Researcher profile

Luca Benini

Luca Benini contributes to research discovery and scholarly infrastructure.

ResearcherAffiliation not importedOpen to collaborate

Trust snapshot

Quick read

Trust 21 - EmergingVerification L1Unclaimed author

74works

0followers

22topics

4close collaborators

Actions

Decide how to stay connected

Follow researcher0

Claiming links this public author record to a researcher profile and unlocks direct collaboration workflows.

Direct collaboration

Open a focused conversation when the fit is right

Claim this author entity first to unlock direct invitations.

Inspect adjacent work, topics, institutions and collaborators without jumping out to a separate graph page.

Building this graph slice

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

MR2-ByteTrack: CNN and Transformer-based Video Object Detection for AI-augmented Embedded Vision Sensor Nodes

Modern smart vision sensors need on-device intelligence to process video streams, as cloud computing is often impractical due to bandwidth, latency, and privacy constraints. However, these sensory systems typically rely on ultra-low-power microcontrollers (MCUs) with limited memory and compute, making conventional video object detection methods, which require feature storage or multi-frame buffering, unfeasible. To address this challenge, we introduce Multi-Resolution Rescored ByteTrack (MR2-ByteTrack), a Video Object Detection (VOD) method tailored for MCU-based embedded vision nodes. MR2-ByteTrack reduces computational cost by alternating between full- and low-resolution inference, while linking detections across frames via ByteTrack and correcting misclassifications through the Rescore algorithm, which applies probability union rules to aggregate detection confidence scores across frames. We apply our approach to both a CNN-based detector and a Transformer-based model, demonstrating its generality across architectures with fundamentally different spatial processing. Experiments on ImageNetVID demonstrate that MR2-ByteTrack maintains accuracy, achieving mAP scores of up to 49.0 for the CNN-based models and 48.7 for the Transformer, while reducing multiply-accumulate operations by as much as 53\% for the CNNs and 32\% for the Transformer. When deployed on GAP9, an ultra-low-power RISC-V multicore MCU, our method yields up to 55\% energy savings compared to processing only full-resolution images, enabling the first real-time Transformer-based VOD on an MCU-class embedded vision node. Code available at https://github.com/Bomps4/Multi_Resolution_Rescored_ByteTrack/tree/IEEE_Access

preprint2026arXiv

RDKV: Rate-Distortion Bit Allocation for Joint Eviction and Quantization of the KV Cache

Large language models (LLMs) have shown strong performance across diverse tasks, but their inference with long input contexts is bottlenecked by memory size and bandwidth. The Key-Value (KV) cache size grows linearly with sequence length and needs to be re-read from off-chip high-bandwidth memory (HBM) to on-chip memory at every decoding step, resulting in memory-bound inference. Existing methods reduce the cache by either eviction or quantization, but typically treat the two in isolation. In this paper, we cast KV cache compression as a rate-distortion problem, under which eviction and quantization are two end-points of the same bit allocation scheme. This exposes the need to optimize them jointly, motivating our method, RDKV (Rate-Distortion KV cache compression). RDKV derives the weight of each token or channel from the distortion that compression induces on the attention computation. Based on these weights, it assigns each token or channel a bit-width ranging from full precision down to zero bits guided by reverse water-filling, applied once after the prefilling stage. Experiments on LongBench, RULER, and InfiniteBench show that RDKV outperforms the best evaluated baseline by 9.1% on average. On LongBench it recovers 97.81% of full-cache accuracy with only 2.48% cache retention. Compared with full-cache FlashAttention-2 decoding, it achieves 4.5x decode speedup and 1.9x peak memory reduction with 128K context length, while maintaining comparable performance.

preprint2026arXiv

TinyMyo: a Tiny Foundation Model for Flexible EMG Signal Processing at the Edge

Objective: Surface electromyography (EMG) is a non-invasive sensing modality widely used in biomechanics, rehabilitation, prosthetic control, and human-machine interfaces. Despite decades of use, achieving robust generalization across subjects, recording systems, and acquisition protocols remains challenging. While foundation models (FMs) are gaining traction for EMG, existing approaches remain limited to single downstream tasks and lack deployability on embedded platforms. This work addresses these limitations. Methods: We present TinyMyo, a lightweight FM based on a Transformer encoder architecture. The model is pre-trained in a self-supervised manner using masked reconstruction on publicly available datasets. With only 3.6M parameters, TinyMyo is designed to support multiple downstream tasks through minimal task-specific head adaptations. Results: We demonstrate generalization across hand gesture classification, hand kinematic regression, speech production and speech recognition, with performance comparable to or surpassing the state of the art (SoA), and model size below 5M parameters. We achieve SoA results compared to previous FM-based works on the NinaPro DB5 (89.4%), UCI-EMG (97.56%), and EPN-612 (96.74%) datasets. We demonstrate the first-time deployment of an EMG FM on an ultra-low power microcontroller (GAP9), with an inference time of 0.785 s, energy of 44.91 mJ and power envelope of 57.18 mW. Conclusion: TinyMyo demonstrates that compact, self-supervised EMG FM can guarantee strong generalization across multiple downstream tasks while remaining compatible with low-power edge devices. Significance: TinyMyo is the first EMG FM for ultra-low power edge devices, enabling scalable and energy-efficient sensing for motor intent decoding, neuromuscular assessment, and biosignal driven human-machine interaction.

preprint2025arXiv

TROOP: At-the-Roofline Performance for Vector Processors on Low Operational Intensity Workloads

The fast evolution of Machine Learning (ML) models requires flexible and efficient hardware solutions as hardwired accelerators face rapid obsolescence. Vector processors are fully programmable and achieve high energy efficiencies by exploiting data parallelism, amortizing instruction fetch and decoding costs. Hence, a promising design choice is to build accelerators based on shared L1-memory clusters of streamlined Vector Processing Elements (VPEs). However, current state-of-the-art VPEs are limited in L1 memory bandwidth and achieve high efficiency only for computational kernels with high data reuse in the Vector Register File (VRF), such as General Matrix Multiplication (GEMM). Performance is suboptimal for workloads with lower data reuse like General Matrix-Vector Multiplication (GEMV). To fully exploit available bandwidth at the L1 memory interface, the VPE micro-architecture must be optimized to achieve near-ideal utilization, i.e., to be as close as possible to the L1 memory roofline (at-the-roofline). In this work, we propose TROOP, a set of hardware optimizations that include decoupled load-store interfaces, improved vector chaining, shadow buffers to hide VRF conflicts, and address scrambling techniques to achieve at-the-roofline performance for VPEs without compromising their area and energy efficiency. We implement TROOP on an open-source streamlined vector processor in a 12nm FinFET technology. TROOP achieves significant speedups of 1.5x, 2.2x, and 2.6x, respectively, for key memory-intensive kernels such as GEMV, DOTP and AXPY, achieving at-the-roofline performance. Additionally, TROOP enhances the energy efficiency by up to 45%, reaching 38 DP-GFLOPs/W (1 GHz, TT, 0.8V) for DOTP while maintaining a high energy efficiency of 61 DP-GFLOPs/W for GEMMs, incurring only a minor area overhead of less than 7%.

preprint2024arXiv

Data-Driven Power Modeling and Monitoring via Hardware Performance Counters Tracking

In the current high-performance and embedded computing era, full-stack energy-centric design is paramount. Use cases require increasingly high performance at an affordable power budget, often under real-time constraints. Extreme heterogeneity and parallelism address these issues but greatly complicate online power consumption assessment, which is essential for dynamic hardware and software stack adaptations. We introduce a novel architecture-agnostic power modeling methodology with state-of-the-art accuracy, low overhead, and high responsiveness. Our methodology identifies the best Performance Monitoring Counters (PMCs) to model the power consumption of each hardware sub-system at each Dynamic Voltage and Frequency Scaling (DVFS) state. The individual linear models are combined into a complete model that effectively describes the power consumption of the whole system, achieving high accuracy and low overhead. Our evaluation reports an average estimation error of 7.5 % for power consumption and 1.3 % for energy. Furthermore, we propose Runmeter, an open-source, PMC-based monitoring framework integrated into the Linux kernel. Runmeter manages PMC samples collection and manipulation, efficiently evaluating our power models at runtime. With a time overhead of only 0.7 % in the worst case, Runmeter provides responsive and accurate power measurements directly in the kernel, which can be employed for actuation policies such as Dynamic Power Management (DPM) and power-aware task scheduling.

preprint2024arXiv

MX: Enhancing RISC-V's Vector ISA for Ultra-Low Overhead, Energy-Efficient Matrix Multiplication

Dense Matrix Multiplication (MatMul) is arguably one of the most ubiquitous compute-intensive kernels, spanning linear algebra, DSP, graphics, and machine learning applications. Thus, MatMul optimization is crucial not only in high-performance processors but also in embedded low-power platforms. Several Instruction Set Architectures (ISAs) have recently included matrix extensions to improve MatMul performance and efficiency at the cost of added matrix register files and units. In this paper, we propose Matrix eXtension (MX), a lightweight approach that builds upon the open-source RISC-V Vector (RVV) ISA to boost MatMul energy efficiency. Instead of adding expensive dedicated hardware, MX uses the pre-existing vector register file and functional units to create a hybrid vector/matrix engine at a negligible area cost (< 3%), which comes from a compact near-FPU tile buffer for higher data reuse, and no clock frequency overhead. We implement MX on a compact and highly energy-optimized RVV processor and evaluate it in both a Dual- and 64-Core cluster in a 12-nm technology node. MX boosts the Dual-Core's energy efficiency by 10% for a double-precision 64x64x64 matrix multiplication with the same FPU utilization (~97%) and by 25% on the 64-Core cluster for the same benchmark on 32-bit data, with a 56% performance gain.

preprint2023arXiv

An Accurate EEGNet-based Motor-Imagery Brain-Computer Interface for Low-Power Edge Computing

This paper presents an accurate and robust embedded motor-imagery brain-computer interface (MI-BCI). The proposed novel model, based on EEGNet, matches the requirements of memory footprint and computational resources of low-power microcontroller units (MCUs), such as the ARM Cortex-M family. Furthermore, the paper presents a set of methods, including temporal downsampling, channel selection, and narrowing of the classification window, to further scale down the model to relax memory requirements with negligible accuracy degradation. Experimental results on the Physionet EEG Motor Movement/Imagery Dataset show that standard EEGNet achieves 82.43%, 75.07%, and 65.07% classification accuracy on 2-, 3-, and 4-class MI tasks in global validation, outperforming the state-of-the-art (SoA) convolutional neural network (CNN) by 2.05%, 5.25%, and 5.48%. Our novel method further scales down the standard EEGNet at a negligible accuracy loss of 0.31% with 7.6x memory footprint reduction and a small accuracy loss of 2.51% with 15x reduction. The scaled models are deployed on a commercial Cortex-M4F MCU taking 101ms and consuming 4.28mJ per inference for operating the smallest model, and on a Cortex-M7 with 44ms and 18.1mJ per inference for the medium-sized model, enabling a fully autonomous, wearable, and accurate low-power BCI.

preprint2023arXiv

HR-SAR-Net: A Deep Neural Network for Urban Scene Segmentation from High-Resolution SAR Data

Synthetic aperture radar (SAR) data is becoming increasingly available to a wide range of users through commercial service providers with resolutions reaching 0.5m/px. Segmenting SAR data still requires skilled personnel, limiting the potential for large-scale use. We show that it is possible to automatically and reliably perform urban scene segmentation from next-gen resolution SAR data (0.15m/px) using deep neural networks (DNNs), achieving a pixel accuracy of 95.19% and a mean IoU of 74.67% with data collected over a region of merely 2.2km${}^2$. The presented DNN is not only effective, but is very small with only 63k parameters and computationally simple enough to achieve a throughput of around 500Mpx/s using a single GPU. We further identify that additional SAR receive antennas and data from multiple flights massively improve the segmentation accuracy. We describe a procedure for generating a high-quality segmentation ground truth from multiple inaccurate building and road annotations, which has been crucial to achieving these segmentation results.

preprint2023arXiv

InfiniWolf: Energy Efficient Smart Bracelet for Edge Computing with Dual Source Energy Harvesting

