Search

This paper introduces BrainFuseNet, a novel lightweight seizure detection network based on the sensor fusion of electroencephalography (EEG) with photoplethysmography (PPG) and accelerometer (ACC) signals, tailored for low-channel count wearable systems. BrainFuseNet utilizes the Sensitivity-Specificity Weighted Cross-Entropy (SSWCE), an innovative loss function incorporating sensitivity and specificity, to address the challenge of heavily unbalanced datasets. The BrainFuseNet-SSWCE approach successfully detects $93.5\%$ seizure events on the CHB-MIT dataset ($76.34\%$ sample-based sensitivity), for EEG-based classification with only four channels. On the PEDESITE dataset, we demonstrate a sample-based sensitivity and false positive rate of $60.66\%$ and $1.18$ FP/h, respectively, when considering EEG data alone. Additionally, we demonstrate that integrating PPG signals increases the sensitivity to $61.22\%$ (successfully detecting $92\%$ seizure events) while decreasing the number of false positives to $1.0$ FP/h. Finally, when ACC data are also considered, the sensitivity increases to $64.28\%$ (successfully detecting $95\%$ seizure events) and the number of false positives drops to only $0.21$ FP/h for sample-based estimations, with less than one false alarm per day when considering event-based estimations. BrainFuseNet is resource-friendly and well-suited for implementation on low-power embedded platforms, and we evaluate its performance on GAP9, a state-of-the-art parallel ultra-low power (PULP) microcontroller for tiny Machine Learning applications on wearables. The implementation on GAP9 achieves an energy efficiency of $21.43$ GMAC/s/W, with an energy consumption per inference of only $0.11$ mJ at high performance ($412.54$ MMAC/s). The BrainFuseNet-SSWCE method demonstrates effective and accurate seizure detection on heavily imbalanced datasets while achieving state-of-the-art performance in the false positive rate and being well-suited for deployment on energy-constrained edge devices.

Spike extraction by blind source separation (BSS) algorithms can successfully extract physiologically meaningful information from the sEMG signal, as they are able to identify motor unit (MU) discharges involved in muscle contractions. However, BSS approaches are currently restricted to isometric contractions, limiting their applicability in real-world scenarios. We present a strategy to track MUs across different dynamic hand gestures using adaptive independent component analysis (ICA): first, a pool of MUs is identified during isometric contractions, and the decomposition parameters are stored; during dynamic gestures, the decomposition parameters are updated online in an unsupervised fashion, yielding the refined MUs; then, a Pan-Tompkins-inspired algorithm detects the spikes in each MUs; finally, the identified spikes are fed to a classifier to recognize the gesture. We validate our approach on a 4-subject, 7-gesture + rest dataset collected with our custom 16-channel dry sEMG armband, achieving an average balanced accuracy of 85.58 $\pm$ 14.91% and macro-F1 score of 85.86 $\pm$ 14.48%. We deploy our solution onto GAP9, a parallel ultra-low-power microcontroller specialized for computation-intensive linear algebra applications at the edge, obtaining an energy consumption of 4.72 mJ @ 240 MHz and a latency of 121.3 ms for each 200 ms-long window of sEMG signal.

Surface electromyography (sEMG) is a State-of-the-Art (SoA) sensing modality for non-invasive human-machine interfaces for consumer, industrial, and rehabilitation use cases. The main limitation of the current sEMG-driven control policies is the sEMG's inherent variability, especially cross-session due to sensor repositioning; this limits the generalization of the Machine/Deep Learning (ML/DL) in charge of the signal-to-command mapping. The other hot front on the ML/DL side of sEMG-driven control is the shift from the classification of fixed hand positions to the regression of hand kinematics and dynamics, promising a more versatile and fluid control. We present an incremental online-training strategy for sEMG-based estimation of simultaneous multi-finger forces, using a small Temporal Convolutional Network suitable for embedded learning-on-device. We validate our method on the HYSER dataset, cross-day. Our incremental online training reaches a cross-day Mean Absolute Error (MAE) of (9.58 ± 3.89)% of the Maximum Voluntary Contraction on HYSER's RANDOM dataset of improvised, non-predefined force sequences, which is the most challenging and closest to real scenarios. This MAE is on par with an accuracy-oriented, non-embeddable offline training exploiting more epochs. Further, we demonstrate that our online training approach can be deployed on the GAP9 ultra-low power microcontroller, obtaining a latency of 1.49 ms and an energy draw of just 40.4 uJ per forward-backward-update step. These results show that our solution fits the requirements for accurate and real-time incremental training-on-device.

Capacitive noncontact electrodes offer a promising solution to improve comfort and unobtrusiveness in continuous monitoring ECG systems which are becoming more widespread in wearable devices. Most implementations based on off-the-shelf components require rather large electrodes and power budgets while custom application-specific integrated circuits (ASICs) face significantly higher adoption challenges. In this article, we present a novel micropower ultrahigh impedance front-end with an equivalent input capacitance of 70 fF in the ECG band, enabling contactless ECG acquisition with electrodes as small as 78.5 mm2 with a power consumption of only $15 \, \mu \text{W}$ . The front-end implements a novel power supply bootstrapping architecture and is integrated in a full acquisition system which includes a 24-bit analog to digital converter (ADC) and a bluetooth low energy (BLE)-capable Cortex M4 processor, achieving up to ten days of continuous operation on a 600 mAh PR44 battery.

Processors using the open RISC-V instruction set architecture (ISA) are finding increasing adoption in the embedded world. Many embedded use cases have real-time constraints and require flexible, predictable, and fast reactive handling of incoming events. However, RISC-V processors are still lagging in this area compared to more mature proprietary architectures, such as ARM Cortex-M and TriCore, which have been tuned for years. The default interrupt controller standardized by RISC-V, the core local interruptor (CLINT), lacks configurability in prioritization and preemption of interrupts. The RISC-V core local interrupt controller (CLIC) specification addresses this concern by enabling preemptible, low-latency vectored interrupts while also envisioning optional extensions to improve interrupt latency. In this work, we implement a CLIC for the CV32E40P, an industrially supported open-source 32-bit microcontroller unit (MCU)-class RISC-V core, and enhance it with fastirq: a custom extension that provides interrupt latency as low as six cycles. We call CV32RT our enhanced core. To the best of our knowledge, CV32RT is the first fully open-source RV32 core with competitive interrupt-handling features compared to the Arm Cortex-M series and TriCore. The proposed extensions are also demonstrated to improve task context switching in real-time operating systems (RTOSs).

