This paper introduces custom RISC-V instructions that merge address indexation, memory access and increment operations into a single load/store instruction to reduce the number of fetched instructions, thus reducing memory accesses. Implemented on the AsterisC CPU core in 18 nm FD-SOI technology, these instructions aim to mitigate the memory wall issue by lowering the number of accesses to instruction memory. Experimental results demonstrate reductions in terms of energy consumption ranging from 8% to 24% and latency improvements up to 30%. This proposal has been applied to a use case: coupling the AsteRISC CPU with a hardware accelerator for edge AI, the proposed custom instructions result in a 19% reduction in energy consumption.

Chlorite minerals, mainly in the form of clay coats, play a critical role in determining the reservoir quality of siliciclastic rocks. They can positively influence reservoir quality by preserving porosity during deep burial, but they can also play a negative role by reducing permeability through pore filling. The main aim of this research is to determine the optimal conditions for chlorite growth in sedimentary basins. This study investigates the Lower Cretaceous turbidite sandstone of the Agat Formation in the North Sea. We used a source-to-sink approach to investigate the impact of sediment source composition, chemical weathering and depositional environment on chlorite formation. Understanding the interplay between these processes can help refine exploration and exploitation strategies, optimise hydrocarbon recovery, and reduce exploration risks. Representative samples from two hydrocarbon fields (the Duva and Agat fields) were investigated using petrography, geochemistry, heavy mineral identification and quantification, and U-Pb geochronology of detrital zircons. Our results show a strong heterogeneity in the sediment provenance between the two turbidite systems. In the Duva field, the sandstone is derived from a mixture of mafic and felsic sources, producing Fe-rich sediments. Intense chemical weathering generates fine fraction materials rich in kaolinite, vermiculite, and hydroxy-interlayered clays, which are transported into shallow marine settings. Subsequent interaction with seawater results in the formation of glauconitic materials, Fe-illite, and phosphatic concretions. These Fe-rich materials are remobilised into deep marine settings, providing precursors for the development of authigenic Fe-clays such as berthierine and chlorite. Conversely, in the Agat field, the sandstone is predominantly sourced from felsic rocks that underwent low chemical weathering, producing sediment rich in quartz and feldspar with a low amount of clays. With few Fe-rich materials transported into the basin, the development of chlorite in the Agat field was less pervasive. Basin configuration and depositional environment exerted additional control on chlorite distribution. In the confined turbidite system (e.g. Duva field), chlorite is typically found as coating, whereas in less confined turbidite systems (e.g. Agat field) chlorite shows complex distribution related to depositional environment and dewatering processes. Our findings demonstrate the importance of considering the entire sediment routing system, from source to sink, when predicting chlorite occurrence and its impact on reservoir quality in deep marine settings. This integrated approach can guide exploration and development efforts in deepwater clastic reservoirs. The influence of sediment provenance and depositional setting on chlorite formation in the Agat Formation.image

In this work we study the sizing and feasibility of a processing element for the calculation of the attention mechanism used in the Transformer models, using floating point representation. We show that the Flash Attention implementation is needed to enable circuit integration at the edge. Different floating point formats, both standard and non standard representations, were explored and synthesized in 18nm FD-SOI CMOS to compare their area and leakage power. The quality of the results is assessed against a pytorch BERT reference model. The best representation found is a non-standard 12-bit floating point architecture that computes a single block of Flash Attention. Once compared to a RISC- V CPU synthesized with the same technology, shows 10 4 less energy consumption and 10 3 less latency, whilst providing a result with low mean error.

Considering the power-hungry nature of speech processing, a keyword spotting (KWS) unit, used to detect multiple spoken words, is often integrated as a front-end layer. KWS systems are always active, and thus, it is extremely important to optimize the devoted power budget. In this context, this article presents a programmable low-power event-driven real-time spectrogram extraction unit tested for the KWS application. This chip, fabricated in 28-nm FD-SOI CMOS technology, has been combined with a software-defined convolutional neural network to demonstrate the recognition of 11 audio classes (ten keywords $+$ background $+$ unknown) with an accuracy equal to 87.9% and an activity-dependent power consumption measured at 391.6 nW, for a 12-keyword/min average speech rate.