This work presents InfiniWolf, a novel multi-sensor smartwatch that can achieve self-sustainability exploiting thermal and solar energy harvesting, performing computationally high demanding tasks. The smartwatch embeds both a System-on-Chip (SoC) with an ARM Cortex-M processor and Bluetooth Low Energy (BLE) and Mr. Wolf, an open-hardware RISC-V based parallel ultra-low-power processor that boosts the processing capabilities on board by more than one order of magnitude, while also increasing energy efficiency. We demonstrate its functionality based on a sample application scenario performing stress detection with multi-layer artificial neural networks on a wearable multi-sensor bracelet. Experimental results show the benefits in terms of energy efficiency and latency of Mr. Wolf over an ARM Cortex-M4F micro-controllers and the possibility, under specific assumptions, to be self-sustainable using thermal and solar energy harvesting while performing up to 24 stress classifications per minute in indoor conditions.

preprint2023arXiv

Q-EEGNet: an Energy-Efficient 8-bit Quantized Parallel EEGNet Implementation for Edge Motor-Imagery Brain--Machine Interfaces

Motor-Imagery Brain--Machine Interfaces (MI-BMIs)promise direct and accessible communication between human brains and machines by analyzing brain activities recorded with Electroencephalography (EEG). Latency, reliability, and privacy constraints make it unsuitable to offload the computation to the cloud. Practical use cases demand a wearable, battery-operated device with low average power consumption for long-term use. Recently, sophisticated algorithms, in particular deep learning models, have emerged for classifying EEG signals. While reaching outstanding accuracy, these models often exceed the limitations of edge devices due to their memory and computational requirements. In this paper, we demonstrate algorithmic and implementation optimizations for EEGNET, a compact Convolutional Neural Network (CNN) suitable for many BMI paradigms. We quantize weights and activations to 8-bit fixed-point with a negligible accuracy loss of 0.4% on 4-class MI, and present an energy-efficient hardware-aware implementation on the Mr.Wolf parallel ultra-low power (PULP) System-on-Chip (SoC) by utilizing its custom RISC-V ISA extensions and 8-core compute cluster. With our proposed optimization steps, we can obtain an overall speedup of 64x and a reduction of up to 85% in memory footprint with respect to a single-core layer-wise baseline implementation. Our implementation takes only 5.82 ms and consumes 0.627 mJ per inference. With 21.0GMAC/s/W, it is 256x more energy-efficient than an EEGNET implementation on an ARM Cortex-M7 (0.082GMAC/s/W).

preprint2022arXiv

A Heterogeneous In-Memory Computing Cluster For Flexible End-to-End Inference of Real-World Deep Neural Networks

Deployment of modern TinyML tasks on small battery-constrained IoT devices requires high computational energy efficiency. Analog In-Memory Computing (IMC) using non-volatile memory (NVM) promises major efficiency improvements in deep neural network (DNN) inference and serves as on-chip memory storage for DNN weights. However, IMC's functional flexibility limitations and their impact on performance, energy, and area efficiency are not yet fully understood at the system level. To target practical end-to-end IoT applications, IMC arrays must be enclosed in heterogeneous programmable systems, introducing new system-level challenges which we aim at addressing in this work. We present a heterogeneous tightly-coupled clustered architecture integrating 8 RISC-V cores, an in-memory computing accelerator (IMA), and digital accelerators. We benchmark the system on a highly heterogeneous workload such as the Bottleneck layer from a MobileNetV2, showing 11.5x performance and 9.5x energy efficiency improvements, compared to highly optimized parallel execution on the cores. Furthermore, we explore the requirements for end-to-end inference of a full mobile-grade DNN (MobileNetV2) in terms of IMC array resources, by scaling up our heterogeneous architecture to a multi-array accelerator. Our results show that our solution, on the end-to-end inference of the MobileNetV2, is one order of magnitude better in terms of execution latency than existing programmable architectures and two orders of magnitude better than state-of-the-art heterogeneous solutions integrating in-memory computing analog cores.

preprint2022arXiv

Adaptive Random Forests for Energy-Efficient Inference on Microcontrollers

Random Forests (RFs) are widely used Machine Learning models in low-power embedded devices, due to their hardware friendly operation and high accuracy on practically relevant tasks. The accuracy of a RF often increases with the number of internal weak learners (decision trees), but at the cost of a proportional increase in inference latency and energy consumption. Such costs can be mitigated considering that, in most applications, inputs are not all equally difficult to classify. Therefore, a large RF is often necessary only for (few) hard inputs, and wasteful for easier ones. In this work, we propose an early-stopping mechanism for RFs, which terminates the inference as soon as a high-enough classification confidence is reached, reducing the number of weak learners executed for easy inputs. The early-stopping confidence threshold can be controlled at runtime, in order to favor either energy saving or accuracy. We apply our method to three different embedded classification tasks, on a single-core RISC-V microcontroller, achieving an energy reduction from 38% to more than 90% with a drop of less than 0.5% in accuracy. We also show that our approach outperforms previous adaptive ML methods for RFs.

preprint2022arXiv

Aerosense: A Self-Sustainable And Long-Range Bluetooth Wireless Sensor Node for Aerodynamic and Aeroacoustic Monitoring on Wind Turbines

This paper presents a low-power, self-sustainable, and modular wireless sensor node for aerodynamic and acoustic measurements on wind turbines and other industrial structures. It includes 40 high-accuracy barometers, 10 microphones, 5 differential pressure sensors, and implements a lossy and a lossless on-board data compression algorithm to decrease the transmission energy cost. The wireless transmitter is based on Bluetooth Low Energy 5.1 tuned for long-range and high throughput while maintaining adequate per-bit energy efficiency (80 nJ). Moreover, we field-assessed the node capability to collect precise and accurate aerodynamic data. Outdoor experimental tests revealed that the system can acquire and sustain a data rate of 850 kbps over 438 m. The power consumption while collecting and streaming all measured data is 120 mW, enabling self-sustainability and long-term in-situ monitoring with a 111 cm^2 photovoltaic panel.

preprint2022arXiv

Bioformers: Embedding Transformers for Ultra-Low Power sEMG-based Gesture Recognition

Human-machine interaction is gaining traction in rehabilitation tasks, such as controlling prosthetic hands or robotic arms. Gesture recognition exploiting surface electromyographic (sEMG) signals is one of the most promising approaches, given that sEMG signal acquisition is non-invasive and is directly related to muscle contraction. However, the analysis of these signals still presents many challenges since similar gestures result in similar muscle contractions. Thus the resulting signal shapes are almost identical, leading to low classification accuracy. To tackle this challenge, complex neural networks are employed, which require large memory footprints, consume relatively high energy and limit the maximum battery life of devices used for classification. This work addresses this problem with the introduction of the Bioformers. This new family of ultra-small attention-based architectures approaches state-of-the-art performance while reducing the number of parameters and operations of 4.9X. Additionally, by introducing a new inter-subjects pre-training, we improve the accuracy of our best Bioformer by 3.39%, matching state-of-the-art accuracy without any additional inference cost. Deploying our best performing Bioformer on a Parallel, Ultra-Low Power (PULP) microcontroller unit (MCU), the GreenWaves GAP8, we achieve an inference latency and energy of 2.72 ms and 0.14 mJ, respectively, 8.0X lower than the previous state-of-the-art neural network, while occupying just 94.2 kB of memory.

preprint2022arXiv

Constrained Few-shot Class-incremental Learning

Continually learning new classes from fresh data without forgetting previous knowledge of old classes is a very challenging research problem. Moreover, it is imperative that such learning must respect certain memory and computational constraints such as (i) training samples are limited to only a few per class, (ii) the computational cost of learning a novel class remains constant, and (iii) the memory footprint of the model grows at most linearly with the number of classes observed. To meet the above constraints, we propose C-FSCIL, which is architecturally composed of a frozen meta-learned feature extractor, a trainable fixed-size fully connected layer, and a rewritable dynamically growing memory that stores as many vectors as the number of encountered classes. C-FSCIL provides three update modes that offer a trade-off between accuracy and compute-memory cost of learning novel classes. C-FSCIL exploits hyperdimensional embedding that allows to continually express many more classes than the fixed dimensions in the vector space, with minimal interference. The quality of class vector representations is further improved by aligning them quasi-orthogonally to each other by means of novel loss functions. Experiments on the CIFAR100, miniImageNet, and Omniglot datasets show that C-FSCIL outperforms the baselines with remarkable accuracy and compression. It also scales up to the largest problem size ever tried in this few-shot setting by learning 423 novel classes on top of 1200 base classes with less than 1.6% accuracy drop. Our code is available at https://github.com/IBM/constrained-FSCIL.

preprint2022arXiv

Embedding Temporal Convolutional Networks for Energy-Efficient PPG-Based Heart Rate Monitoring

Photoplethysmography (PPG) sensors allow for non-invasive and comfortable heart-rate (HR) monitoring, suitable for compact wrist-worn devices. Unfortunately, Motion Artifacts (MAs) severely impact the monitoring accuracy, causing high variability in the skin-to-sensor interface. Several data fusion techniques have been introduced to cope with this problem, based on combining PPG signals with inertial sensor data. Until know, both commercial and reasearch solutions are computationally efficient but not very robust, or strongly dependent on hand-tuned parameters, which leads to poor generalization performance. % In this work, we tackle these limitations by proposing a computationally lightweight yet robust deep learning-based approach for PPG-based HR estimation. Specifically, we derive a diverse set of Temporal Convolutional Networks (TCN) for HR estimation, leveraging Neural Architecture Search (NAS). Moreover, we also introduce ActPPG, an adaptive algorithm that selects among multiple HR estimators depending on the amount of MAs, to improve energy efficiency. We validate our approaches on two benchmark datasets, achieving as low as 3.84 Beats per Minute (BPM) of Mean Absolute Error (MAE) on PPGDalia, which outperforms the previous state-of-the-art. Moreover, we deploy our models on a low-power commercial microcontroller (STM32L4), obtaining a rich set of Pareto optimal solutions in the complexity vs. accuracy space.

preprint2022arXiv

Energy-Efficient Adaptive Machine Learning on IoT End-Nodes With Class-Dependent Confidence

Energy-efficient machine learning models that can run directly on edge devices are of great interest in IoT applications, as they can reduce network pressure and response latency, and improve privacy. An effective way to obtain energy-efficiency with small accuracy drops is to sequentially execute a set of increasingly complex models, early-stopping the procedure for "easy" inputs that can be confidently classified by the smallest models. As a stopping criterion, current methods employ a single threshold on the output probabilities produced by each model. In this work, we show that such a criterion is sub-optimal for datasets that include classes of different complexity, and we demonstrate a more general approach based on per-classes thresholds. With experiments on a low-power end-node, we show that our method can significantly reduce the energy consumption compared to the single-threshold approach.

preprint2022arXiv

Energy-Efficient Tree-Based EEG Artifact Detection

In the context of epilepsy monitoring, EEG artifacts are often mistaken for seizures due to their morphological similarity in both amplitude and frequency, making seizure detection systems susceptible to higher false alarm rates. In this work we present the implementation of an artifact detection algorithm based on a minimal number of EEG channels on a parallel ultra-low-power (PULP) embedded platform. The analyses are based on the TUH EEG Artifact Corpus dataset and focus on the temporal electrodes. First, we extract optimal feature models in the frequency domain using an automated machine learning framework, achieving a 93.95% accuracy, with a 0.838 F1 score for a 4 temporal EEG channel setup. The achieved accuracy levels surpass state-of-the-art by nearly 20%. Then, these algorithms are parallelized and optimized for a PULP platform, achieving a 5.21 times improvement of energy-efficient compared to state-of-the-art low-power implementations of artifact detection frameworks. Combining this model with a low-power seizure detection algorithm would allow for 300h of continuous monitoring on a 300 mAh battery in a wearable form factor and power budget. These results pave the way for implementing affordable, wearable, long-term epilepsy monitoring solutions with low false-positive rates and high sensitivity, meeting both patients' and caregivers' requirements.

preprint2022arXiv

Exploring Scalable, Distributed Real-Time Anomaly Detection for Bridge Health Monitoring