Measurement of muscle activity via wearable instrumentation is of great interest for medical and sport science applications: examples include the control of prostheses, robotics, or therapy and training. This article focuses on the measurement of pennation angles in lateral gastrocnemius (LG) muscles directly from raw ultrasound (US) data. We rely on a reduced number of channels (32) to demonstrate accurate and real-time-compatible predictions of pennation angles for low-power wearable US devices. B-mode images are reconstructed by means of a delay-and-sum beamformer, and an automated computer vision tool (AEPUS) is employed to perform annotations. The labels are then used as ground truth for training an XGBoost regressor based on statistical features extracted from raw US data. Feature importance analyses and deployment-oriented XGBoost model design allow reducing the memory footprint down to only 11 kB. Experimental verification demonstrates a root-mean-square error (RMSE) of 1.6°, comparable to manual annotations by experts. To evaluate the performance on a low-power embedded processor, the complete algorithm is implemented on a recently released RISC-V-based parallel low-power processor (GAP9, from Greenwaves Technologies). Experimental measurements show an inference time and energy consumption as low as 1.31 ms and 43.32 $\mu \text{J}$ , respectively, with an average power envelope of 33.14 mW and a peak power of only 44 mW. The achieved accuracy, low memory footprint, and low power demonstrate the feasibility of pennation angle measurements directly on the probe for resource-constrained smart wearable US devices with a small number of channels.

The long-term, continuous analysis of electroencephalography (EEG) signals on wearable devices to automatically detect seizures in epileptic patients is a high-potential application field for deep neural networks, and specifically for transformers, which are highly suited for end-to-end time series processing without handcrafted feature extraction. In this work, we propose a small-scale transformer detector, the EEGformer, compatible with unobtrusive acquisition setups that use only the temporal channels. EEGformer is the result of a hardware-oriented design exploration, aiming for efficient execution on tiny low-power micro-controller units (MCUs) and low latency and false alarm rate to increase patient and caregiver acceptance.Tests conducted on the CHB-MIT dataset show a 20% reduction of the onset detection latency with respect to the state-of-the-art model for temporal acquisition, with a competitive 73% seizure detection probability and 0.15 false-positive-per-hour (FP/h). Further investigations on a novel and challenging scalp EEG dataset result in the successful detection of 88% of the annotated seizure events, with 0.45 FP/h.We evaluate the deployment of the EEGformer on three commercial low-power computing platforms: the single-core Apollo4 MCU and the GAP8 and GAP9 parallel MCUs. The most efficient implementation (on GAP9) results in as low as 13.7 ms and 0.31 mJ per inference, demonstrating the feasibility of deploying the EEGformer on wearable seizure detection systems with reduced channel count and multi-day battery duration.

A brain–machine interface (BMI) based on motor imagery (MI) enables the control of devices using brain signals while the subject imagines performing a movement. It plays a key role in prosthesis control and motor rehabilitation. To improve user comfort, preserve data privacy, and reduce the system’s latency, a new trend in wearable BMIs is to execute algorithms on low-power microcontroller units (MCUs) embedded on edge devices to process the electroencephalographic (EEG) data in real-time close to the sensors. However, most of the classification models presented in the literature are too resource-demanding for low-power MCUs. This article proposes an efficient convolutional neural network (CNN) for EEG-based MI classification that achieves comparable accuracy while being orders of magnitude less resource-demanding and significantly more energy-efficient than state-of-the-art (SoA) models. To further reduce the model complexity, we propose an automatic channel selection method based on spatial filters and quantize both weights and activations to 8-bit precision with negligible accuracy loss. Finally, we implement and evaluate the proposed models on leading-edge parallel ultralow-power (PULP) MCUs. The final two-class solution consumes as little as $30 ~\mu \text{J}$ /inference with a runtime of 2.95 ms/inference and an accuracy of 82.51% while using $6.4\times $ fewer EEG channels, becoming the new SoA for embedded MI-BMI and defining a new Pareto frontier in the three-way trade-off among accuracy, resource cost, and power usage.

Emerging artificial intelligence-enabled Internet-of-Things (AI-IoT) system-on-chip (SoC) for augmented reality, personalized healthcare, and nanorobotics need to run many diverse tasks within a power envelope of a few tens of mW over a wide range of operating conditions: compute-intensive but strongly quantized deep neural network (DNN) inference, as well as signal processing and control requiring high-precision floating point. We present MARSELLUS, an all-digital heterogeneous SoC for AI-IoT end-nodes fabricated in GlobalFoundries 22 nm FDX that combines: 1) a general-purpose cluster of 16 RISC-V digital signal processing (DSP) cores attuned for the execution of a diverse range of workloads exploiting 4- and 2-bit arithmetic extensions (XpulpNN), combined with fused multiply accumulate (MAC) and LOAD operations and floating-point support; 2) a 2–8 bit reconfigurable binary engine (RBE) to accelerate $A3 3$ and $A1 1$ (pointwise) convolutions in DNNs; 3) a set of ON-chip monitoring (OCM) blocks connected to an adaptive body biasing (ABB) generator and a hardware control loop, enabling on- the-fly adaptation of transistor threshold voltages. MARSELLUS achieves up to 180 Gop/s or 3.32 Top/s/W on 2-bit precision arithmetic in software, and up to 637 Gop/s or 12.4 Top/s/W on hardware-accelerated DNN layers.

Data transfers are essential in today's computing systems as latency and complex memory access patterns are increasingly challenging to manage. Direct memory access engines (DMAES) are critically needed to transfer data independently of the processing elements, hiding latency and achieving high throughput even for complex access patterns to high-latency memory. With the prevalence of heterogeneous systems, DMAEs must operate efficiently in increasingly diverse environments. This work proposes a modular and highly configurable open-source DMAE architecture called intelligent DMA (iDMA), split into three parts that can be composed and customized independently. The front-end implements the control plane binding to the surrounding system. The mid-end accelerates complex data transfer patterns such as multi-dimensional transfers, scattering, or gathering. The back-end interfaces with the on-chip communication fabric (data plane). We assess the efficiency of iDMA in various instantiations: In high-performance systems, we achieve speedups of up to 15.8$\boldsymbol{\times}$× with only 1% additional area compared to a base system without a DMAE. We achieve an area reduction of 10% while improving ML inference performance by 23% in ultra-low-energy edge AI systems over an existing DMAE solution. We provide area, timing, latency, and performance characterization to guide its instantiation in various systems.