This paper presents an autonomous road traffic monitoring Edge AI SoC implemented in 18nm FD-SOI CMOS. The SoC includes a RISC-V CPU, a 128 Processing Elements (PE) Single Instruction Multiple Data (SIMD) Tensor Processing Unit (TPU) AI accelerator, with QVGA and control interfaces. It uses Adaptative-Body-Bias (ABB) process and temperature compensation. The circuit operates within 0.5 to $0.87 \mathrm{~V},-40$ to $140^{\circ} \mathrm{C}, 50$ to 500 MHz ranges. The TPU efficiency energy per inference and latency are $3.52 \mu \mathrm{~J}$ and 0.616 ms when running MobileNet v1 stripped by 2 layers, and $14.8 \mu \mathrm{~J}$ and 2.27 ms when running traffic monitoring application. For this latter application, the circuit total power is 6.38 mW at 15 fps 0.5 V supply.

This paper presents an autonomous road traffic monitoring Edge AI SoC implemented in 18nm FD-SOI CMOS. The SoC includes a RISC-V CPU, a 128 Processing Elements (PE) Single Instruction Multiple Data (SIMD) Tensor Processing Unit (TPU) AI accelerator, with QVGA and control interfaces. It uses Adaptative-Body-Bias (ABB) process and temperature compensation. The circuit operates within 0.5 to 0.87 V, -40 to 140 degrees C,50 to 500 MHz ranges. The TPU efficiency energy per inference and latency are 3.52 mu J and 0.616 ms when running MobileNet v1 stripped by 2 layers, and 14.8 mu J and 2.27 ms when running traffic monitoring application. For this latter application, the circuit total power is 6.38 mW at 15 fps 0.5V supply.

This paper presents the enhancements brought to AsteRISC, a highly modulable RISC-V core. We study the energy impact of the M extension and show that data gating, multiple-cycle constraints and frequency scaling help in reducing by a factor 2 the energy consumption of the resulting M-extended core. The core implementation, done in 18nm FD-SOI technology, has a lower dynamic power than Riscy and a commercialized reference core but also a better energy efficiency. Among the different implementations, the most efficient reached a 1.37 mu W energy consumption post Signoff with a 727 CoreMark score at 0.77V.

The preservation of reservoir quality in deeply buried sandstones largely depends on the presence of grain-coating clays that can inhibit quartz overgrowth and cementation. These grain-coating clays are the focus of today's hydrocarbon exploration as they enable new oil and gas discoveries in sedimentary basins. However, the origin and distribution of these clays in turbidites are poorly documented in scientific literature. Our study addresses this knowledge gap by investigating the mechanisms and parameters governing the clay coat development in turbidite sandstones, with a focus on the Agat Formation (Fm) in the Norwegian North Sea. Petrographic observations, scanning electron microscopy, and X-ray diffraction analyses were used to identify the origin of clay minerals and their distribution, as well as to understand the diagenetic evolution of the sandstone and its impact on reservoir quality. Based on the sedimentary facies description, the Agat Fm comprises four principal members deposited in various depositional environments, ranging from distal lobe fringe to amalgamated proximal lobes and weakly confined channels. Chlorite and kaolinite are the main authigenic clay minerals in these sandstones, associated with a variable amount of inherited detrital clays consisting in Fe-bearing kaolinite and mixed-layer minerals composed of illite and hydroxyl-interlayered clay minerals formed in acidic soil environments. Well-developed chlorite coats are associated with the highest reservoir quality (φ > 10%), while discontinuous chlorite coats are linked to extensive quartz cementation and poor reservoir quality (φ < 5%). Samples with abundant pore-filling chlorite also exhibit poor reservoir quality (φ < 7%). Proximal lobe deposits and weakly confined channels show well-developed chlorite coats, whereas discontinuous chlorite coats are mainly found in proximal lobe deposits associated with high velocity escaping fluids during dewatering. Pore-filling chlorites are typically present in distal lobe fringes and levee deposits. Chlorite in the Agat Fm is mainly derived from the replacement of inherited detrital clays via an intermediate berthierine precursor forming during early diagenesis. The dissolution of Fe-rich grains such as biotite and ferric detrital clays under reducing conditions governed the kinetics of berthierine/chlorite growth. The late diagenesis is characterized by extensive feldspar dissolution in a closed diagenetic system, leading to the precipitation of kaolinite and quartz cement. The presence of chlorite coatings around detrital grains preserved the porosity from extensive quartz cementation during deep burial, highlighting the role of chlorite in maintaining the quality of deeply buried turbidite reservoirs.