Modern real-time Structural Health Monitoring systems can generate a considerable amount of information that must be processed and evaluated for detecting early anomalies and generating prompt warnings and alarms about the civil infrastructure conditions. The current cloud-based solutions cannot scale if the raw data has to be collected from thousands of buildings. This paper presents a full-stack deployment of an efficient and scalable anomaly detection pipeline for SHM systems which does not require sending raw data to the cloud but relies on edge computation. First, we benchmark three algorithmic approaches of anomaly detection, i.e., Principal Component Analysis (PCA), Fully-Connected AutoEncoder (FC-AE), and Convolutional AutoEncoder (C-AE). Then, we deploy them on an edge-sensor, the STM32L4, with limited computing capabilities. Our approach decreases network traffic by $\approx8\cdot10^5\times$ , from 780KB/hour to less than 10 Bytes/hour for a single installation and minimize network and cloud resource utilization, enabling the scaling of the monitoring infrastructure. A real-life case study, a highway bridge in Italy, demonstrates that combining near-sensor computation of anomaly detection algorithms, smart pre-processing, and low-power wide-area network protocols (LPWAN) we can greatly reduce data communication and cloud computing costs, while anomaly detection accuracy is not adversely affected.

preprint2022arXiv

FANN-on-MCU: An Open-Source Toolkit for Energy-Efficient Neural Network Inference at the Edge of the Internet of Things

The growing number of low-power smart devices in the Internet of Things is coupled with the concept of "Edge Computing", that is moving some of the intelligence, especially machine learning, towards the edge of the network. Enabling machine learning algorithms to run on resource-constrained hardware, typically on low-power smart devices, is challenging in terms of hardware (optimized and energy-efficient integrated circuits), algorithmic and firmware implementations. This paper presents FANN-on-MCU, an open-source toolkit built upon the Fast Artificial Neural Network (FANN) library to run lightweight and energy-efficient neural networks on microcontrollers based on both the ARM Cortex-M series and the novel RISC-V-based Parallel Ultra-Low-Power (PULP) platform. The toolkit takes multi-layer perceptrons trained with FANN and generates code targeted at execution on low-power microcontrollers either with a floating-point unit (i.e., ARM Cortex-M4F and M7F) or without (i.e., ARM Cortex M0-M3 or PULP-based processors). This paper also provides an architectural performance evaluation of neural networks on the most popular ARM Cortex-M family and the parallel RISC-V processor called Mr. Wolf. The evaluation includes experimental results for three different applications using a self-sustainable wearable multi-sensor bracelet. Experimental results show a measured latency in the order of only a few microseconds and a power consumption of few milliwatts while keeping the memory requirements below the limitations of the targeted microcontrollers. In particular, the parallel implementation on the octa-core RISC-V platform reaches a speedup of 22x and a 69% reduction in energy consumption with respect to a single-core implementation on Cortex-M4 for continuous real-time classification.

preprint2022arXiv

GVSoC: A Highly Configurable, Fast and Accurate Full-Platform Simulator for RISC-V based IoT Processors

The last few years have seen the emergence of IoT processors: ultra-low power systems-on-chips (SoCs) combining lightweight and flexible micro-controller units (MCUs), often based on open-ISA RISC-V cores, with application-specific accelerators to maximize performance and energy efficiency. Overall, this heterogeneity level requires complex hardware and a full-fledged software stack to orchestrate the execution and exploit platform features. For this reason, enabling agile design space exploration becomes a crucial asset for this new class of low-power SoCs. In this scenario, high-level simulators play an essential role in breaking the speed and design effort bottlenecks of cycle-accurate simulators and FPGA prototypes, respectively, while preserving functional and timing accuracy. We present GVSoC, a highly configurable and timing-accurate event-driven simulator that combines the efficiency of C++ models with the flexibility of Python configuration scripts. GVSoC is fully open-sourced, with the intent to drive future research in the area of highly parallel and heterogeneous RISC-V based IoT processors, leveraging three foundational features: Python-based modular configuration of the hardware description, easy calibration of platform parameters for accurate performance estimation, and high-speed simulation. Experimental results show that GVSoC enables practical functional and performance analysis and design exploration at the full-platform level (processors, memory, peripherals and IOs) with a speed-up of 2500x with respect to cycle-accurate simulation with errors typically below 10% for performance analysis.

preprint2022arXiv

HEROv2: Full-Stack Open-Source Research Platform for Heterogeneous Computing

Heterogeneous computers integrate general-purpose host processors with domain-specific accelerators to combine versatility with efficiency and high performance. To realize the full potential of heterogeneous computers, however, many hardware and software design challenges have to be overcome. While architectural and system simulators can be used to analyze heterogeneous computers, they are faced with unavoidable compromises between simulation speed and performance modeling accuracy. In this work we present HEROv2, an FPGA-based research platform that enables accurate and fast exploration of heterogeneous computers consisting of accelerators based on clusters of 32-bit RISC-V cores and an application-class 64-bit ARMv8 or RV64 host processor. HEROv2 allows to seamlessly share data between 64-bit hosts and 32-bit accelerators and comes with a fully open-source on-chip network, a unified heterogeneous programming interface, and a mixed-data-model, mixed-ISA heterogeneous compiler based on LLVM. We evaluate HEROv2 in four case studies from the application level over toolchain and system architecture down to accelerator microarchitecture. We demonstrate how HEROv2 enables effective research and development on the full stack of heterogeneous computing. For instance, the compiler can tile loops and infer data transfers to and from the accelerators, which leads to a speedup of up to 4.4x compared to the original program and in most cases is only 15 % slower than a handwritten implementation, which requires 2.6x more code.

preprint2022arXiv

HULK-V: a Heterogeneous Ultra-low-power Linux capable RISC-V SoC

IoT applications span a wide range in performance and memory footprint, under tight cost and power constraints. High-end applications rely on power-hungry Systems-on-Chip (SoCs) featuring powerful processors, large LPDDR/DDR3/4/5 memories, and supporting full-fledged Operating Systems (OS). On the contrary, low-end applications typically rely on Ultra-Low-Power ucontrollers with a "close to metal" software environment and simple micro-kernel-based runtimes. Emerging applications and trends of IoT require the "best of both worlds": cheap and low-power SoC systems with a well-known and agile software environment based on full-fledged OS (e.g., Linux), coupled with extreme energy efficiency and parallel digital signal processing capabilities. We present HULK-V: an open-source Heterogeneous Linux-capable RISC-V-based SoC coupling a 64-bit RISC-V processor with an 8-core Programmable Multi-Core Accelerator (PMCA), delivering up to 13.8 GOps, up to 157 GOps/W and accelerating the execution of complex DSP and ML tasks by up to 112x over the host processor. HULK-V leverages a lightweight, fully digital memory hierarchy based on HyperRAM IoT DRAM that exposes up to 512 MB of DRAM memory to the host CPU. Featuring HyperRAMs, HULK-V doubles the energy efficiency without significant performance loss compared to featuring power-hungry LPDDR memories, requiring expensive and large mixed-signal PHYs. HULK-V, implemented in Global Foundries 22nm FDX technology, is a fully digital ultra-low-cost SoC running a 64-bit Linux software stack with OpenMP host-to-PMCA offload within a power envelope of just 250 mW.

preprint2022arXiv

In-memory Realization of In-situ Few-shot Continual Learning with a Dynamically Evolving Explicit Memory

Continually learning new classes from a few training examples without forgetting previous old classes demands a flexible architecture with an inevitably growing portion of storage, in which new examples and classes can be incrementally stored and efficiently retrieved. One viable architectural solution is to tightly couple a stationary deep neural network to a dynamically evolving explicit memory (EM). As the centerpiece of this architecture, we propose an EM unit that leverages energy-efficient in-memory compute (IMC) cores during the course of continual learning operations. We demonstrate for the first time how the EM unit can physically superpose multiple training examples, expand to accommodate unseen classes, and perform similarity search during inference, using operations on an IMC core based on phase-change memory (PCM). Specifically, the physical superposition of a few encoded training examples is realized via in-situ progressive crystallization of PCM devices. The classification accuracy achieved on the IMC core remains within a range of 1.28%--2.5% compared to that of the state-of-the-art full-precision baseline software model on both the CIFAR-100 and miniImageNet datasets when continually learning 40 novel classes (from only five examples per class) on top of 60 old classes.

preprint2022arXiv

Kraken: A Direct Event/Frame-Based Multi-sensor Fusion SoC for Ultra-Efficient Visual Processing in Nano-UAVs

Small-size unmanned aerial vehicles (UAV) have the potential to dramatically increase safety and reduce cost in applications like critical infrastructure maintenance and post-disaster search and rescue. Many scenarios require UAVs to shrink toward nano and pico-size form factors. The key open challenge to achieve true autonomy on Nano-UAVs is to run complex visual tasks like object detection, tracking, navigation and obstacle avoidance fully on board, at high speed and robustness, under tight payload and power constraints. With the Kraken SoC, fabricated in 22nm FDX technology, we demonstrate a multi-visual-sensor capability exploiting both event-based and BW/RGB imagers, combining their output for multi-functional visual tasks previously impossible on a single low-power chip for Nano-UAVs. Kraken is an ultra-low-power, heterogeneous SoC architecture integrating three acceleration engines and a vast set of peripherals to enable efficient interfacing with standard frame-based sensors and novel event-based DVS. Kraken enables highly sparse event-driven sub-uJ/inf SNN inference on a dedicated neuromorphic energy-proportional accelerator. Moreover, it can perform frame-based inference by combining a 1.8TOp\s\W 8-cores RISC-V processor cluster with mixed-precision DNN extensions with a 1036TOp\s\W} TNN accelerator.

preprint2022arXiv

Leveraging Tactile Sensors for Low Latency Embedded Smart Hands for Prosthetic and Robotic Applications

Tactile sensing is a crucial perception mode for robots and human amputees in need of controlling a prosthetic device. Today robotic and prosthetic systems are still missing the important feature of accurate tactile sensing. This lack is mainly due to the fact that the existing tactile technologies have limited spatial and temporal resolution and are either expensive or not scalable. In this paper, we present the design and the implementation of a hardware-software embedded system called SmartHand. It is specifically designed to enable the acquisition and the real-time processing of high-resolution tactile information from a hand-shaped multi-sensor array for prosthetic and robotic applications. During data collection, our system can deliver a high throughput of 100 frames per second, which is 13.7x higher than previous related work. We collected a new tactile dataset while interacting with daily-life objects during five different sessions. We propose a compact yet accurate convolutional neural network that requires one order of magnitude less memory and 15.6x fewer computations compared to related work without degrading classification accuracy. The top-1 and top-3 cross-validation accuracies are respectively 98.86% and 99.83%. We further analyze the inter-session variability and obtain the best top-3 leave-one-out-validation accuracy of 77.84%. We deploy the trained model on a high-performance ARM Cortex-M7 microcontroller achieving an inference time of only 100 ms minimizing the response latency. The overall measured power consumption is 505 mW. Finally, we fabricate a new control sensor and perform additional experiments to provide analyses on sensor degradation and slip detection. This work is a step forward in giving robotic and prosthetic devices a sense of touch and demonstrates the practicality of a smart embedded system empowered by tiny machine learning.

preprint2022arXiv

Monte Cimone: Paving the Road for the First Generation of RISC-V High-Performance Computers

The new open and royalty-free RISC-V ISA is attracting interest across the whole computing continuum, from microcontrollers to supercomputers. High-performance RISC-V processors and accelerators have been announced, but RISC-V-based HPC systems will need a holistic co-design effort, spanning memory, storage hierarchy interconnects and full software stack. In this paper, we describe Monte Cimone, a fully-operational multi-blade computer prototype and hardware-software test-bed based on U740, a double-precision capable multi-core, 64-bit RISC-V SoC. Monte Cimone does not aim to achieve strong floating-point performance, but it was built with the purpose of "priming the pipe" and exploring the challenges of integrating a multi-node RISC-V cluster capable of providing an HPC production stack including interconnect, storage and power monitoring infrastructure on RISC-V hardware. We present the results of our hardware/software integration effort, which demonstrate a remarkable level of software and hardware readiness and maturity - showing that the first generation of RISC-V HPC machines may not be so far in the future.

preprint2022arXiv

Pruning In Time (PIT): A Lightweight Network Architecture Optimizer for Temporal Convolutional Networks

Temporal Convolutional Networks (TCNs) are promising Deep Learning models for time-series processing tasks. One key feature of TCNs is time-dilated convolution, whose optimization requires extensive experimentation. We propose an automatic dilation optimizer, which tackles the problem as a weight pruning on the time-axis, and learns dilation factors together with weights, in a single training. Our method reduces the model size and inference latency on a real SoC hardware target by up to 7.4x and 3x, respectively with no accuracy drop compared to a network without dilation. It also yields a rich set of Pareto-optimal TCNs starting from a single model, outperforming hand-designed solutions in both size and accuracy.