Wearable ultrasound (US) is a novel sensing approach that shows promise in multiple application domains, and specifically in hand gesture recognition (HGR). In fact, US enables to collect information from deep musculoskeletal structures at high spatiotemporal resolution and high signal-to-noise ratio, making it a perfect candidate to complement surface electromyography for improved accuracy performance and on-the-edge classification. However, existing wearable solutions for US-based gesture recognition are not sufficiently low power for continuous, long-term operation. On top of that, practical hardware limitations of wearable US devices (limited power budget, reduced wireless throughput, and restricted computational power) set the need for the compressed size of models for feature extraction and classification. To overcome these limitations, this article presents a novel end-to-end approach for feature extraction from raw musculoskeletal US data suited for edge computing, coupled with an armband for HGR based on a truly wearable (12 cm2, 9 g), ultralow-power (ULP) (16 mW) US probe. The proposed approach uses a 1-D convolutional autoencoder (CAE) to compress raw US data by $20\times $ while preserving the main amplitude features of the envelope signal. The latent features of the autoencoder are used to train an XGBoost classifier for HGR on datasets collected with a custom US armband, considering armband removal/repositioning in between sessions. Our approach achieves a classification accuracy of 96%. Furthermore, the proposed unsupervised feature extraction approach offers generalization capabilities for intersubject use, as demonstrated by testing the pretrained encoder on a different subject and conducting posttraining analysis, revealing that the operations performed by the encoder are subject-independent. The autoencoder is also quantized to 8-bit integers and deployed on a ULP wearable US probe along with the XGBoost classifier, allowing for a gesture recognition rate $\geq 25$ Hz and leading to 21% lower power consumption [at 30 frames/s (FPS)] compared to the conventional approach (raw data transmission and remote processing).

Extended reality (XR) applications are machine learning (ML)-intensive, featuring deep neural networks (DNNs) with millions of weights, tightly latency-bound (10–20 ms end-to-end), and power-constrained (low tens of mW average power). While ML performance and efficiency can be achieved by introducing neural engines within low-power systems-on-chip (SoCs), system-level power for nontrivial DNNs depends strongly on the energy of non-volatile memory (NVM) access for network weights. This work introduces Siracusa, a near-sensor heterogeneous SoC for next-generation XR devices manufactured in 16 nm CMOS. Siracusa couples an octa-core cluster of RISC-V digital signal processing (DSP) cores with a novel tightly coupled “At-Memory” integration between a state-of-the-art digital neural engine called N-EUREKA and an on-chip NVM based on magnetoresistive random access memory (MRAM), achieving 1.7 $\times $ higher throughput and 3 $\times $ better energy efficiency than XR SoCs using NVM as background memory. The fabricated SoC prototype achieves an area efficiency of 65.2 GOp/s/mm2 and a peak energy efficiency of 8.84 TOp/J for DNN inference while supporting complex, heterogeneous application workloads, which combine ML with conventional signal processing and control.

Hierarchical very-large-scale integration (VLSI) flows are an understudied yet critical approach to achieving design closure at giga-scale complexity and gigahertz frequency targets. This article proposes a novel hierarchical physical design flow enabling the building of high-density and commercial-quality two-tier face-to-face-bonded hierarchical 3-D ICs. Complemented with an automated floorplanning solution, the flow allows for system-level physical and architectural exploration of 3-D designs. As a result, we significantly reduce the associated manufacturing cost compared to existing 3-D implementation flows and, for the first time, achieve cost competitiveness against the 2-D reference in large modern designs. Experimental results on complex industrial and open manycore processors demonstrate in two advanced nodes that the proposed flow provides major power, performance, and area/cost (PPAC) improvements of 1.2 - 2.2 $\times $ compared with 2-D, where all metrics are improved simultaneously, including up to $\mathrm {20 \%}$ power savings.

Vector processing is highly effective in boosting processor performance and efficiency for data-parallel workloads. In this paper, we present Ara2, the first fully open-source vector processor to support the RISC-V V 1.0 frozen ISA. We evaluate Ara2's performance on a diverse set of data-parallel kernels for various problem sizes and vector-unit configurations, achieving an average functional-unit utilization of 95% on the most computationally intensive kernels. We pinpoint performance boosters and bottlenecks, including the scalar core, memories, and vector architecture, providing insights into the main vector architecture's performance drivers. Leveraging the openness of the design, we implement Ara2 in a 22nm technology, characterize its PPA metrics on various configurations (2-16 lanes), and analyze its microarchitecture and implementation bottlenecks. Ara2 achieves a state-of-the-art energy efficiency of 37.8 DP-GFLOPS/W (0.8V) and 1.35GHz of clock frequency (critical path: $\sim$∼40 FO4 gates). Finally, we explore the performance and energy-efficiency trade-offs of multi-core vector processors: we find that multiple vector cores help overcome the scalar core issue-rate bound that limits short-vector performance. For example, a cluster of eight 2-lane Ara2 (16 FPUs) achieves more than 3x better performance than a 16-lane single-core Ara2 (16 FPUs) when executing a 32x32x32 matrix multiplication, with 1.5x improved energy efficiency.

The rapid advancement of energy-efficient parallel ultra-low-power (ULP) $\mu $ controllers units (MCUs) is enabling the development of autonomous nano-sized unmanned aerial vehicles (nano-UAVs). These sub-10cm drones represent the next generation of unobtrusive robotic helpers and ubiquitous smart sensors. However, nano-UAVs face significant power and payload constraints while requiring advanced computing capabilities akin to standard drones, including real-time Machine Learning (ML) performance and the safe co-existence of general-purpose and real-time OSs. Although some advanced parallel ULP MCUs offer the necessary ML computing capabilities within the prescribed power limits, they rely on small main memories (< 1MB) and $\mu $ controller-class CPUs with no virtualization or security features, and hence only support simple bare-metal runtimes. In this work, we present Shaheen, a 9mm $^{\textbf {2}}~200$ mW SoC implemented in 22nm FDX technology. Differently from state-of-the-art MCUs, Shaheen integrates a Linux-capable RV64 core, compliant with the v1.0 ratified Hypervisor extension and equipped with timing channel protection, along with a low-cost and low-power memory controller exposing up to 512MB of off-chip low-cost low-power HyperRAM directly to the CPU. At the same time, it integrates a fully programmable energy- and area-efficient multi-core cluster of RV32 cores optimized for general-purpose DSP as well as reduced- and mixed-precision ML. To the best of the authors’ knowledge, it is the first silicon prototype of a ULP SoC coupling the RV64 and RV32 cores in a heterogeneous host+accelerator architecture fully based on the RISC-V ISA. We demonstrate the capabilities of the proposed SoC on a wide range of benchmarks relevant to nano-UAV applications including general-purpose DSP as well as inference and online learning of quantized DNNs. The cluster can deliver up to 90GOp/s and up to 1.8TOp/s/W on 2-bit integer kernels and up to 7.9GFLOp/s and up to 150GFLOp/s/W on 16-bit FP kernels.