Considering the power-hungry nature of speech processing, a keyword spotting (KWS) unit, used to detect single spoken words, is often integrated as a front-end layer. KWS systems are always active, thus it is an extreme importance to optimize the devoted power budget. In this context, this paper presents a programmable low-power event-driven real-time spectrogram extraction unit tested for KWS application. This systemon-chip (SoC), fabricated in 28nm FD-SOI CMOS technology, has been combined with a software defined convolutional neural network (CNN) to demonstrate the recognition of 11 audio classes (10 keywords + background noise) with an accuracy equal to 90.08% and an activity-dependent power consumption measured at 291. 6nW, for a 12 keywords/minute average speech rate.

This work predicts the long-term on-orbit upset rate of 28-nm ultra-thin body and buried oxide (UTBB) fully depleted silicon on insulator (FD-SOI) technology. This prediction is based on investigating the effect of aging degradation mechanism on the heavy-ion single event upset (SEU) radiation sensitivity of 28-nm UTBB FD-SOI technology utilizing a custom designed test vehicle: ProArray. A comprehensive framework is presented to investigate the underlying mechanisms of the impact of the NBTI degradation mechanism on the heavy-ion radiation susceptibility. The framework can be used to predict the long-term on-orbit upset rate of 28-nm UTBB FD-SOI technology if there is no available experimental data.

It is well known that chlorite controls the reservoir quality of deeply buried sandstones by limiting secondary quartz overgrowths and by preserving the porosity network. However because of their depositional history, the impact of chlorite coatings in turbiditic sandstones, appears quite uncertain. The objective of this paper is to investigate the petrographic/mineralogical characteristics of chlorite coatings and their source material in the Early Cretaceous Agat Formation composed of massive turbiditic sandstones hosting two reservoirs of contrasting quality. This study is based on drill-hole to micrometer scale investigations using QEMSCAN mapping, scanning electron microscopy equipped with an energy dispersive spectrometer, X-ray diffraction, and FEG SEM observation of ultrathin sections.Chlorite of Ib polytype, inherited from berthierine precursor, is observed in both reservoirs. Its homogeneous composition suggests that the geochemical controls of chloritization are the same for the two reservoirs. Grain replacing chlorite, clay replacing chlorite and chlorite coatings are observed in both reservoirs but the grain coatings present different textural organizations. In the upper reservoir, the clay coating (5 µm width) is dominated by chlorite platelets whose growth seems to be controlled by geometrical selection. In the lower reservoir, the chlorite coating (of several tens of µm) differs by an additional inner part of the diagenetic-recrystallized clay coat.The preburial material consists of ferric illite (+/- glauconitic pelloids). The deformation of illitic grains suggests that illitization predated the transportation and/or incorporation in the turbidic flows. The illite shape (large particles with exfoliated micaceous cleavages or vermicular shapes) is considered to be inherited from an earlier supergene vermiculitization of phyllosilicates in saprolites. As ferric iron dominates in ferric illite compared to the dominance of ferrous iron in berthierine/chlorite, a change in redox conditions is needed for Fe illite-berthierine transformation and could be linked to oil degradation in the reservoir.