preprint2022arXiv

Q-PPG: Energy-Efficient PPG-based Heart Rate Monitoring on Wearable Devices

Hearth Rate (HR) monitoring is increasingly performed in wrist-worn devices using low-cost photoplethysmography (PPG) sensors. However, Motion Artifacts (MAs) caused by movements of the subject's arm affect the performance of PPG-based HR tracking. This is typically addressed coupling the PPG signal with acceleration measurements from an inertial sensor. Unfortunately, most standard approaches of this kind rely on hand-tuned parameters, which impair their generalization capabilities and their applicability to real data in the field. In contrast, methods based on deep learning, despite their better generalization, are considered to be too complex to deploy on wearable devices. In this work, we tackle these limitations, proposing a design space exploration methodology to automatically generate a rich family of deep Temporal Convolutional Networks (TCNs) for HR monitoring, all derived from a single "seed" model. Our flow involves a cascade of two Neural Architecture Search (NAS) tools and a hardware-friendly quantizer, whose combination yields both highly accurate and extremely lightweight models. When tested on the PPG-Dalia dataset, our most accurate model sets a new state-of-the-art in Mean Absolute Error. Furthermore, we deploy our TCNs on an embedded platform featuring a STM32WB55 microcontroller, demonstrating their suitability for real-time execution. Our most accurate quantized network achieves 4.41 Beats Per Minute (BPM) of Mean Absolute Error (MAE), with an energy consumption of 47.65 mJ and a memory footprint of 412 kB. At the same time, the smallest network that obtains a MAE < 8 BPM, among those generated by our flow, has a memory footprint of 1.9 kB and consumes just 1.79 mJ per inference.

preprint2022arXiv

RedMulE: A Compact FP16 Matrix-Multiplication Accelerator for Adaptive Deep Learning on RISC-V-Based Ultra-Low-Power SoCs

The fast proliferation of extreme-edge applications using Deep Learning (DL) based algorithms required dedicated hardware to satisfy extreme-edge applications' latency, throughput, and precision requirements. While inference is achievable in practical cases, online finetuning and adaptation of general DL models are still highly challenging. One of the key stumbling stones is the need for parallel floating-point operations, which are considered unaffordable on sub-100 mW extreme-edge SoCs. We tackle this problem with RedMulE (Reduced-precision matrix Multiplication Engine), a parametric low-power hardware accelerator for FP16 matrix multiplications - the main kernel of DL training and inference - conceived for tight integration within a cluster of tiny RISC-V cores based on the PULP (Parallel Ultra-Low-Power) architecture. In 22 nm technology, a 32-FMA RedMulE instance occupies just 0.07 mm^2 (14% of an 8-core RISC-V cluster) and achieves up to 666 MHz maximum operating frequency, for a throughput of 31.6 MAC/cycle (98.8% utilization). We reach a cluster-level power consumption of 43.5 mW and a full-cluster energy efficiency of 688 16-bit GFLOPS/W. Overall, RedMulE features up to 4.65x higher energy efficiency and 22x speedup over SW execution on 8 RISC-V cores.

preprint2022arXiv

Reducing Neural Architecture Search Spaces with Training-Free Statistics and Computational Graph Clustering

The computational demands of neural architecture search (NAS) algorithms are usually directly proportional to the size of their target search spaces. Thus, limiting the search to high-quality subsets can greatly reduce the computational load of NAS algorithms. In this paper, we present Clustering-Based REDuction (C-BRED), a new technique to reduce the size of NAS search spaces. C-BRED reduces a NAS space by clustering the computational graphs associated with its architectures and selecting the most promising cluster using proxy statistics correlated with network accuracy. When considering the NAS-Bench-201 (NB201) data set and the CIFAR-100 task, C-BRED selects a subset with 70% average accuracy instead of the whole space's 64% average accuracy.

preprint2022arXiv

Robust and Energy-efficient PPG-based Heart-Rate Monitoring

A wrist-worn PPG sensor coupled with a lightweight algorithm can run on a MCU to enable non-invasive and comfortable monitoring, but ensuring robust PPG-based heart-rate monitoring in the presence of motion artifacts is still an open challenge. Recent state-of-the-art algorithms combine PPG and inertial signals to mitigate the effect of motion artifacts. However, these approaches suffer from limited generality. Moreover, their deployment on MCU-based edge nodes has not been investigated. In this work, we tackle both the aforementioned problems by proposing the use of hardware-friendly Temporal Convolutional Networks (TCN) for PPG-based heart estimation. Starting from a single "seed" TCN, we leverage an automatic Neural Architecture Search (NAS) approach to derive a rich family of models. Among them, we obtain a TCN that outperforms the previous state-of-the-art on the largest PPG dataset available (PPGDalia), achieving a Mean Absolute Error (MAE) of just 3.84 Beats Per Minute (BPM). Furthermore, we tested also a set of smaller yet still accurate (MAE of 5.64 - 6.29 BPM) networks that can be deployed on a commercial MCU (STM32L4) which require as few as 5k parameters and reach a latency of 17.1 ms consuming just 0.21 mJ per inference.

preprint2022arXiv

RUAD: unsupervised anomaly detection in HPC systems

The increasing complexity of modern high-performance computing (HPC) systems necessitates the introduction of automated and data-driven methodologies to support system administrators' effort toward increasing the system's availability. Anomaly detection is an integral part of improving the availability as it eases the system administrator's burden and reduces the time between an anomaly and its resolution. However, current state-of-the-art (SoA) approaches to anomaly detection are supervised and semi-supervised, so they require a human-labelled dataset with anomalies - this is often impractical to collect in production HPC systems. Unsupervised anomaly detection approaches based on clustering, aimed at alleviating the need for accurate anomaly data, have so far shown poor performance. In this work, we overcome these limitations by proposing RUAD, a novel Recurrent Unsupervised Anomaly Detection model. RUAD achieves better results than the current semi-supervised and unsupervised SoA approaches. This is achieved by considering temporal dependencies in the data and including long-short term memory cells in the model architecture. The proposed approach is assessed on a complete ten-month history of a Tier-0 system (Marconi100 from CINECA with 980 nodes). RUAD achieves an area under the curve (AUC) of 0.763 in semi-supervised training and an AUC of 0.767 in unsupervised training, which improves upon the SoA approach that achieves an AUC of 0.747 in semi-supervised training and an AUC of 0.734 in unsupervised training. It also vastly outperforms the current SoA unsupervised anomaly detection approach based on clustering, achieving the AUC of 0.548.

preprint2022arXiv

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.

preprint2022arXiv

SNE: an Energy-Proportional Digital Accelerator for Sparse Event-Based Convolutions

Event-based sensors are drawing increasing attention due to their high temporal resolution, low power consumption, and low bandwidth. To efficiently extract semantically meaningful information from sparse data streams produced by such sensors, we present a 4.5TOP/s/W digital accelerator capable of performing 4-bits-quantized event-based convolutional neural networks (eCNN). Compared to standard convolutional engines, our accelerator performs a number of operations proportional to the number of events contained into the input data stream, ultimately achieving a high energy-to-information processing proportionality. On the IBM-DVS-Gesture dataset, we report 80uJ/inf to 261uJ/inf, respectively, when the input activity is 1.2% and 4.9%. Our accelerator consumes 0.221pJ/SOP, to the best of our knowledge it is the lowest energy/OP reported on a digital neuromorphic engine.

preprint2022arXiv

Soft Tiles: Capturing Physical Implementation Flexibility for Tightly-Coupled Parallel Processing Clusters

Modern high-performance computing architectures (Multicore, GPU, Manycore) are based on tightly-coupled clusters of processing elements, physically implemented as rectangular tiles. Their size and aspect ratio strongly impact the achievable operating frequency and energy efficiency, but they should be as flexible as possible to achieve a high utilization for the top-level die floorplan. In this paper, we explore the flexibility range for a high-performance cluster of RISC-V cores with shared L1 memory used to build scalable accelerators, with the goal of establishing a hierarchical implementation methodology where clusters can be modeled as soft tiles to achieve optimal die utilization.

preprint2022arXiv

Spatz: A Compact Vector Processing Unit for High-Performance and Energy-Efficient Shared-L1 Clusters

While parallel architectures based on clusters of Processing Elements (PEs) sharing L1 memory are widespread, there is no consensus on how lean their PE should be. Architecting PEs as vector processors holds the promise to greatly reduce their instruction fetch bandwidth, mitigating the Von Neumann Bottleneck (VNB). However, due to their historical association with supercomputers, classical vector machines include micro-architectural tricks to improve the Instruction Level Parallelism (ILP), which increases their instruction fetch and decode energy overhead. In this paper, we explore for the first time vector processing as an option to build small and efficient PEs for large-scale shared-L1 clusters. We propose Spatz, a compact, modular 32-bit vector processing unit based on the integer embedded subset of the RISC-V Vector Extension version 1.0. A Spatz-based cluster with four Multiply-Accumulate Units (MACUs) needs only 7.9 pJ per 32-bit integer multiply-accumulate operation, 40% less energy than an equivalent cluster built with four Snitch scalar cores. We analyzed Spatz' performance by integrating it within MemPool, a large-scale many-core shared-L1 cluster. The Spatz-based MemPool system achieves up to 285 GOPS when running a 256x256 32-bit integer matrix multiplication, 70% more than the equivalent Snitch-based MemPool system. In terms of energy efficiency, the Spatz-based MemPool system achieves up to 266 GOPS/W when running the same kernel, more than twice the energy efficiency of the Snitch-based MemPool system, which reaches 128 GOPS/W. Those results show the viability of lean vector processors as high-performance and energy-efficient PEs for large-scale clusters with tightly-coupled L1 memory.

preprint2022arXiv

Sub-mW Keyword Spotting on an MCU: Analog Binary Feature Extraction and Binary Neural Networks

Keyword spotting (KWS) is a crucial function enabling the interaction with the many ubiquitous smart devices in our surroundings, either activating them through wake-word or directly as a human-computer interface. For many applications, KWS is the entry point for our interactions with the device and, thus, an always-on workload. Many smart devices are mobile and their battery lifetime is heavily impacted by continuously running services. KWS and similar always-on services are thus the focus when optimizing the overall power consumption. This work addresses KWS energy-efficiency on low-cost microcontroller units (MCUs). We combine analog binary feature extraction with binary neural networks. By replacing the digital preprocessing with the proposed analog front-end, we show that the energy required for data acquisition and preprocessing can be reduced by 29x, cutting its share from a dominating 85% to a mere 16% of the overall energy consumption for our reference KWS application. Experimental evaluations on the Speech Commands Dataset show that the proposed system outperforms state-of-the-art accuracy and energy efficiency, respectively, by 1% and 4.3x on a 10-class dataset while providing a compelling accuracy-energy trade-off including a 2% accuracy drop for a 71x energy reduction.

preprint2022arXiv

TCN Mapping Optimization for Ultra-Low Power Time-Series Edge Inference

Temporal Convolutional Networks (TCNs) are emerging lightweight Deep Learning models for Time Series analysis. We introduce an automated exploration approach and a library of optimized kernels to map TCNs on Parallel Ultra-Low Power (PULP) microcontrollers. Our approach minimizes latency and energy by exploiting a layer tiling optimizer to jointly find the tiling dimensions and select among alternative implementations of the causal and dilated 1D-convolution operations at the core of TCNs. We benchmark our approach on a commercial PULP device, achieving up to 103X lower latency and 20.3X lower energy than the Cube-AI toolkit executed on the STM32L4 and from 2.9X to 26.6X lower energy compared to commercial closed-source and academic open-source approaches on the same hardware target.

preprint2022arXiv

Training Quantised Neural Networks with STE Variants: the Additive Noise Annealing Algorithm

Training quantised neural networks (QNNs) is a non-differentiable optimisation problem since weights and features are output by piecewise constant functions. The standard solution is to apply the straight-through estimator (STE), using different functions during the inference and gradient computation steps. Several STE variants have been proposed in the literature aiming to maximise the task accuracy of the trained network. In this paper, we analyse STE variants and study their impact on QNN training. We first observe that most such variants can be modelled as stochastic regularisations of stair functions; although this intuitive interpretation is not new, our rigorous discussion generalises to further variants. Then, we analyse QNNs mixing different regularisations, finding that some suitably synchronised smoothing of each layer map is required to guarantee pointwise compositional convergence to the target discontinuous function. Based on these theoretical insights, we propose additive noise annealing (ANA), a new algorithm to train QNNs encompassing standard STE and its variants as special cases. When testing ANA on the CIFAR-10 image classification benchmark, we find that the major impact on task accuracy is not due to the qualitative shape of the regularisations but to the proper synchronisation of the different STE variants used in a network, in accordance with the theoretical results.