Shared L1 memory clusters are a common architectural pattern (e.g., in GPGPUs) for building efficient and flexible multi-processing-element (PE) engines. However, it is a common belief that these tightly-coupled clusters would not scale beyond a few tens of PEs. In this work, we tackle scaling shared L1 clusters to hundreds of PEs while supporting a flexible and productive programming model and maintaining high efficiency. We present MemPool, a manycore system with 256 RV32IMAXpulpimg “Snitch” cores featuring application-tunable functional units. We designed and implemented an efficient low-latency PE to L1-memory interconnect, an optimized instruction path to ensure each PE's independent execution, and a powerful DMA engine and system interconnect to stream data in and out. MemPool is easy to program, with all the cores sharing a global view of a large, multi-banked, L1 scratchpad memory, accessible within at most five cycles in the absence of conflicts. We provide multiple runtimes to program MemPool at different abstraction levels and illustrate its versatility with a wide set of applications. MemPool runs at 600 MHz (60 gate delays) in typical conditions (TT/0.80 V/25 ${}^{\boldsymbol{\circ}}$∘C) in 22 nm FDX technology and achieves a performance of up to 229 GOPS or 180 GOPS/W with less than 2% of execution stalls.

Sparse linear algebra is crucial in many application domains, but challenging to handle efficiently in both software and hardware, with one- and two-sided operand sparsity handled with distinct approaches. In this work, we enhance an existing memory-streaming RISC-V ISA extension to accelerate both one- and two-sided operand sparsity on widespread sparse tensor formats like compressed sparse row (CSR) and compressed sparse fiber (CSF) by accelerating the underlying operations of streaming indirection, intersection, and union. Our extensions enable single-core speedups over an optimized RISC-V baseline of up to 7.0x, 7.7x, and 9.8x on sparse-dense multiply, sparse-sparse multiply, and sparse-sparse addition, respectively, and peak FPU utilizations of up to 80% on sparse-dense problems. On an eight-core cluster, sparse-dense and sparse-sparse matrix-vector multiply using real-world matrices are up to 4.9x and 5.9x faster and up to 2.9x and 3.0x more energy efficient. We explore further applications for our extensions, such as stencil codes and graph pattern matching. Compared to recent CPU, GPU, and accelerator approaches, our extensions enable higher flexibility on data representation, degree of sparsity, and dataflow at a minimal hardware footprint, adding only 1.8% in area to a compute cluster. A cluster with our extensions running CSR matrix-vector multiplication achieves 9.9x and 1.7x higher peak floating-point utilizations than recent highly optimized sparse data structures and libraries for CPU and GPU, respectively, even when accounting for off-chip main memory (HBM) and on-chip interconnect latency and bandwidth effects.

Nanosize drones hold enormous potential to explore unknown and complex environments. Their small size makes them agile and safe for operation close to humans and allows them to navigate through narrow spaces. However, their tiny size and payload restrict the possibilities for onboard computation and sensing, making fully autonomous flight extremely challenging. The first step toward full autonomy is reliable obstacle avoidance, which has proven to be challenging by itself in a generic indoor environment. Current approaches utilize vision-based or 1-D sensors to support nanodrone perception algorithms. This article presents a lightweight obstacle avoidance system based on a novel millimeter form factor 64 pixels multizone time-of-flight (ToF) sensor and a generalized model-free control policy. In-field tests are based on the Crazyflie 2.1, extended by a custom multizone ToF deck, featuring a total flight mass of 35 g. The algorithm only uses 0.3% of the onboard processing power (${210}\,{\mu }\mathrm{{s}}$ execution time) with a frame rate of 15 f/s. The presented autonomous nanosize drone reaches 100% reliability at 0.5 m/s in a generic and previously unexplored indoor environment.

Virtualization is a key technology used in a wide range of applications, from cloud computing to embedded systems. Over the last few years, mainstream computer architectures were extended with hardware virtualization support, giving rise to a set of virtualization technologies (e.g., Intel VT and Arm VE) that are now proliferating in modern processors and systems on chip (SoCs). In this article, we describe our work on hardware virtualization support in the RISC-V CVA6 core. Our contribution is multifold and encompasses architecture, microarchitecture, and design space exploration (DSE). In particular, we highlight the design of a set of microarchitectural enhancements [i.e., G-stage translation lookaside buffer (GTLB) and second-level TLB (L2 TLB)] to alleviate the virtualization performance overhead. We also perform a DSE and accompanying postlayout simulations (based on 22-nm FDX technology) to assess performance, power, and area (PPA). Furthermore, we map design variants on a field-programmable gate array (FPGA) platform (Genesys 2) to assess the functional performance–area tradeoff. Based on the DSE, we select an optimal design point for the CVA6 with hardware virtualization support. For this optimal hardware configuration, we collected functional performance results by running the MiBench benchmark on Linux atop Bao hypervisor for a single-core configuration. We observed a performance speedup of up to 16% (approximately 12.5% on average) compared with virtualization-aware nonoptimized design at the minimal cost of 0.78% in area and 0.33% in power. Finally, all works described in this article are publicly available and open-sourced for the community to further evaluate additional design configurations and software stacks.

The nature and heterogeneity of modern workloads force hardware designers to choose between general-purpose processors, which come with superior flexibility, and highly-tailored accelerators that boost performance and power efficiency at the cost of extreme specialization. One of the most promising solutions that couple the flexibility of a processor with the performance and efficiency of an accelerator is the vector processor architecture. Since RISC-V has only recently frozen its vector ISA extension, no open-source RISC-V-based vector processor has been fabricated and characterized. This brief presents the Yun SoC, featuring the first implementation of an open-source RISC-V-based vector processor in TSMC’s 65-nm technology. Our efficient 4-lane design achieves almost peak theoretical performance on large matrix multiplication problems with an FPU utilization of almost 90%. Yun, with a critical path of 30 FO4 inverter delays, achieves a peak performance of 2.83 GFLOPSDP (at 400 MHz and 1.5 V), a leading-edge area efficiency of 3 GFLOPSP/cycle/MGE, and a peak energy efficiency of 10.8 GFLOPSDP (at 100 MHz and 0.85 V). Yun supports integer (64-bit, 32-bit, 16-bit, and 8-bit)) and floating-point (64-bit and 32-bit) SIMD data formats, as required by ML and data analytics workloads.

Power and cost constraints in the Internet-of-Things (IoT) extreme-edge and TinyML domains, coupled with increasing performance requirements, motivate a trend toward heterogeneous architectures. These designs use energy-efficient application-class host processors to coordinate compute-specialized multicore accelerators, amortizing the architectural costs of operating system support and external communication. This brief presents Cheshire, a lightweight and modular 64-bit Linux-capable host platform designed for the seamless plug-in of domain-specific accelerators. It features a unique low-pin-count DRAM interface, a last-level cache configurable as scratchpad memory, and a DMA engine enabling efficient data movement to or from accelerators or DRAM. It also provides numerous optional IO peripherals including UART, SPI, I2C, VGA, and GPIOs. Cheshire’s synthesizable RTL description, comprising all of its peripherals and its fully digital DRAM interface, is available free and open-source. We implemented and fabricated Cheshire as a silicon demonstrator called Neo in TSMC’s 65nm CMOS technology. At 1.2V, Neo achieves clock frequencies of up to 325 MHz while not exceeding 300 mW in total power on data-intensive computational workloads. Its RPC DRAM interface consumes only 250 pJ/B and incurs only 3.5 kGE in area for its PHY while attaining a peak transfer rate of 750 MB/s at 200 MHz.