Keyword spotting (KWS) is one of the major tasks required in online sensing and surveillance devices. This operation requires the processing and analysis of the audio signals that are complex signals with a random distribution of the information in time. In that context, this paper proposes an event-driven low-power time-based feature extractor computing the spectral energy distribution in the form of a spectrogram. A convolutional neural network (CNN) then performs the classification to detect a specific keyword among multiple learned keywords. This system has been tested on a DE2-115 Inter® FPGA board to prove the hardware implementability and synthesized on 60nm CMOS technology for power computation. Furthermore, the combined simulation of the FPGA implementation and the Matlab® model has demonstrated that the system performs the recognition of up to 10 keywords with an accuracy of 90.4% and a power budget of 9.24μW, for an average speech rate of 60 words/minute.

Integrated circuits (ICs) are a keystone for most critical applications operating in high-level radiation environments, spanning from high-energy nuclear applications up to space applications. The long-term reliability of these applications is essential for safe operation. However, the radiation effects for ICs are commonly investigated using fresh circuits, leaving the coupled effect of radiation and aging degradation unknown. This article investigates the impact of negative bias temperature instability (NBTI) aging degradation mechanism on the heavy-ion single event upset (SEU) radiation susceptibility of 28-nm ultra-thin body and buried oxide (UTBB) fully depleted silicon on insulator (FD-SOI) technology using a custom-designed test vehicle. NBTI aging degradation mechanism has been experimentally proven to increase the SEU sensitivity up to $2\times $ for 28-nm UTBB FD-SOI flip-flops. A comprehensive framework is presented to analyze the underlying mechanisms for the impact of NBTI, which includes NBTI aging mechanism modeling, SEU SPICE simulation, TCAD irradiation simulation, and Monte-Carlo simulation of radiation effects. The framework offered a quantitative prediction of the effect of NBTI degradation mechanism on the heavy-ion SEU radiation sensitivity.

The turbidite sandstone of the Agat Formation is one of the most important hydrocarbon reservoirs in the Lower Cretaceous succession of the Norwegian North Sea. The diagenetic history of these sandstones has greatly impacted the reservoir quality, resulting in a high heterogeneity in porosity and permeability. To investigate this impact, sixty thin sections and rock samples were investigated through petrographic observations, geochemical microanalysis, and X-ray diffraction. Three main diagenetic events occurred during burial and influenced the reservoir properties: (1) mutual growth of chlorite, siderite, and apatite, (2) calcite cementation, and (3) dissolution of K-feldspar and calcite. Early diagenesis occurs under strongly reducing conditions, leading to the formation of Fe-chlorite/berthierine, siderite, pyrite, and apatite forming coatings around the detrital grains. Chlorite coatings have a key role in controlling the reservoir quality in the Agat Formation. In the lower part of the reservoir, chlorite forms thick coats (15-25 mu m) that strongly reduce permeability. Conversely, in the upper part of the reservoir, chlorite coats are thinner (4-10 mu m), favoring porosity preservation by inhibiting quartz overgrowth. X-ray diffraction and electron microscope observations indicate Fe-rich chlorite of Ib polytype, suggesting solid-state transformation from a berthierine precursor. The kinetics of berthierine/chlorite growth are governed by the dissolution of Fe-rich grains and fine fraction materials found as matrix and early coats. These grains are abundant in the lower part of the reservoir favoring the formation of thick chlorite coats. Chlorite was followed by pervasive calcite cementation in deep burial conditions creating low permeability barriers in the reservoir. The source of this calcite is internal and probably related to the dissolution of an early diagenetic calcite cement forming at shallow depth, with a minor contribution from marine carbonates. During late diagenesis, K-feldspar grains and calcite cements underwent intense dissolution creating secondary porosity and consequently increasing permeability. The input of acidic fluids associated with the maturation of the source rocks could have facilitated these dissolution reactions. Our results highlight the importance of inherited mineralogy in controlling the distribution of chlorite, carbonates, and secondary porosity in the Agat Formation. This study provides useful indicators to help predict diagenetic reactions that can occur in deep marine silici-clastic reservoirs.