preprint2022arXiv

Vau da muntanialas: Energy-efficient multi-die scalable acceleration of RNN inference

Recurrent neural networks such as Long Short-Term Memories (LSTMs) learn temporal dependencies by keeping an internal state, making them ideal for time-series problems such as speech recognition. However, the output-to-input feedback creates distinctive memory bandwidth and scalability challenges in designing accelerators for RNNs. We present Muntaniala, an RNN accelerator architecture for LSTM inference with a silicon-measured energy-efficiency of 3.25$TOP/s/W$ and performance of 30.53$GOP/s$ in UMC 65 $nm$ technology. The scalable design of Muntaniala allows running large RNN models by combining multiple tiles in a systolic array. We keep all parameters stationary on every die in the array, drastically reducing the I/O communication to only loading new features and sharing partial results with other dies. For quantifying the overall system power, including I/O power, we built Vau da Muntanialas, to the best of our knowledge, the first demonstration of a systolic multi-chip-on-PCB array of RNN accelerator. Our multi-die prototype performs LSTM inference with 192 hidden states in 330$μs$ with a total system power of 9.0$mW$ at 10$MHz$ consuming 2.95$μJ$. Targeting the 8/16-bit quantization implemented in Muntaniala, we show a phoneme error rate (PER) drop of approximately 3% with respect to floating-point (FP) on a 3L-384NH-123NI LSTM network on the TIMIT dataset.

preprint2021arXiv

A 5 μW Standard Cell Memory-based Configurable Hyperdimensional Computing Accelerator for Always-on Smart Sensing

Hyperdimensional computing (HDC) is a brain-inspired computing paradigm based on high-dimensional holistic representations of vectors. It recently gained attention for embedded smart sensing due to its inherent error-resiliency and suitability to highly parallel hardware implementations. In this work, we propose a programmable all-digital CMOS implementation of a fully autonomous HDC accelerator for always-on classification in energy-constrained sensor nodes. By using energy-efficient standard cell memory (SCM), the design is easily cross-technology mappable. It achieves extremely low power, 5 $μW$ in typical applications, and an energy-efficiency improvement over the state-of-the-art (SoA) digital architectures of up to 3$\times$ in post-layout simulations for always-on wearable tasks such as EMG gesture recognition. As part of the accelerator's architecture, we introduce novel hardware-friendly embodiments of common HDC-algorithmic primitives, which results in 3.3$\times$ technology scaled area reduction over the SoA, achieving the same accuracy levels in all examined targets. The proposed architecture also has a fully configurable datapath using microcode optimized for HDC stored on an integrated SCM based configuration memory, making the design "general-purpose" in terms of HDC algorithm flexibility. This flexibility allows usage of the accelerator across novel HDC tasks, for instance, a newly designed HDC applied to the task of ball bearing fault detection.

preprint2021arXiv

A TinyML Platform for On-Device Continual Learning with Quantized Latent Replays

In the last few years, research and development on Deep Learning models and techniques for ultra-low-power devices in a word, TinyML has mainly focused on a train-then-deploy assumption, with static models that cannot be adapted to newly collected data without cloud-based data collection and fine-tuning. Latent Replay-based Continual Learning (CL) techniques[1] enable online, serverless adaptation in principle, but so farthey have still been too computation and memory-hungry for ultra-low-power TinyML devices, which are typically based on microcontrollers. In this work, we introduce a HW/SW platform for end-to-end CL based on a 10-core FP32-enabled parallel ultra-low-power (PULP) processor. We rethink the baseline Latent Replay CL algorithm, leveraging quantization of the frozen stage of the model and Latent Replays (LRs) to reduce their memory cost with minimal impact on accuracy. In particular, 8-bit compression of the LR memory proves to be almost lossless (-0.26% with 3000LR) compared to the full-precision baseline implementation, but requires 4x less memory, while 7-bit can also be used with an additional minimal accuracy degradation (up to 5%). We also introduce optimized primitives for forward and backward propagation on the PULP processor. Our results show that by combining these techniques, continual learning can be achieved in practice using less than 64MB of memory an amount compatible with embedding in TinyML devices. On an advanced 22nm prototype of our platform, called VEGA, the proposed solution performs onaverage 65x faster than a low-power STM32 L4 microcontroller, being 37x more energy efficient enough for a lifetime of 535h when learning a new mini-batch of data once every minute.

preprint2021arXiv

ChewBaccaNN: A Flexible 223 TOPS/W BNN Accelerator

Binary Neural Networks enable smart IoT devices, as they significantly reduce the required memory footprint and computational complexity while retaining a high network performance and flexibility. This paper presents ChewBaccaNN, a 0.7 mm$^2$ sized binary convolutional neural network (CNN) accelerator designed in GlobalFoundries 22 nm technology. By exploiting efficient data re-use, data buffering, latch-based memories, and voltage scaling, a throughput of 241 GOPS is achieved while consuming just 1.1 mW at 0.4V/154MHz during inference of binary CNNs with up to 7x7 kernels, leading to a peak core energy efficiency of 223 TOPS/W. ChewBaccaNN's flexibility allows to run a much wider range of binary CNNs than other accelerators, drastically improving the accuracy-energy trade-off beyond what can be captured by the TOPS/W metric. In fact, it can perform CIFAR-10 inference at 86.8% accuracy with merely 1.3 $μJ$, thus exceeding the accuracy while at the same time lowering the energy cost by 2.8x compared to even the most efficient and much larger analog processing-in-memory devices, while keeping the flexibility of running larger CNNs for higher accuracy when needed. It also runs a binary ResNet-18 trained on the 1000-class ILSVRC dataset and improves the energy efficiency by 4.4x over accelerators of similar flexibility. Furthermore, it can perform inference on a binarized ResNet-18 trained with 8-bases Group-Net to achieve a 67.5% Top-1 accuracy with only 3.0 mJ/frame -- at an accuracy drop of merely 1.8% from the full-precision ResNet-18.

preprint2021arXiv

CUTIE: Beyond PetaOp/s/W Ternary DNN Inference Acceleration with Better-than-Binary Energy Efficiency

We present a 3.1 POp/s/W fully digital hardware accelerator for ternary neural networks. CUTIE, the Completely Unrolled Ternary Inference Engine, focuses on minimizing non-computational energy and switching activity so that dynamic power spent on storing (locally or globally) intermediate results is minimized. This is achieved by 1) a data path architecture completely unrolled in the feature map and filter dimensions to reduce switching activity by favoring silencing over iterative computation and maximizing data re-use, 2) targeting ternary neural networks which, in contrast to binary NNs, allow for sparse weights which reduce switching activity, and 3) introducing an optimized training method for higher sparsity of the filter weights, resulting in a further reduction of the switching activity. Compared with state-of-the-art accelerators, CUTIE achieves greater or equal accuracy while decreasing the overall core inference energy cost by a factor of 4.8x-21x.

preprint2021arXiv

Memory-Aware Partitioning of Machine Learning Applications for Optimal Energy Use in Batteryless Systems

Sensing systems powered by energy harvesting have traditionally been designed to tolerate long periods without energy. As the Internet of Things (IoT) evolves towards a more transient and opportunistic execution paradigm, reducing energy storage costs will be key for its economic and ecologic viability. However, decreasing energy storage in harvesting systems introduces reliability issues. Transducers only produce intermittent energy at low voltage and current levels, making guaranteed task completion a challenge. Existing ad hoc methods overcome this by buffering enough energy either for single tasks, incurring large data-retention overheads, or for one full application cycle, requiring a large energy buffer. We present Julienning: an automated method for optimizing the total energy cost of batteryless applications. Using a custom specification model, developers can describe transient applications as a set of atomically executed kernels with explicit data dependencies. Our optimization flow can partition data- and energy-intensive applications into multiple execution cycles with bounded energy consumption. By leveraging interkernel data dependencies, these energy-bounded execution cycles minimize the number of system activations and nonvolatile data transfers, and thus the total energy overhead. We validate our methodology with two batteryless cameras running energy-intensive machine learning applications. Results demonstrate that compared to ad hoc solutions, our method can reduce the required energy storage by over 94% while only incurring a 0.12% energy overhead.

preprint2021arXiv

MemPool-3D: Boosting Performance and Efficiency of Shared-L1 Memory Many-Core Clusters with 3D Integration

Three-dimensional integrated circuits promise power, performance, and footprint gains compared to their 2D counterparts, thanks to drastic reductions in the interconnects' length through their smaller form factor. We can leverage the potential of 3D integration by enhancing MemPool, an open-source many-core design with 256 cores and a shared pool of L1 scratchpad memory connected with a low-latency interconnect. MemPool's baseline 2D design is severely limited by routing congestion and wire propagation delay, making the design ideal for 3D integration. In architectural terms, we increase MemPool's scratchpad memory capacity beyond the sweet spot for 2D designs, improving performance in a common digital signal processing kernel. We propose a 3D MemPool design that leverages a smart partitioning of the memory resources across two layers to balance the size and utilization of the stacked dies. In this paper, we explore the architectural and the technology parameter spaces by analyzing the power, performance, area, and energy efficiency of MemPool instances in 2D and 3D with 1 MiB, 2 MiB, 4 MiB, and 8 MiB of scratchpad memory in a commercial 28 nm technology node. We observe a performance gain of 9.1% when running a matrix multiplication on the MemPool-3D design with 4 MiB of scratchpad memory compared to the MemPool 2D counterpart. In terms of energy efficiency, we can implement the MemPool-3D instance with 4 MiB of L1 memory on an energy budget 15% smaller than its 2D counterpart, and even 3.7% smaller than the MemPool-2D instance with one-fourth of the L1 scratchpad memory capacity.

preprint2021arXiv

Mixed-Precision Quantization and Parallel Implementation of Multispectral Riemannian Classification for Brain--Machine Interfaces

With Motor-Imagery (MI) Brain--Machine Interfaces (BMIs) we may control machines by merely thinking of performing a motor action. Practical use cases require a wearable solution where the classification of the brain signals is done locally near the sensor using machine learning models embedded on energy-efficient microcontroller units (MCUs), for assured privacy, user comfort, and long-term usage. In this work, we provide practical insights on the accuracy-cost tradeoff for embedded BMI solutions. Our proposed Multispectral Riemannian Classifier reaches 75.1% accuracy on 4-class MI task. We further scale down the model by quantizing it to mixed-precision representations with a minimal accuracy loss of 1%, which is still 3.2% more accurate than the state-of-the-art embedded convolutional neural network. We implement the model on a low-power MCU with parallel processing units taking only 33.39ms and consuming 1.304mJ per classification.

preprint2021arXiv

Sound Event Detection with Binary Neural Networks on Tightly Power-Constrained IoT Devices

Sound event detection (SED) is a hot topic in consumer and smart city applications. Existing approaches based on Deep Neural Networks are very effective, but highly demanding in terms of memory, power, and throughput when targeting ultra-low power always-on devices. Latency, availability, cost, and privacy requirements are pushing recent IoT systems to process the data on the node, close to the sensor, with a very limited energy supply, and tight constraints on the memory size and processing capabilities precluding to run state-of-the-art DNNs. In this paper, we explore the combination of extreme quantization to a small-footprint binary neural network (BNN) with the highly energy-efficient, RISC-V-based (8+1)-core GAP8 microcontroller. Starting from an existing CNN for SED whose footprint (815 kB) exceeds the 512 kB of memory available on our platform, we retrain the network using binary filters and activations to match these memory constraints. (Fully) binary neural networks come with a natural drop in accuracy of 12-18% on the challenging ImageNet object recognition challenge compared to their equivalent full-precision baselines. This BNN reaches a 77.9% accuracy, just 7% lower than the full-precision version, with 58 kB (7.2 times less) for the weights and 262 kB (2.4 times less) memory in total. With our BNN implementation, we reach a peak throughput of 4.6 GMAC/s and 1.5 GMAC/s over the full network, including preprocessing with Mel bins, which corresponds to an efficiency of 67.1 GMAC/s/W and 31.3 GMAC/s/W, respectively. Compared to the performance of an ARM Cortex-M4 implementation, our system has a 10.3 times faster execution time and a 51.1 times higher energy-efficiency.