Computationally intensive algorithms such as Deep Neural Networks (DNNs) are becoming killer applications for edge devices. Porting heavily data-parallel algorithms on resource-constrained and battery-powered devices while retaining the flexibility granted by instruction processor-based architectures poses several challenges related to memory footprint, computational throughput, and energy efficiency. Low-bitwidth and mixed-precision arithmetic have been proven to be valid strategies for tackling these problems. We present Dustin, a fully programmable compute cluster integrating 16 RISC-V cores capable of 2- to 32-bit arithmetic and all possible mixed-precision combinations. In addition to a conventional Multiple-Instruction Multiple-Data (MIMD) processing paradigm, Dustin introduces a Vector Lockstep Execution Mode (VLEM) to minimize power consumption in highly data-parallel kernels. In VLEM, a single leader core fetches instructions and broadcasts them to the 15 follower cores. Clock gating Instruction Fetch (IF) stages and private caches of the follower cores leads to 38% power reduction. The cluster, implemented in 65 nm CMOS technology, achieves a peak performance of 58 GOPS and a peak efficiency of 1.15 TOPS/W.

High performance and energy efficiency are critical requirements for Internet of Things (IoT) end-nodes. Exploiting tightly coupled clusters of programmable processors (CMPs) has recently emerged as a suitable solution to address this challenge. One of the main bottlenecks limiting the performance and energy efficiency of these systems is the instruction cache architecture due to its criticality in terms of timing (i.e., maximum operating frequency), bandwidth, and power. We propose a hierarchical instruction cache tailored to ultralow-power (ULP) tightly coupled processor clusters where a relatively large cache (L1.5) is shared by L1 private (PR) caches through a two-cycle latency interconnect. To address the performance loss caused by the L1 capacity misses, we introduce a next-line prefetcher with cache probe filtering (CPF) from L1 to L1.5. We optimize the core instruction fetch (IF) stage by removing the critical core-to-L1 combinational path. We present a detailed comparison of instruction cache architectures’ performance and energy efficiency for parallel ULP (PULP) clusters. Focusing on the implementation, our two-level instruction cache provides better scalability than existing shared caches, delivering up to 20% higher operating frequency. On average, the proposed two-level cache improves maximum performance by up to 17% compared to the state-of-the-art while delivering similar energy efficiency for most relevant applications.

Microarchitectural timing channels enable unwanted information flow across security boundaries, violating fundamental security assumptions. They leverage timing variations of several state-holding microarchitectural components and have been demonstrated across instruction set architectures and hardware implementations. Analogously to memory protection, (Ge et al. 2019) have proposed time protection for preventing information leakage via timing channels. They also showed that time protection calls for hardware support. This work leverages the open and extensible RISC-V instruction set architecture (ISA) to introduce the temporal fence instruction fence.t, which provides the required mechanisms by clearing vulnerable microarchitectural state and guaranteeing a history-independent context-switch latency. We propose and discuss three different implementations of fence.t and implement them on an experimental version of the seL4 microkernel (Klein et al. 2014) and CVA6, an open-source, in-order, application class, 64-bit RISC-V core (Zaruba and Benini 2019). We find that a complete, systematic, ISA-supported erasure of all non-architectural core components is the most effective implementation while featuring a low implementation effort, a minimal performance overhead of less than 1%, and negligible hardware costs.

Neural Architecture Search (NAS) is quickly becoming the go-to approach to optimize the structure of Deep Learning (DL) models for complex tasks such as Image Classification or Object Detection. However, many other relevant applications of DL, especially at the edge, are based on time-series processing and require models with unique features, for which NAS is less explored. This work focuses in particular on Temporal Convolutional Networks (TCNs), a convolutional model for time-series processing that has recently emerged as a promising alternative to more complex recurrent architectures. We propose the first NAS tool that explicitly targets the optimization of the most peculiar architectural parameters of TCNs, namely dilation, receptive-field and number of features in each layer. The proposed approach searches for networks that offer good trade-offs between accuracy and number of parameters/operations, enabling an efficient deployment on embedded platforms. Moreover, its fundamental feature is that of being lightweight in terms of search complexity, making it usable even with limited hardware resources. We test the proposed NAS on four real-world, edge-relevant tasks, involving audio and bio-signals: (i) PPG-based Heart-Rate Monitoring, (ii) ECG-based Arrythmia Detection, (iii) sEMG-based Hand-Gesture Recognition, and (iv) Keyword Spotting. Results show that, starting from a single seed network, our method is capable of obtaining a rich collection of Pareto optimal architectures, among which we obtain models with the same accuracy as the seed, and 15.9-152× fewer parameters. Moreover, the NAS finds solutions that Pareto-dominate state-of-the-art hand-tuned models for 3 out of the 4 benchmarks, and are Pareto-optimal on the fourth (sEMG). Compared to three state-of-the-art NAS tools, ProxylessNAS, MorphNet and FBNetV2, our method explores a larger search space for TCNs (up to $10^{12} \times$1012×) and obtains superior solutions, while requiring low GPU memory and search time. We deploy our NAS outputs on two distinct edge devices, the multicore GreenWaves Technology GAP8 IoT processor and the single-core STMicroelectronics STM32H7 microcontroller. With respect to the state-of-the-art hand-tuned models, we reduce latency and energy of up to 5.5× and 3.8× on the two targets respectively, without any accuracy loss.

Disentangling the attributes of a sensory signal is central to sensory perception and cognition and hence is a critical task for future artificial intelligence systems. Here we present a compute engine capable of efficiently factorizing high-dimensional holographic representations of combinations of such attributes, by exploiting the computation-in-superposition capability of brain-inspired hyperdimensional computing, and the intrinsic stochasticity associated with analogue in-memory computing based on nanoscale memristive devices. Such an iterative in-memory factorizer is shown to solve at least five orders of magnitude larger problems that cannot be solved otherwise, as well as substantially lowering the computational time and space complexity. We present a large-scale experimental demonstration of the factorizer by employing two in-memory compute chips based on phase-change memristive devices. The dominant matrix–vector multiplication operations take a constant time, irrespective of the size of the matrix, thus reducing the computational time complexity to merely the number of iterations. Moreover, we experimentally demonstrate the ability to reliably and efficiently factorize visual perceptual representations.

