering at terahertz frequencies. A comprehensive design methodology is presented encompassing technological, unit cell design, digital metamaterial synthesis, and programmability aspects. By setting up and dynamically adjusting a phase gradient along the metasurface plane, the resulting device achieves beam steering at all practical directions. The proposed design is studied through analytical models and validated numerically, showing beam widths and steering errors well below 10 degrees and 5% in most cases. Finally, design guidelines are extracted through a scalability analysis involving the metasurface size and number of unit cell states.

physics.app-ph↗

arXiv2020

Dataflow-Architecture Co-Design for 2.5D DNN Accelerators using Wireless Network-on-Package

Deep neural network (DNN) models continue to grow in size and complexity, demanding higher computational power to enable real-time inference. To efficiently deliver such computational demands, hardware accelerators are being developed and deployed across scales. This naturally requires an efficient scale-out mechanism for increasing compute density as required by the application. 2.5D integration over interposer has emerged as a promising solution, but as we show in this work, the limited interposer bandwidth and multiple hops in the Network-on-Package (NoP) can diminish the benefits of the approach. To cope with this challenge, we propose WIENNA, a wireless NoP-based 2.5D DNN accelerator. In WIENNA, the wireless NoP connects an array of DNN accelerator chiplets to the global buffer chiplet, providing high-bandwidth multicasting capabilities. Here, we also identify the dataflow style that most efficienty exploits the wireless NoP's high-bandwidth multicasting capability on each layer. With modest area and power overheads, WIENNA achieves 2.2X--5.1X higher throughput and 38.2% lower energy than an interposer-based NoP design.

cs.AR↗

arXiv2020

Workload Characterization of Programmable Metasurfaces

Metasurfaces are envisaged to play a key role in next-generation wireless systems due to their powerful control over electromagnetic waves. The last decade has witnessed huge advances in this regard, shifting from static to programmable metasurfaces. The HyperSurface (HSF) paradigm takes one step further by integrating a network of controllers within the device with the aim of adding intelligence, connectivity, and autonomy. However, little is known about the traffic that this network will have to support as the target electromagnetic function or boundary conditions change. In this paper, we lay down the foundations of a methodology to characterize the workload of programmable metasurfaces and then employ it to analyze the case of beam steering HSFs. We observe that traffic is bursty and highly dependent on the position of the target. These results will enable the early-stage evaluation of intra-HSF networks, as well as the estimation of the system performance and cost.

cs.NI↗