preprint2020arXiv

A 0.5GHz 0.35mW LDO-Powered Constant-Slope Phase Interpolator with 0.22$\%$ INL

Clock generators are an essential and critical building block of any communication link, whether it be wired or wireless, and they are increasingly critical given the push for lower I/O power and higher bandwidth in Systems-on-Chip (SoCs) for the Internet-of-Things (IoT). One recurrent issue with clock generators is multiple-phase generation, especially for low-power applications; several methods of phase generation have been proposed, one of which is phase interpolation. We propose a phase interpolator (PI) that employs the concept of constant-slope operation. Consequently, a low-power highly-linear operation is coupled with the wide dynamic range (i.e. phase wrapping) capabilities of a PI. Furthermore, the PI is powered by a low-dropout regulator (LDO) supporting fast transient operation. Implemented in 65-nm CMOS technology, it consumes 350$μ$W at a 1.2-V supply and a 0.5-GHz clock; it achieves energy efficiency 4$\times$-15$\times$ lower than state-of-the-art (SoA) digital-to-time converters (DTCs) and an integral non-linearity (INL) of 2.5$\times$-3.1$\times$ better than SoA PIs, striking a good balance between linearity and energy efficiency.

preprint2020arXiv

Always-On 674uW @ 4GOP/s Error Resilient Binary Neural Networks with Aggressive SRAM Voltage Scaling on a 22nm IoT End-Node

Binary Neural Networks (BNNs) have been shown to be robust to random bit-level noise, making aggressive voltage scaling attractive as a power-saving technique for both logic and SRAMs. In this work, we introduce the first fully programmable IoT end-node system-on-chip (SoC) capable of executing software-defined, hardware-accelerated BNNs at ultra-low voltage. Our SoC exploits a hybrid memory scheme where error-vulnerable SRAMs are complemented by reliable standard-cell memories to safely store critical data under aggressive voltage scaling. On a prototype in 22nm FDX technology, we demonstrate that both the logic and SRAM voltage can be dropped to 0.5Vwithout any accuracy penalty on a BNN trained for the CIFAR-10 dataset, improving energy efficiency by 2.2X w.r.t. nominal conditions. Furthermore, we show that the supply voltage can be dropped to 0.42V (50% of nominal) while keeping more than99% of the nominal accuracy (with a bit error rate ~1/1000). In this operating point, our prototype performs 4Gop/s (15.4Inference/s on the CIFAR-10 dataset) by computing up to 13binary ops per pJ, achieving 22.8 Inference/s/mW while keeping within a peak power envelope of 674uW - low enough to enable always-on operation in ultra-low power smart cameras, long-lifetime environmental sensors, and insect-sized pico-drones.

preprint2020arXiv

Arnold: an eFPGA-Augmented RISC-V SoC for Flexible and Low-Power IoT End-Nodes

A wide range of Internet of Things (IoT) applications require powerful, energy-efficient and flexible end-nodes to acquire data from multiple sources, process and distill the sensed data through near-sensor data analytics algorithms, and transmit it wirelessly. This work presents Arnold: a 0.5 V to 0.8 V, 46.83 uW/MHz, 600 MOPS fully programmable RISC-V Microcontroller unit (MCU) fabricated in 22 nm Globalfoundries GF22FDX (GF22FDX) technology, coupled with a stateof-the-art (SoA) microcontroller to an embedded Field Programmable Gate Array (FPGA). We demonstrate the flexibility of the System-OnChip (SoC) to tackle the challenges of many emerging IoT applications, such as (i) interfacing sensors and accelerators with non-standard interfaces, (ii) performing on-the-fly pre-processing tasks on data streamed from peripherals, and (iii) accelerating near-sensor analytics, encryption, and machine learning tasks. A unique feature of the proposed SoC is the exploitation of body-biasing to reduce leakage power of the embedded FPGA (eFPGA) fabric by up to 18x at 0.5 V, achieving SoA state bitstream-retentive sleep power for the eFPGA fabric, as low as 20.5 uW. The proposed SoC provides 3.4x better performance and 2.9x better energy efficiency than other fabricated heterogeneous re-configurable SoCs of the same class.

preprint2020arXiv

Bonseyes AI Pipeline -- bringing AI to you. End-to-end integration of data, algorithms and deployment tools

Next generation of embedded Information and Communication Technology (ICT) systems are collaborative systems able to perform autonomous tasks. The remarkable expansion of the embedded ICT market, together with the rise and breakthroughs of Artificial Intelligence (AI), have put the focus on the Edge as it stands as one of the keys for the next technological revolution: the seamless integration of AI in our daily life. However, training and deployment of custom AI solutions on embedded devices require a fine-grained integration of data, algorithms, and tools to achieve high accuracy. Such integration requires a high level of expertise that becomes a real bottleneck for small and medium enterprises wanting to deploy AI solutions on the Edge which, ultimately, slows down the adoption of AI on daily-life applications. In this work, we present a modular AI pipeline as an integrating framework to bring data, algorithms, and deployment tools together. By removing the integration barriers and lowering the required expertise, we can interconnect the different stages of tools and provide a modular end-to-end development of AI products for embedded devices. Our AI pipeline consists of four modular main steps: i) data ingestion, ii) model training, iii) deployment optimization and, iv) the IoT hub integration. To show the effectiveness of our pipeline, we provide examples of different AI applications during each of the steps. Besides, we integrate our deployment framework, LPDNN, into the AI pipeline and present its lightweight architecture and deployment capabilities for embedded devices. Finally, we demonstrate the results of the AI pipeline by showing the deployment of several AI applications such as keyword spotting, image classification and object detection on a set of well-known embedded platforms, where LPDNN consistently outperforms all other popular deployment frameworks.

preprint2020arXiv

Combining Learning and Optimization for Transprecision Computing

The growing demands of the worldwide IT infrastructure stress the need for reduced power consumption, which is addressed in so-called transprecision computing by improving energy efficiency at the expense of precision. For example, reducing the number of bits for some floating-point operations leads to higher efficiency, but also to a non-linear decrease of the computation accuracy. Depending on the application, small errors can be tolerated, thus allowing to fine-tune the precision of the computation. Finding the optimal precision for all variables in respect of an error bound is a complex task, which is tackled in the literature via heuristics. In this paper, we report on a first attempt to address the problem by combining a Mathematical Programming (MP) model and a Machine Learning (ML) model, following the Empirical Model Learning methodology. The ML model learns the relation between variables precision and the output error; this information is then embedded in the MP focused on minimizing the number of bits. An additional refinement phase is then added to improve the quality of the solution. The experimental results demonstrate an average speedup of 6.5\% and a 3\% increase in solution quality compared to the state-of-the-art. In addition, experiments on a hardware platform capable of mixed-precision arithmetic (PULPissimo) show the benefits of the proposed approach, with energy savings of around 40\% compared to fixed-precision.

preprint2020arXiv

EEG-TCNet: An Accurate Temporal Convolutional Network for Embedded Motor-Imagery Brain-Machine Interfaces

In recent years, deep learning (DL) has contributed significantly to the improvement of motor-imagery brain-machine interfaces (MI-BMIs) based on electroencephalography(EEG). While achieving high classification accuracy, DL models have also grown in size, requiring a vast amount of memory and computational resources. This poses a major challenge to an embedded BMI solution that guarantees user privacy, reduced latency, and low power consumption by processing the data locally. In this paper, we propose EEG-TCNet, a novel temporal convolutional network (TCN) that achieves outstanding accuracy while requiring few trainable parameters. Its low memory footprint and low computational complexity for inference make it suitable for embedded classification on resource-limited devices at the edge. Experimental results on the BCI Competition IV-2a dataset show that EEG-TCNet achieves 77.35% classification accuracy in 4-class MI. By finding the optimal network hyperparameters per subject, we further improve the accuracy to 83.84%. Finally, we demonstrate the versatility of EEG-TCNet on the Mother of All BCI Benchmarks (MOABB), a large scale test benchmark containing 12 different EEG datasets with MI experiments. The results indicate that EEG-TCNet successfully generalizes beyond one single dataset, outperforming the current state-of-the-art (SoA) on MOABB by a meta-effect of 0.25.

preprint2020arXiv

Energy-Efficient Hardware-Accelerated Synchronization for Shared-L1-Memory Multiprocessor Clusters

The steeply growing performance demands for highly power- and energy-constrained processing systems such as end-nodes of the internet-of-things (IoT) have led to parallel near-threshold computing (NTC), joining the energy-efficiency benefits of low-voltage operation with the performance typical of parallel systems. Shared-L1-memory multiprocessor clusters are a promising architecture, delivering performance in the order of GOPS and over 100 GOPS/W of energy-efficiency. However, this level of computational efficiency can only be reached by maximizing the effective utilization of the processing elements (PEs) available in the clusters. Along with this effort, the optimization of PE-to-PE synchronization and communication is a critical factor for performance. In this work, we describe a light-weight hardware-accelerated synchronization and communication unit (SCU) for tightly-coupled clusters of processors. We detail the architecture, which enables fine-grain per-PE power management, and its integration into an eight-core cluster of RISC-V processors. To validate the effectiveness of the proposed solution, we implemented the eight-core cluster in advanced 22nm FDX technology and evaluated performance and energy-efficiency with tunable microbenchmarks and a set of real-life applications and kernels. The proposed solution allows synchronization-free regions as small as 42 cycles, over 41 times smaller than the baseline implementation based on fast test-and-set access to L1 memory when constraining the microbenchmarks to 10% synchronization overhead. When evaluated on the real-life DSP-applications, the proposed SCU improves performance by up to 92% and 23% on average and energy efficiency by up to 98% and 39% on average.

preprint2020arXiv

Extending the RISC-V ISA for Efficient RNN-based 5G Radio Resource Management

Radio Resource Management (RRM) in 5G mobile communication is a challenging problem for which Recurrent Neural Networks (RNN) have shown promising results. Accelerating the compute-intensive RNN inference is therefore of utmost importance. Programmable solutions are desirable for effective 5G-RRM top cope with the rapidly evolving landscape of RNN variations. In this paper, we investigate RNN inference acceleration by tuning both the instruction set and micro-architecture of a micro-controller-class open-source RISC-V core. We couple HW extensions with software optimizations to achieve an overall improvement in throughput and energy efficiency of 15$\times$ and 10$\times$ w.r.t. the baseline core on a wide range of RNNs used in various RRM tasks.

preprint2020arXiv

FPnew: An Open-Source Multi-Format Floating-Point Unit Architecture for Energy-Proportional Transprecision Computing

The slowdown of Moore's law and the power wall necessitates a shift towards finely tunable precision (a.k.a. transprecision) computing to reduce energy footprint. Hence, we need circuits capable of performing floating-point operations on a wide range of precisions with high energy-proportionality. We present FPnew, a highly configurable open-source transprecision floating-point unit (TP-FPU) capable of supporting a wide range of standard and custom FP formats. To demonstrate the flexibility and efficiency of FPnew in general-purpose processor architectures, we extend the RISC-V ISA with operations on half-precision, bfloat16, and an 8bit FP format, as well as SIMD vectors and multi-format operations. Integrated into a 32-bit RISC-V core, our TP-FPU can speed up execution of mixed-precision applications by 1.67x w.r.t. an FP32 baseline, while maintaining end-to-end precision and reducing system energy by 37%. We also integrate FPnew into a 64-bit RISC-V core, supporting five FP formats on scalars or 2, 4, or 8-way SIMD vectors. For this core, we measured the silicon manufactured in Globalfoundries 22FDX technology across a wide voltage range from 0.45V to 1.2V. The unit achieves leading-edge measured energy efficiencies between 178 Gflop/sW (on FP64) and 2.95 Tflop/sW (on 8-bit mini-floats), and a performance between 3.2 Gflop/s and 25.3 Gflop/s.

preprint2020arXiv

In-memory hyperdimensional computing

Hyperdimensional computing (HDC) is an emerging computational framework that takes inspiration from attributes of neuronal circuits such as hyperdimensionality, fully distributed holographic representation, and (pseudo)randomness. When employed for machine learning tasks such as learning and classification, HDC involves manipulation and comparison of large patterns within memory. Moreover, a key attribute of HDC is its robustness to the imperfections associated with the computational substrates on which it is implemented. It is therefore particularly amenable to emerging non-von Neumann paradigms such as in-memory computing, where the physical attributes of nanoscale memristive devices are exploited to perform computation in place. Here, we present a complete in-memory HDC system that achieves a near optimum trade-off between design complexity and classification accuracy based on three prototypical HDC related learning tasks, namely, language classification, news classification, and hand gesture recognition from electromyography signals. Comparable accuracies to software implementations are demonstrated, experimentally, using 760,000 phase-change memory devices performing analog in-memory computing.