This article presents a low-power, self-sustainable, and modular wireless sensor node for aerod- ynamic and acoustic measurements on wind turbines and other industrial structures. It includes 40 high-accuracy barometers, ten microphones, and five 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 (BLE) tuned for long range and high throu- ghput 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 kb\ $\ \text{s}$ over 438 m. The power consumption while collecting and streaming all the measured data is 120 mW, enabling self-sustainability and long-term in situ monitoring with a 111 cm 2 photovoltaic panel.

Smart sensors have become pervasive in railway transportation applications, particularly in Europe where digital technologies are increasingly being applied to the railway signaling field. In the future, safety-critical track-side train detection equipment, such as track circuits and axle counters, will be eliminated in favor of an accurate position estimate supplied by the train. However, the best approach to calculate an accurate position estimate remains an open research question, especially due to the high availability and reliability required. This article describes two static experiments performed with a global navigation satellite system (GNSS) module, which demonstrate that the real-world accuracy achievable with GNSS and real time kinematic (RTK) alone is not sufficient for safety-critical applications, meaning further complementary sensors are required. Furthermore, a custom sensor node containing a GNSS module with RTK and an inertial measurement unit (IMU) has been used to acquire several data sets from an operating passenger train during dynamic tests on a railway line between Formigine and Modena, in Emilia-Romagna, Italy. These labeled GNSS and IMU data have been made freely available to the scientific community. The positioning accuracy of the GNSS and RTK measurements is evaluated, providing an in-depth study of the localization error and satellite coverage on the entire route. We demonstrate, with experimental evaluation, that centimeter accuracy (1.9 ± 0.8) cm) is achievable under favorable static conditions, while accuracy can deteriorate to 8 m with RTK in urban scenarios with many reflections and poor sky view, worse than with GNSS alone. Under controlled conditions we show that shielding the GNSS receiver without RTK with a grounded metal plate causes a reduction in accuracy from 0.8 ± 0.04 to 3.7 ± 0.55 m in the least and most shielded case respectively. Our dynamic tests on a train show that although at least meter-level accuracy (1.08 ± 1.3 m) is achievable with GNSS and RTK alone under dynamic conditions, a sensor fusion approach is necessary to accurately localize trains when GNSS conditions are poor or GNSS is unavailable.

Tiny machine learning (TinyML) applications impose µJ/inference constraints, with a maximum power consumption of tens of megawatt. It is extremely challenging to meet these requirements at a reasonable accuracy level. This work addresses the challenge with a flexible, fully digital ternary neural network (TNN) accelerator in a reduced instruction set computer-five (RISC-V)-based System-on-Chip (SoC). Besides supporting ternary convolutional neural networks, we introduce extensions to the accelerator design that enable the processing of time-dilated temporal convolutional neural networks (TCNs). The design achieves 5.5-µJ/inference, 12.2 mW, 8,000 inferences/s at 0.5 V for a dynamic vision sensor (DVS)-based TCN and an accuracy of 94.5%, and 2.72-µJ/inference, 12.2 mW, 3,200 inferences/s at 0.5 V for a nontrivial 9-layer, 96 channels-per-layer convolutional network with CIFAR-10 accuracy of 86%. The peak energy efficiency is 1,036 TOp/s/W, outperforming the state-of-the-art silicon-proven TinyML quantized accelerators by 1.67× while achieving competitive accuracy.

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 $29\times $ , 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 $4.3\times $ on a 10-class dataset while providing a compelling accuracy-energy trade-off including a 2% accuracy drop for a $71\times $ energy reduction. [ABSTRACT FROM AUTHOR]

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 $11.5\times $ performance and $9.5\times $ 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.

Remanufacturing of end-of-life products and parts is seen as a solution in the transition towards a circular economy. There are many proposed tools and methods to facilitate the application of this circular strategy, however, among them, there is a lack of support tools for practitioners that include multiple perspectives related to the value chain and circular economy. In fact, remanufacturing strategy, economic and environmental trade-offs, and circularity indicators are rarely integrated within one framework. In this paper, an approach is presented taking advantage of the state-of-the-art research on green profit model and circularity indicators; in other words, these tools are used together to unlock the circular potential in manufacturing practice. In this way, typical problems of production planning and control in remanufacturing processes are interconnected with the goals of sustainable development, also considering product design and end-of-life strategy choices. The presented framework represents a promising support to be used in industrial practice. A case study based on PV panel infrastructure allows a better comprehension of the research outputs and assesses the validity of the support provided by the framework in the deployment of circular economy strategies.

In the last few years, research and development on Deep Learning models & 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 (Pellegrini et al., 2020) enable online, serverless adaptation in principle, but so far they 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 $4\times $ 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, together with data tiling strategies to fully exploit its memory hierarchy, while maximizing efficiency. 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 on average $65 \times $ faster than a low-power STM32 L4 microcontroller, being $37\times $ more energy efficient – enough for a lifetime of 535h when learning a new mini-batch of data once every minute.

The evolution of energy-efficient ultra-low-power (ULP) parallel processors and the diffusion of convolutional neural networks (CNNs) are fueling the advent of autonomous driving nano-sized unmanned aerial vehicles (UAVs). These sub-10cm robotic platforms are envisioned as next-generation ubiquitous smart-sensors and unobtrusive robotic-helpers. However, the limited computational/memory resources available aboard nano-UAVs introduce the challenge of minimizing and optimizing vision-based CNNs – which to date require error-prone, labor-intensive iterative development flows. This work explores methodologies and software tools to streamline and automate all the deployment of vision-based CNN navigation on a ULP multicore system-on-chip acting as a mission computer on a Crazyflie 2.1 nano-UAV. We focus on the deployment of PULP-Dronet (Palossi et al., 2019), a state-of-the-art CNN for autonomous navigation of nano-UAVs, from the initial training to the final closed-loop evaluation. Compared to the original hand-crafted CNN, our results show a $2\times $ reduction of memory footprint and a speedup of $1.6\times $ in inference time while guaranteeing the same prediction accuracy and significantly improving the behavior in the field, achieving: i) obstacle avoidance with a peak braking-speed of 1.65m/s and improving the speed/braking-space ratio of the baseline, ii) free flight in a familiar environment up to 1.96m/s (0.5m/s for the baseline), and iii) lane following on a path featuring a 90deg turn – all while using for computation less than 1.6% of the drone’s power budget. To foster new applications and future research, we open-source all the software design in a ready-to-run project compatible with the Crazyflie 2.1.

