Search

Quantization is widely employed in both cloud and edge systems to reduce the memory occupation, latency, and energy consumption of deep neural networks. In particular, mixed-precision quantization, i.e., the use of different bit-widths for different portions of the network, has been shown to provide excellent efficiency gains with limited accuracy drops, especially with optimized bit-width assignments determined by automated Neural Architecture Search (NAS) tools. State-of-the-art mixed-precision works layer-wise, i.e., it uses different bit-widths for the weights and activations tensors of each network layer. In this work, we widen the search space, proposing a novel NAS that selects the bit-width of each weight tensor channel independently. This gives the tool the additional flexibility of assigning a higher precision only to the weights associated with the most informative features. Testing on the MLPerf Tiny benchmark suite, we obtain a rich collection of Pareto-optimal models in the accuracy vs model size and accuracy vs energy spaces. When deployed on the MPIC RISC-V edge processor, our networks reduce the memory and energy for inference by up to 63% and 27% respectively compared to a layer-wise approach, for the same accuracy.

This paper presents a technique, alternative to performance-drivensynthesis, that allows to drastically increase the averagethroughput of combinational logic blocks by transforming fixed-latencyunits into variable-latency ones that run with a fasterclock cycle.The transformation is fully automatic and can beused in conjunction with traditional design techniques, such aspipelining, to improve the overall performance of speed-criticalsystems.Results, obtained on a large set of benchmark circuits,are very promising.

In modern data centers, storage system failures are major contributors to downtimes and maintenance costs. Predicting these failures by collecting measurements from disks and analyzing them with machine learning techniques can effectively reduce their impact, enabling timely maintenance. While there is a vast literature on this subject, most approaches attempt to predict hard disk failures using either classic machine learning solutions, such as Random Forests (RFs) or deep Recurrent Neural Networks (RNNs). In this work, we address hard disk failure prediction using Temporal Convolutional Networks (TCNs), a novel type of deep neural network for time series analysis. Using a real-world dataset, we show that TCNs outperform both RFs and RNNs. Specifically, we can improve the Fault Detection Rate (FDR) of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\approx $$\end{document}≈7.5% (FDR = 89.1%) compared to the state-of-the-art, while simultaneously reducing the False Alarm Rate (FAR = 0.052%). Moreover, we explore the network architecture design space showing that TCNs are consistently superior to RNNs for a given model size and complexity and that even relatively small TCNs can reach satisfactory performance. All the codes to reproduce the results presented in this paper are available at https://github.com/ABurrello/tcn-hard-disk-failure-prediction.

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., Published in: 2021 IFIP/IEEE 29th International Conference on Very Large Scale Integration (VLSI-SoC), 2021

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.

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.

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 $103 \times $ lower latency and $20.3 \times $ lower energy than the Cube-AI toolkit executed on the STM32L4 and from $2.9 \times $ to $26.6 \times $ lower energy compared to commercial closed-source and academic open-source approaches on the same hardware target.

In recent years, the introduction and exploitation of innovative information technologies in industrial contexts have led to the continuous growth of digital shop floor environments. The new Industry 4.0 model allows smart factories to become very advanced IT industries, generating an ever-increasing amount of valuable data. As a consequence, the necessity of powerful and reliable software architectures is becoming prominent along with data-driven methodologies to extract useful and hidden knowledge supporting the decision-making process. This article discusses the latest software technologies needed to collect, manage, and elaborate all data generated through innovative Internet-of-Things (IoT) architectures deployed over the production line, with the aim of extracting useful knowledge for the orchestration of high-level control services that can generate added business value. This survey covers the entire data life cycle in manufacturing environments, discussing key functional and methodological aspects along with a rich and properly classified set of technologies and tools, useful to add intelligence to data-driven services. Therefore, it serves both as a first guided step toward the rich landscape of the literature for readers approaching this field and as a global yet detailed overview of the current state of the art in the Industry 4.0 domain for experts. As a case study, we discuss, in detail, the deployment of the proposed solutions for two research project demonstrators, showing their ability to mitigate manufacturing line interruptions and reduce the corresponding impacts and costs.

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., Comment: Published in: 2020 27th IEEE International Conference on Electronics, Circuits and Systems (ICECS)

Proximity beacons are small, low-power devices capable of transmitting information at a limited distance via Bluetooth low energy protocol. These beacons are typically used to broadcast small amounts of location-dependent data (e.g., advertisements) or to detect nearby objects. However, researchers have shown that beacons can also be used for indoor localization converting the received signal strength indication (RSSI) to distance information. In this work, we study the effectiveness of proximity beacons for accurately locating objects within a manufacturing plant by performing extensive experiments in a real industrial environment. To this purpose, we compare localization algorithms based either on trilateration or environment fingerprinting combined with a machine-learning based regressor (k-nearest neighbors, support-vector machines, or multi-layer perceptron). Each algorithm is analyzed in two different types of industrial environments. For each environment, various configurations are explored, where a configuration is characterized by the number of beacons per square meter and the density of fingerprint points. In addition, the fingerprinting approach is based on a preliminary site characterization, it may lead to location errors in the presence of environment variations (e.g., movements of large objects). For this reason, the robustness of fingerprinting algorithms against such variations is also assessed. Our results show that fingerprint solutions outperform trilateration, showing also a good resilience to environmental variations. Given the similar error obtained by all three fingerprint approaches, we conclude that k-NN is the preferable algorithm due to its simple deployment and low number of hyper-parameters.

The capability of accurately estimating an upper bound of the maximum current drawn by a digital macroblock from the ground or power supply line constitutes a major asset of automatic power-gating flows. In fact, the maximum current information is essential to properly size the sleep transistor in such a way that speed degradation and signal integrity violations are avoided. Loose upper bounds can be determined with a reasonable computational cost, but they lead to oversized sleep transistors. On the other hand, exact computation of the maximum drawn current is an NP-hard problem, even when conservative simplifying assumptions are made on gate-level current profiles. In this paper, we present a scalable algorithm for tightening upper bound computation, with a controlled and tunable computational cost. The algorithm exploits state-of-the-art commercial timing analysis engines, and it is tightly integrated into an industrial power-gating flow for leakage power reduction. The results we have obtained on large circuits demonstrate the scalability and effectiveness of our estimation approach.

Leakage power has become a serious concern in nanometer CMOS technologies, and power-gating has shown to offer a viable solution to the problem with a small penalty in performance. This paper focuses on leakage power reduction through automatic insertion of sleep transistors for power-gating. In particular, we propose a novel, layout-aware methodology that facilitates sleep transistor insertion and virtual-ground routing on row-based layouts. We also introduce a clustering algorithm that is able to handle simultaneously timing and area constraints, and we extend it to the case of multi- Vt sleep transistors to increase leakage savings. The results we have obtained on a set of benchmark circuits show that the leakage savings we can achieve are, by far, superior to those obtained using existing power-gating solutions and with much tighter timing and area constraints.

Power-gating is one of the most promising and widely adopted solutions for controlling sub-threshold leakage power in nanometer circuits. Although single-cycle power-mode transition reduces wake-up latency, it develops large discharge current spikes, thereby causing IR-drop and inductive ground bounce for the neighboring circuit blocks, which can suffer from power plane integrity degradation. We propose a new reactivation solution that helps in controlling power supply fluctuations and achieving minimum reactivation times. Our structure limits the turn-on current below a given threshold through a sequential activation of the sleep transistors (STs), which are connected in parallel and sized using a novel optimal sizing algorithm. We also introduce a distributed physical implementation, which allows minimum layout disruption after ST insertion and minimizes routing congestion.