preprint2020arXiv

Leveraging Automated Mixed-Low-Precision Quantization for tiny edge microcontrollers

The severe on-chip memory limitations are currently preventing the deployment of the most accurate Deep Neural Network (DNN) models on tiny MicroController Units (MCUs), even if leveraging an effective 8-bit quantization scheme. To tackle this issue, in this paper we present an automated mixed-precision quantization flow based on the HAQ framework but tailored for the memory and computational characteristics of MCU devices. Specifically, a Reinforcement Learning agent searches for the best uniform quantization levels, among 2, 4, 8 bits, of individual weight and activation tensors, under the tight constraints on RAM and FLASH embedded memory sizes. We conduct an experimental analysis on MobileNetV1, MobileNetV2 and MNasNet models for Imagenet classification. Concerning the quantization policy search, the RL agent selects quantization policies that maximize the memory utilization. Given an MCU-class memory bound of 2MB for weight-only quantization, the compressed models produced by the mixed-precision engine result as accurate as the state-of-the-art solutions quantized with a non-uniform function, which is not tailored for CPUs featuring integer-only arithmetic. This denotes the viability of uniform quantization, required for MCU deployments, for deep weights compression. When also limiting the activation memory budget to 512kB, the best MobileNetV1 model scores up to 68.4% on Imagenet thanks to the found quantization policy, resulting to be 4% more accurate than the other 8-bit networks fitting the same memory constraints.

preprint2020arXiv

LLHD: A Multi-level Intermediate Representation for Hardware Description Languages

Modern Hardware Description Languages (HDLs) such as SystemVerilog or VHDL are, due to their sheer complexity, insufficient to transport designs through modern circuit design flows. Instead, each design automation tool lowers HDLs to its own Intermediate Representation (IR). These tools are monolithic and mostly proprietary, disagree in their implementation of HDLs, and while many redundant IRs exists, no IR today can be used through the entire circuit design flow. To solve this problem, we propose the LLHD multi-level IR. LLHD is designed as simple, unambiguous reference description of a digital circuit, yet fully captures existing HDLs. We show this with our reference compiler on designs as complex as full CPU cores. LLHD comes with lowering passes to a hardware-near structural IR, which readily integrates with existing tools. LLHD establishes the basis for innovation in HDLs and tools without redundant compilers or disjoint IRs. For instance, we implement an LLHD simulator that runs up to 2.4x faster than commercial simulators but produces equivalent, cycle-accurate results. An initial vertically-integrated research prototype is capable of representing all levels of the IR, implements lowering from the behavioural to the structural IR, and covers a sufficient subset of SystemVerilog to support a full CPU design.

preprint2020arXiv

Memory-Latency-Accuracy Trade-offs for Continual Learning on a RISC-V Extreme-Edge Node

AI-powered edge devices currently lack the ability to adapt their embedded inference models to the ever-changing environment. To tackle this issue, Continual Learning (CL) strategies aim at incrementally improving the decision capabilities based on newly acquired data. In this work, after quantifying memory and computational requirements of CL algorithms, we define a novel HW/SW extreme-edge platform featuring a low power RISC-V octa-core cluster tailored for on-demand incremental learning over locally sensed data. The presented multi-core HW/SW architecture achieves a peak performance of 2.21 and 1.70 MAC/cycle, respectively, when running forward and backward steps of the gradient descent. We report the trade-off between memory footprint, latency, and accuracy for learning a new class with Latent Replay CL when targeting an image classification task on the CORe50 dataset. For a CL setting that retrains all the layers, taking 5h to learn a new class and achieving up to 77.3% of precision, a more efficient solution retrains only part of the network, reaching an accuracy of 72.5% with a memory requirement of 300 MB and a computation latency of 1.5 hours. On the other side, retraining only the last layer results in the fastest (867 ms) and less memory hungry (20 MB) solution but scoring 58% on the CORe50 dataset. Thanks to the parallelism of the low-power cluster engine, our HW/SW platform results 25x faster than typical MCU device, on which CL is still impractical, and demonstrates an 11x gain in terms of energy consumption with respect to mobile-class solutions.

preprint2020arXiv

MemPool: A Shared-L1 Memory Many-Core Cluster with a Low-Latency Interconnect

A key challenge in scaling shared-L1 multi-core clusters towards many-core (more than 16 cores) configurations is to ensure low-latency and efficient access to the L1 memory. In this work we demonstrate that it is possible to scale up the shared-L1 architecture: We present MemPool, a 32 bit many-core system with 256 fast RV32IMA "Snitch" cores featuring application-tunable execution units, running at 700 MHz in typical conditions (TT/0.80 V/25°C). MemPool is easy to program, with all the cores sharing a global view of a large L1 scratchpad memory pool, accessible within at most 5 cycles. In MemPool's physical-aware design, we emphasized the exploration, design, and optimization of the low-latency processor-to-L1-memory interconnect. We compare three candidate topologies, analyzing them in terms of latency, throughput, and back-end feasibility. The chosen topology keeps the average latency at fewer than 6 cycles, even for a heavy injected load of 0.33 request/core/cycle. We also propose a lightweight addressing scheme that maps each core private data to a memory bank accessible within one cycle, which leads to performance gains of up to 20% in real-world signal processing benchmarks. The addressing scheme is also highly efficient in terms of energy consumption since requests to local banks consume only half of the energy required to access remote banks. Our design achieves competitive performance with respect to an ideal, non-implementable full-crossbar baseline.

preprint2020arXiv

Optimizing Temporal Convolutional Network inference on FPGA-based accelerators

Convolutional Neural Networks are extensively used in a wide range of applications, commonly including computer vision tasks like image and video classification, recognition, and segmentation. Recent research results demonstrate that multilayer(deep) networks involving mono-dimensional convolutions and dilation can be effectively used in time series and sequences classification and segmentation, as well as in tasks involving sequence modelling. These structures, commonly referred to as Temporal Convolutional Networks (TCNs), have been demonstrated to consistently outperform Recurrent Neural Networks in terms of accuracy and training time [1]. While FPGA-based inference accelerators for classic CNNs are widespread, literature is lacking in a quantitative evaluation of their usability on inference for TCN models. In this paper we present such an evaluation, considering a CNN accelerator with specific features supporting TCN kernels as a reference and a set of state-of-the-art TCNs as a benchmark. Experimental results show that, during TCN execution, operational intensity can be critical for the overall performance. We propose a convolution scheduling based on batch processing that can boost efficiency up to 96% of theoretical peak performance. Overall we can achieve up to 111,8 GOPS/s and power efficiency of 33,9 GOPS/s/W on an Ultrascale+ ZU3EG (up to 10x speedup and 3x power efficiency improvement with respect to pure software implementation).

preprint2020arXiv

pAElla: Edge-AI based Real-Time Malware Detection in Data Centers

The increasing use of Internet-of-Things (IoT) devices for monitoring a wide spectrum of applications, along with the challenges of "big data" streaming support they often require for data analysis, is nowadays pushing for an increased attention to the emerging edge computing paradigm. In particular, smart approaches to manage and analyze data directly on the network edge, are more and more investigated, and Artificial Intelligence (AI) powered edge computing is envisaged to be a promising direction. In this paper, we focus on Data Centers (DCs) and Supercomputers (SCs), where a new generation of high-resolution monitoring systems is being deployed, opening new opportunities for analysis like anomaly detection and security, but introducing new challenges for handling the vast amount of data it produces. In detail, we report on a novel lightweight and scalable approach to increase the security of DCs/SCs, that involves AI-powered edge computing on high-resolution power consumption. The method -- called pAElla -- targets real-time Malware Detection (MD), it runs on an out-of-band IoT-based monitoring system for DCs/SCs, and involves Power Spectral Density of power measurements, along with AutoEncoders. Results are promising, with an F1-score close to 1, and a False Alarm and Malware Miss rate close to 0%. We compare our method with State-of-the-Art MD techniques and show that, in the context of DCs/SCs, pAElla can cover a wider range of malware, significantly outperforming SoA approaches in terms of accuracy. Moreover, we propose a methodology for online training suitable for DCs/SCs in production, and release open dataset and code.

preprint2020arXiv

Performance-Aware Predictive-Model-Based On-Chip Body-Bias Regulation Strategy for an ULP Multi-Core Cluster in 28nm UTBB FD-SOI

The performance and reliability of Ultra-Low-Power (ULP) computing platforms are adversely affected by environmental temperature and process variations. Mitigating the effect of these phenomena becomes crucial when these devices operate near-threshold, due to the magnification of process variations and to the strong temperature inversion effect that affects advanced technology nodes in low-voltage corners, which causes huge overhead due to margining for timing closure. Supporting an extended range of reverse and forward body-bias, UTBB FD-SOI technology provides a powerful knob to compensate for such variations. In this work we propose a methodology to maximize energy efficiency at run-time exploiting body biasing on a ULP platform operating near-threshold. The proposed method relies on on-line performance measurements by means of Process Monitoring Blocks (PMBs) coupled with an on-chip low-power body bias generator. We correlate the measurement performed by the PMBs to the maximum achievable frequency of the system, deriving a predictive model able to estimate it with an error of 9.7% at 0.7V. To minimize the effect of process variations we propose a calibration procedure that allows to use a PMB model affected by only the temperature-induced error, which reduces the frequency estimation error by 2.4x (from 9.7% to 4%). We finally propose a controller architecture relying on the derived models to automatically regulate at run-time the body bias voltage. We demonstrate that adjusting the body bias voltage against environmental temperature variations leads up to 2X reduction in the leakage power and a 15% improvement on the global energy consumption when the system operates at 0.7V and 170MHz

preprint2020arXiv

Prevention of Microarchitectural Covert Channels on an Open-Source 64-bit RISC-V Core

Covert channels enable information leakage across security boundaries of the operating system. Microarchitectural covert channels exploit changes in execution timing resulting from competing access to limited hardware resources. We use the recent experimental support for time protection, aimed at preventing covert channels, in the seL4 microkernel and evaluate the efficacy of the mechanisms against five known channels on Ariane, an open-source 64-bit application-class RISC-V core. We confirm that without hardware support, these defences are expensive and incomplete. We show that the addition of a single-instruction extension to the RISC-V ISA, that flushes microarchitectural state, can enable the OS to close all five evaluated covert channels with low increase in context switch costs and negligible hardware overhead. We conclude that such a mechanism is essential for security.

preprint2020arXiv

RPR: Random Partition Relaxation for Training; Binary and Ternary Weight Neural Networks

We present Random Partition Relaxation (RPR), a method for strong quantization of neural networks weight to binary (+1/-1) and ternary (+1/0/-1) values. Starting from a pre-trained model, we quantize the weights and then relax random partitions of them to their continuous values for retraining before re-quantizing them and switching to another weight partition for further adaptation. We demonstrate binary and ternary-weight networks with accuracies beyond the state-of-the-art for GoogLeNet and competitive performance for ResNet-18 and ResNet-50 using an SGD-based training method that can easily be integrated into existing frameworks.

preprint2020arXiv

Stream Semantic Registers: A Lightweight RISC-V ISA Extension Achieving Full Compute Utilization in Single-Issue Cores

Single-issue processor cores are very energy efficient but suffer from the von Neumann bottleneck, in that they must explicitly fetch and issue the loads/storse necessary to feed their ALU/FPU. Each instruction spent on moving data is a cycle not spent on computation, limiting ALU/FPU utilization to 33% on reductions. We propose "Stream Semantic Registers" to boost utilization and increase energy efficiency. SSR is a lightweight, non-invasive RISC-V ISA extension which implicitly encodes memory accesses as register reads/writes, eliminating a large number of loads/stores. We implement the proposed extension in the RTL of an existing multi-core cluster and synthesize the design for a modern 22nm technology. Our extension provides a significant, 2x to 5x, architectural speedup across different kernels at a small 11% increase in core area. Sequential code runs 3x faster on a single core, and 3x fewer cores are needed in a cluster to achieve the same performance. The utilization increase to almost 100% in leads to a 2x energy efficiency improvement in a multi-core cluster. The extension reduces instruction fetches by up to 3.5x and instruction cache power consumption by up to 5.6x. Compilers can automatically map loop nests to SSRs, making the changes transparent to the programmer.