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 $\mu \text{W}$ 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 Electromyography (EMG) hand 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-algorithm for the task of ball bearing fault detection. [ABSTRACT FROM AUTHOR]

Successful motor-imagery brain–computer interface (MI-BCI) algorithms either extract a large number of handcrafted features and train a classifier, or combine feature extraction and classification within deep convolutional neural networks (CNNs). Both approaches typically result in a set of real-valued weights, that pose challenges when targeting real-time execution on tightly resource-constrained devices. We propose methods for each of these approaches that allow transforming real-valued weights to binary numbers for efficient inference. Our first method, based on sparse bipolar random projection, projects a large number of real-valued Riemannian covariance features to a binary space, where a linear SVM classifier can be learned with binary weights too. By tuning the dimension of the binary embedding, we achieve almost the same accuracy in 4-class MI (≤1.27% lower) compared to models with float16 weights, yet delivering a more compact model with simpler operations to execute. Second, we propose to use memory-augmented neural networks (MANNs) for MI-BCI such that the augmented memory is binarized. Our method replaces the fully connected layer of CNNs with a binary augmented memory using bipolar random projection, or learned projection. Our experimental results on EEGNet, an already compact CNN for MI-BCI, show that it can be compressed by $1.28\times $ at iso-accuracy using the random projection. On the other hand, using the learned projection provides 3.89% higher accuracy but increases the memory size by $28.10\times $ .

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 0.5V without 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 than 99% of the nominal accuracy (with a bit error rate ~1/1000). In this operating point, our prototype performs 4Gop/s (15.4 Inference/s on the CIFAR-10 dataset) by computing up to 13 binary 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. [ABSTRACT FROM AUTHOR]

Convolutional Neural Networks (CNNs) 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 multi-layer (deep) network 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 modeling. These structures, commonly referred to as Temporal Convolutional Networks (TCNs), represent an extremely promising alternative to recurrent architectures, commonly used across a broad range of sequence modeling tasks. 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 a power efficiency of 33,8 GOPS/s/W on an Ultrascale+ ZU3EG (up to $10\times$ speedup and $3\times$ power efficiency improvement with respect to pure software implementation).

In the wake of the success of convolutional neural networks in image classification, object recognition, speech recognition, etc., the demand for deploying these compute-intensive ML models on embedded and mobile systems with tight power and energy constraints at low cost, as well as for boosting throughput in data centers, is growing rapidly. This has sparked a surge of research into specialized hardware accelerators. Their performance is typically limited by I/O bandwidth, power consumption is dominated by I/O transfers to off-chip memory, and on-chip memories occupy a large part of the silicon area. We introduce and evaluate a novel, hardware-friendly, and lossless compression scheme for the feature maps present within convolutional neural networks. We present hardware architectures and synthesis results for the compressor and decompressor in 65 nm. With a throughput of one 8-bit word/cycle at 600 MHz, they fit into 2.8 kGE and 3.0 kGE of silicon area, respectively- together the size of less than seven 8-bit multiply-add units at the same throughput. We show that an average compression ratio of 5.1× for AlexNet, 4× for VGG-16, 2.4× for ResNet-34 and 2.2× for MobileNetV2 can be achieved-a gain of 45-70% over existing methods. Our approach also works effectively for various number formats, has a low frame-to-frame variance on the compression ratio, and achieves compression factors for gradient map compression during training that are even better than for inference.

Deep neural networks have achieved impressive results in computer vision and machine learning. Unfortunately, state-of-the-art networks are extremely compute and memory intensive, which makes them unsuitable for mW-devices such as IoT end-nodes. Aggressive quantization of these networks dramatically reduces the computation and memory footprint. Binary-weight neural networks (BWNs) follow this trend, pushing weight quantization to the limit. Hardware accelerators for BWNs presented up to now have focused on core efficiency, disregarding I/O bandwidth, and system-level efficiency that are crucial for the deployment of accelerators in ultra-low power devices. We present Hyperdrive: a BWN accelerator dramatically reducing the I/O bandwidth exploiting a novel binary-weight streaming approach, which can be used for an arbitrarily sized convolutional neural network architecture and input resolution by exploiting the natural scalability of the compute units both at chip-level and system-level by arranging Hyperdrive chips systolically in a 2D mesh while processing the entire feature map together in parallel. Hyperdrive achieves 4.3 TOp/s/W system-level efficiency (i.e., including I/Os)— $3.1\times $ higher than state-of-the-art BWN accelerators, even if its core uses resource-intensive FP16 arithmetic for increased robustness.

Broadband current sensors are key components in numerous applications, including power conversion, motor control, and smart-metering. We present a compressive sensing (CS) current sensor system-on-chip (SoC) designed and fabricated in STM 0.16 $\mu \text{m}$ Bipolar-CMOS-DMOS technology. The SoC is capable of measuring currents with amplitudes of up to 10 A peak with a sensing bandwidth of 1 MHz. Two broadband current sensing cores, each consisting of a Hall-effect probe, an AFE, and a 2 MS/s 9-bit ADC, are monolithically integrated together with a digital multi-mode compressive sensing encoder (DCSE) for data-rate reduction. We focus on the evaluation of CS as data compression codec for current sensing applications and on the details of the DCSE architecture designed for general purpose use. We introduce multi-block decoding that is a decoding modality to improve the reconstruction quality of “off-grid sparse” signals commonly occurring in practice. Moreover, we provide measurements of both the compression performance and power consumption of the SoC employed in two exemplary real-world applications; namely, sparse current sensing, and electrical appliance detection for non-intrusive load monitoring based on compressive measurements. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (CNNs) obtain outstanding results in tasks that require human-level understanding of data, like image or speech recognition. However, their computational load is significant, motivating the development of CNN-specialized accelerators. This work presents NEURAghe, a flexible and efficient hardware/software solution for the acceleration of CNNs on Zynq SoCs. NEURAgheleverages the synergistic usage of Zynq ARM cores and of a powerful and flexible Convolution-Specific Processor deployed on the reconfigurable logic. The Convolution-Specific Processor embeds both a convolution engine and a programmable soft core, releasing the ARM processors from most of the supervision duties and allowing the accelerator to be controlled by software at an ultra-fine granularity. This methodology opens the way for cooperative heterogeneous computing: While the accelerator takes care of the bulk of the CNN workload, the ARM cores can seamlessly execute hard-to-accelerate parts of the computational graph, taking advantage of the NEON vector engines to further speed up computation. Through the companion NeuDNN SW stack, NEURAghesupports end-to-end CNN-based classification with a peak performance of 169GOps/s, and an energy efficiency of 17GOps/W. Thanks to our heterogeneous computing model, our platform improves upon the state-of-the-art, achieving a frame rate of 5.5 frames per second (fps) on the end-to-end execution of VGG-16 and 6.6fps on ResNet-18.

We report an always-on event-driven asynchronous wake-up circuit with trainable pattern recognition capabilities to duty-cycle power-constrained Internet-of-Things (IoT) sensor nodes. The wake-up circuit is based on a level-crossing analog-to-digital converter (LC-ADC) employed as a feature-extraction block with automatic activity-sampling rate scaling behavior. A novel asynchronous digital logic classifier for sequential pattern recognition is presented. It is driven by the LC-ADC activity and trained to minimize classification errors due to falsely detected events. As proof-of-concept, a prototype of the wake-up circuit is fabricated in 130nm CMOS technology within 0.054 mm2 of active area, covering up to 2.6 kHz of input signal bandwidth. The prototype has been first validated by interfacing it with a commercial accelerometer to classify hand gestures in real-time, reaching 81% of accuracy with only 2.2 $\mu \text{W}$ at 1-V supply. To highlight the flexibility of the design, a second application, detecting pathologic electrocardiogram beats is also discussed.

Duty-cycled low-rate Internet-of-things (IoT) sensors are employed in diverse applications, requiring configurable and energy-efficient on-chip and on-demand clock synthesis. We present an all-digital frequency-locked loop (AD-FLL) capable of generating an accurate clock selectively in stand-alone operation or locked to a 32kHz reference. We report measurement results of two prototypes in 65nm and 28nm CMOS offering a configurable clock multiplication factor of up to 32 786, resulting in a wide tuning-range from a few MHz to 2.4GHz and 1.6GHz, respectively. The challenges of slow start-up and deterministic jitter are addressed by a fast hybrid-mode start-up procedure and by various jitter reduction modes. We also introduce the concept of Transient Clocking that leverages the capabilities of the proposed AD-FLL to make a system operational after cold-start or wake-up before the supply voltage has stabilized. We study two application examples that highlight the versatility of the concept in IoT applications and show its potential to amortize the time and energy cost of typical system start-up tasks, like state-restoration or wake-up event classification. [ABSTRACT FROM PUBLISHER]

Near-sensor data analytics is a promising direction for internet-of-things endpoints, as it minimizes energy spent on communication and reduces network load - but it also poses security concerns, as valuable data are stored or sent over the network at various stages of the analytics pipeline. Using encryption to protect sensitive data at the boundary of the on-chip analytics engine is a way to address data security issues. To cope with the combined workload of analytics and encryption in a tight power envelope, we propose Fulmine, a system-on-chip (SoC) based on a tightly-coupled multi-core cluster augmented with specialized blocks for compute-intensive data processing and encryption functions, supporting software programmability for regular computing tasks. The Fulmine SoC, fabricated in 65-nm technology, consumes less than 20mW on average at 0.8V achieving an efficiency of up to 70pJ/B in encryption, 50pJ/px in convolution, or up to 25MIPS/mW in software. As a strong argument for real-life flexible application of our platform, we show experimental results for three secure analytics use cases: secure autonomous aerial surveillance with a state-of-the-art deep convolutional neural network (CNN) consuming 3.16pJ per equivalent reduced instruction set computer operation, local CNN-based face detection with secured remote recognition in 5.74pJ/op, and seizure detection with encrypted data collection from electroencephalogram within 12.7pJ/op. [ABSTRACT FROM PUBLISHER]

This work presents the research findings of the Multithermal ERC-Advanced project, in terms of multi-scale thermal control of complex large scale computing platforms such as High Performance Computing Systems, and datacenters. In this respect, the challenges and opportunities concerning thermal control of computing systems are discussed, along with the proposed innovative solutions. Control and management problem is divided at different hierarchical, and scale, levels (compute node and system). Then, given the control knobs and sensors available at each level, advanced control and system identification techniques are proposed, to achieve a holistic thermal management of complex computing platforms, with benefits for thermal stability guarantees and energy consumption optimization. Such features are of utmost importance for advancements in next generation large scale computing solutions.

Data produced by wearable sensors is key in contexts such as performance enhancement and training help for sports and fitness, continuous monitoring for aging people and for chronic disease management, and in gaming and entertainment. Unfortunately, wearable devices currently in the market are either incapable of complex functionality or severely impaired by short battery lifetime. In this work, we present a smartwatch platform based on an ultralow-power (ULP) heterogeneous system composed of a TI MSP430 microcontroller, the PULP programmable parallel accelerator, and a set of ULP sensors, including a camera. The embedded PULP accelerator enables state-of-the-art context classification based on convolutional neural networks to be applied within a sub-10-mW system power envelope. Our methodology enables to reach high accuracy in context classification over five classes (up to 84%, with three classes over five reaching more than 90% accuracy), while consuming 2.2 mJ per classification, or an ultralow energy consumption of less than 91 $\mu$ J per classification with an accuracy of 64%–3.2 $\times$ better than chance. Our results suggest that the proposed heterogeneous platform can provide up to 500$\times$ speedup with respect to the MSP430 within a similar power envelope, which would enable complex computer vision algorithms to be executed in highly power-constrained scenarios.

Embedded memory remains a major bottleneck in current integrated circuit design in terms of silicon area, power dissipation, and performance; however, static random access memories (SRAMs) are almost exclusively supplied by a small number of vendors through memory generators, targeted at rather generic design specifications. As an alternative, standard cell memories (SCMs) can be defined, synthesized, and placed and routed as an integral part of a given digital system, providing complete design flexibility, good energy efficiency, low-voltage operation, and even area efficiency for small memory blocks. Yet implementing an SCM block with a standard digital flow often fails to exploit the distinct and regular structure of such an array, leaving room for optimization. In this article, we present a design methodology for optimizing the physical implementation of SCM macros as part of the standard design flow. This methodology introduces controlled placement, leading to a structured, noncongested layout with close to 100 placement utilization, resulting in a smaller silicon footprint, reduced wire length, and lower power consumption compared to SCMs without controlled placement. This methodology is demonstrated on SCM macros of various sizes and aspect ratios in a state-of-the-art 28nm fully depleted silicon-on-insulator technology, and compared with equivalent macros designed with the noncontrolled, standard flow, as well as with foundry-supplied SRAM macros. The controlled SCMs provide an average 25 reduction in area as compared to noncontrolled implementations while achieving a smaller size than SRAM macros of up to 1Kbyte. Power and performance comparisons of controlled SCM blocks of a commonly found 256 × 32 (1 Kbyte) memory with foundry-provided SRAMs show greater than 65 and 10 reduction in read and write power, respectively, while providing faster access than their SRAM counterparts, despite being of an aspect ratio that is typically unfavorable for SCMs. In addition, the SCM blocks function correctly with a supply voltage as low as 0.3V, well below the lower limit of even the SRAM macros optimized for low-voltage operation. The controlled placement methodology is applied within a full-chip physical implementation flow of an OpenRISC-based test chip, providing more than 50 power reduction compared to equivalently sized compiled SRAMs under a benchmark application.