preprint2019arXiv

Ara: A 1 GHz+ Scalable and Energy-Efficient RISC-V Vector Processor with Multi-Precision Floating Point Support in 22 nm FD-SOI

In this paper, we present Ara, a 64-bit vector processor based on the version 0.5 draft of RISC-V's vector extension, implemented in GlobalFoundries 22FDX FD-SOI technology. Ara's microarchitecture is scalable, as it is composed of a set of identical lanes, each containing part of the processor's vector register file and functional units. It achieves up to 97% FPU utilization when running a 256 x 256 double precision matrix multiplication on sixteen lanes. Ara runs at more than 1 GHz in the typical corner (TT/0.80V/25 oC) achieving a performance up to 33 DP-GFLOPS. In terms of energy efficiency, Ara achieves up to 41 DP-GFLOPS/W under the same conditions, which is slightly superior to similar vector processors found in literature. An analysis on several vectorizable linear algebra computation kernels for a range of different matrix and vector sizes gives insight into performance limitations and bottlenecks for vector processors and outlines directions to maintain high energy efficiency even for small matrix sizes where the vector architecture achieves suboptimal utilization of the available FPUs.

preprint2019arXiv

Online Anomaly Detection in HPC Systems

Reliability is a cumbersome problem in High Performance Computing Systems and Data Centers evolution. During operation, several types of fault conditions or anomalies can arise, ranging from malfunctioning hardware to improper configurations or imperfect software. Currently, system administrator and final users have to discover it manually. Clearly this approach does not scale to large scale supercomputers and facilities: automated methods to detect faults and unhealthy conditions is needed. Our method uses a type of neural network called autoncoder trained to learn the normal behavior of a real, in-production HPC system and it is deployed on the edge of each computing node. We obtain a very good accuracy (values ranging between 90% and 95%) and we also demonstrate that the approach can be deployed on the supercomputer nodes without negatively affecting the computing units performance.

preprint2018arXiv

Anomaly Detection using Autoencoders in High Performance Computing Systems

Anomaly detection in supercomputers is a very difficult problem due to the big scale of the systems and the high number of components. The current state of the art for automated anomaly detection employs Machine Learning methods or statistical regression models in a supervised fashion, meaning that the detection tool is trained to distinguish among a fixed set of behaviour classes (healthy and unhealthy states). We propose a novel approach for anomaly detection in High Performance Computing systems based on a Machine (Deep) Learning technique, namely a type of neural network called autoencoder. The key idea is to train a set of autoencoders to learn the normal (healthy) behaviour of the supercomputer nodes and, after training, use them to identify abnormal conditions. This is different from previous approaches which where based on learning the abnormal condition, for which there are much smaller datasets (since it is very hard to identify them to begin with). We test our approach on a real supercomputer equipped with a fine-grained, scalable monitoring infrastructure that can provide large amount of data to characterize the system behaviour. The results are extremely promising: after the training phase to learn the normal system behaviour, our method is capable of detecting anomalies that have never been seen before with a very good accuracy (values ranging between 88% and 96%).

preprint2018arXiv

Learning to infer: RL-based search for DNN primitive selection on Heterogeneous Embedded Systems

Deep Learning is increasingly being adopted by industry for computer vision applications running on embedded devices. While Convolutional Neural Networks' accuracy has achieved a mature and remarkable state, inference latency and throughput are a major concern especially when targeting low-cost and low-power embedded platforms. CNNs' inference latency may become a bottleneck for Deep Learning adoption by industry, as it is a crucial specification for many real-time processes. Furthermore, deployment of CNNs across heterogeneous platforms presents major compatibility issues due to vendor-specific technology and acceleration libraries. In this work, we present QS-DNN, a fully automatic search based on Reinforcement Learning which, combined with an inference engine optimizer, efficiently explores through the design space and empirically finds the optimal combinations of libraries and primitives to speed up the inference of CNNs on heterogeneous embedded devices. We show that, an optimized combination can achieve 45x speedup in inference latency on CPU compared to a dependency-free baseline and 2x on average on GPGPU compared to the best vendor library. Further, we demonstrate that, the quality of results and time "to-solution" is much better than with Random Search and achieves up to 15x better results for a short-time search.

preprint2018arXiv

Pricing Schemes for Energy-Efficient HPC Systems: Design and Exploration

Energy efficiency is of paramount importance for the sustainability of HPC systems. Energy consumption limits the peak performance of supercomputers and accounts for a large share of total cost of ownership. Consequently, system owners and final users have started exploring mechanisms to trade off performance for power consumption, for example through frequency and voltage scaling. However, only a limited number of studies have been devoted to explore the economic viability of performance scaling solutions and to devise pricing mechanisms fostering a more energy-conscious usage of resources, without adversely impacting return-of-investment on the HPC facility. We present a parametrized model to analyze the impact of frequency scaling on energy and to assess the potential total cost benefits for the HPC facility and the user. We evaluate four pricing schemes, considering both facility manager and the user perspectives. We then perform a design space exploration considering current and near-future HPC systems and technologies.

Luca Benini

Quick read

Decide how to stay connected

How to connect with this researcher

Open a focused conversation when the fit is right

See the researcher in context

Building this graph slice

74 published item(s)

MR2-ByteTrack: CNN and Transformer-based Video Object Detection for AI-augmented Embedded Vision Sensor Nodes

RDKV: Rate-Distortion Bit Allocation for Joint Eviction and Quantization of the KV Cache

TinyMyo: a Tiny Foundation Model for Flexible EMG Signal Processing at the Edge

TROOP: At-the-Roofline Performance for Vector Processors on Low Operational Intensity Workloads

Data-Driven Power Modeling and Monitoring via Hardware Performance Counters Tracking

MX: Enhancing RISC-V&#39;s Vector ISA for Ultra-Low Overhead, Energy-Efficient Matrix Multiplication

An Accurate EEGNet-based Motor-Imagery Brain-Computer Interface for Low-Power Edge Computing

HR-SAR-Net: A Deep Neural Network for Urban Scene Segmentation from High-Resolution SAR Data

InfiniWolf: Energy Efficient Smart Bracelet for Edge Computing with Dual Source Energy Harvesting

Q-EEGNet: an Energy-Efficient 8-bit Quantized Parallel EEGNet Implementation for Edge Motor-Imagery Brain--Machine Interfaces

A Heterogeneous In-Memory Computing Cluster For Flexible End-to-End Inference of Real-World Deep Neural Networks

Adaptive Random Forests for Energy-Efficient Inference on Microcontrollers

Aerosense: A Self-Sustainable And Long-Range Bluetooth Wireless Sensor Node for Aerodynamic and Aeroacoustic Monitoring on Wind Turbines

Bioformers: Embedding Transformers for Ultra-Low Power sEMG-based Gesture Recognition

Constrained Few-shot Class-incremental Learning

Embedding Temporal Convolutional Networks for Energy-Efficient PPG-Based Heart Rate Monitoring

Energy-Efficient Adaptive Machine Learning on IoT End-Nodes With Class-Dependent Confidence

Energy-Efficient Tree-Based EEG Artifact Detection

Exploring Scalable, Distributed Real-Time Anomaly Detection for Bridge Health Monitoring

FANN-on-MCU: An Open-Source Toolkit for Energy-Efficient Neural Network Inference at the Edge of the Internet of Things

GVSoC: A Highly Configurable, Fast and Accurate Full-Platform Simulator for RISC-V based IoT Processors

HEROv2: Full-Stack Open-Source Research Platform for Heterogeneous Computing

HULK-V: a Heterogeneous Ultra-low-power Linux capable RISC-V SoC

In-memory Realization of In-situ Few-shot Continual Learning with a Dynamically Evolving Explicit Memory

Kraken: A Direct Event/Frame-Based Multi-sensor Fusion SoC for Ultra-Efficient Visual Processing in Nano-UAVs

Leveraging Tactile Sensors for Low Latency Embedded Smart Hands for Prosthetic and Robotic Applications

Monte Cimone: Paving the Road for the First Generation of RISC-V High-Performance Computers

Pruning In Time (PIT): A Lightweight Network Architecture Optimizer for Temporal Convolutional Networks

Q-PPG: Energy-Efficient PPG-based Heart Rate Monitoring on Wearable Devices

RedMulE: A Compact FP16 Matrix-Multiplication Accelerator for Adaptive Deep Learning on RISC-V-Based Ultra-Low-Power SoCs

Reducing Neural Architecture Search Spaces with Training-Free Statistics and Computational Graph Clustering

Robust and Energy-efficient PPG-based Heart-Rate Monitoring

RUAD: unsupervised anomaly detection in HPC systems

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

SNE: an Energy-Proportional Digital Accelerator for Sparse Event-Based Convolutions

Soft Tiles: Capturing Physical Implementation Flexibility for Tightly-Coupled Parallel Processing Clusters

Spatz: A Compact Vector Processing Unit for High-Performance and Energy-Efficient Shared-L1 Clusters

Sub-mW Keyword Spotting on an MCU: Analog Binary Feature Extraction and Binary Neural Networks

TCN Mapping Optimization for Ultra-Low Power Time-Series Edge Inference

Training Quantised Neural Networks with STE Variants: the Additive Noise Annealing Algorithm

Vau da muntanialas: Energy-efficient multi-die scalable acceleration of RNN inference

A 5 μW Standard Cell Memory-based Configurable Hyperdimensional Computing Accelerator for Always-on Smart Sensing

A TinyML Platform for On-Device Continual Learning with Quantized Latent Replays

ChewBaccaNN: A Flexible 223 TOPS/W BNN Accelerator

CUTIE: Beyond PetaOp/s/W Ternary DNN Inference Acceleration with Better-than-Binary Energy Efficiency

Memory-Aware Partitioning of Machine Learning Applications for Optimal Energy Use in Batteryless Systems

MemPool-3D: Boosting Performance and Efficiency of Shared-L1 Memory Many-Core Clusters with 3D Integration

Mixed-Precision Quantization and Parallel Implementation of Multispectral Riemannian Classification for Brain--Machine Interfaces

Sound Event Detection with Binary Neural Networks on Tightly Power-Constrained IoT Devices

A 0.5GHz 0.35mW LDO-Powered Constant-Slope Phase Interpolator with 0.22$\%$ INL

Always-On 674uW @ 4GOP/s Error Resilient Binary Neural Networks with Aggressive SRAM Voltage Scaling on a 22nm IoT End-Node

Arnold: an eFPGA-Augmented RISC-V SoC for Flexible and Low-Power IoT End-Nodes

Bonseyes AI Pipeline -- bringing AI to you. End-to-end integration of data, algorithms and deployment tools

Combining Learning and Optimization for Transprecision Computing

EEG-TCNet: An Accurate Temporal Convolutional Network for Embedded Motor-Imagery Brain-Machine Interfaces

Energy-Efficient Hardware-Accelerated Synchronization for Shared-L1-Memory Multiprocessor Clusters

Extending the RISC-V ISA for Efficient RNN-based 5G Radio Resource Management

FPnew: An Open-Source Multi-Format Floating-Point Unit Architecture for Energy-Proportional Transprecision Computing

In-memory hyperdimensional computing

Leveraging Automated Mixed-Low-Precision Quantization for tiny edge microcontrollers

LLHD: A Multi-level Intermediate Representation for Hardware Description Languages

Memory-Latency-Accuracy Trade-offs for Continual Learning on a RISC-V Extreme-Edge Node

MemPool: A Shared-L1 Memory Many-Core Cluster with a Low-Latency Interconnect

Optimizing Temporal Convolutional Network inference on FPGA-based accelerators

pAElla: Edge-AI based Real-Time Malware Detection in Data Centers

Performance-Aware Predictive-Model-Based On-Chip Body-Bias Regulation Strategy for an ULP Multi-Core Cluster in 28nm UTBB FD-SOI

Prevention of Microarchitectural Covert Channels on an Open-Source 64-bit RISC-V Core

RPR: Random Partition Relaxation for Training; Binary and Ternary Weight Neural Networks

Stream Semantic Registers: A Lightweight RISC-V ISA Extension Achieving Full Compute Utilization in Single-Issue Cores

Ara: A 1 GHz+ Scalable and Energy-Efficient RISC-V Vector Processor with Multi-Precision Floating Point Support in 22 nm FD-SOI

Online Anomaly Detection in HPC Systems

Anomaly Detection using Autoencoders in High Performance Computing Systems

MX: Enhancing RISC-V's Vector ISA for Ultra-Low Overhead, Energy-Efficient Matrix Multiplication