Sediment gravity flows transport large volumes of sand and clay minerals into submarine systems, which store some of the world's major reserves of oil and gas. However, knowledge about grain‐coating clay mineral formation and its role in preserving reservoir quality in deep marine settings is poorly documented. Here we present a case study on the Agat Formation, a deep marine deposit interpreted as a series of turbidites, using a multimethod approach including petrographical, petrophysical and sedimentological data. This study investigates the occurrence and origin of chlorite coating and demonstrates how extensive chlorite coating substantially affects reservoir quality. The presence of green marine clay pellets suggests an initial shallow marine origin and sedimentological evidence reveals that the sediments were later remobilized by gravity flows and deposited at their present location. We suggest that the precursor clay coating was emplaced prior to sediment remobilization because of the presence of clay coating on grain contacts and all detrital components, the continuous nature of coating and the lack of clay bridges between the grains. Therefore, the origin of chlorite coating in deep marine environments may be recognized using the characteristic properties of inherited precursor clay coating. Chlorite coating thickness varies between an upper and lower sand unit, with an average of ca. 4.5 µm and ca. 24 µm, respectively. Permeability is significantly reduced in the interval with exceedingly thick chlorite coating but shows only a subtle decrease in helium porosity. This study enlightens the importance of crucially evaluating porosity in sandstones with thick chlorite coating using a multimethod approach. The results from this study can be useful in future exploration endeavours in the area and in other deep marine systems with a similar setting worldwide.

This paper presents circuit monitoring, reviewing different classes of silicon monitoring solutions, the specificity of each class, and where they best fit in the life-cycle of circuit design. A diversified circuit monitoring strategy is presented, that encompasses Early-Silicon versus Computer-Aided-Design (CAD) calibration, In-Die High Volume Manufacturing (HVM) testing and In-Field Dynamic monitoring, coupled with Adaptative Voltage Scaling (AVS) and/or Adaptative Body-Bias (ABB) live compensation. Illustrations of Silicon measurement and CAD simulation analysis from 40nm CMOS and 28nm FD-SOI, general purpose and automotive derivative, are presented.

This work presents a compact voltage and frequency scalable clock generator for low-power digital SoC clocking. Named Direct Digital Sampling and Synthesis (DDSS), the open-loop generator implemented in 28nm FD-SOI operates from 0.45V to 1.1V with measured jitter from 2.0% to 5.1% UI. Its low power consumption of 0.40pJ/cycle at 57MHz 0.5V combined with the ability to perform fast frequency changes makes this circuit an alternative to PLLs for fast Dynamic Voltage and Frequency Scaling (DVFS) strategies in low power SoCs.

Analyzing image content usually comes at the expense of a power consumption incompatible with battery-powered systems. Aiming at proposing a solution to this problem, this paper presents an imager with full on-chip object recognition, consuming sub-10μW using standard 4T pixels in 90nm imaging CMOS technology, opening the path for both wake-up and high-quality imaging. It combines multi-modality event-of-interest detection with self-controlled capabilities, a key for low-power applications. It embeds a log-domain auto-exposure algorithm to increase on-chip automation. The power consumption figures range from 3.0 to 5.7μW at 5fps for a QQVGA resolution while enabling background subtraction and single-scale object recognition. This typically shows a measured 94% accuracy for a face detection use case.

Temperature monitoring is critical to the operation of all SoCs, as it provides information to adjust logic timing, optimize power management, or calibrate analog circuits. The temperature information should be obtained with a fine spatial and temporal resolution, which requires low-area sensors with a fast conversion time. Digital MOS sensors offer such performance and take full advantage of process scaling; however, they often require costly two-point calibration to achieve the desired accuracy. This chapter presents a digital sensor in 28 nm FD-SOI process which takes advantage of the technology’s extended body-biasing capabilities for process compensation. Through an NMOS-only oscillator, a single regulator provides a low-noise supply and NWell biasing. The probe is complemented by an on-chip digital logic backend to compensate for non-linearities. The whole system achieves an accuracy of −1.4 ∘C/1.3 ∘C, a per-probe area of 1044 µm2, and accommodates a wide operating range (0.62–1.2 V) and satisfying power (2.0 nJ/Sa) and accuracy.

This section reports open loop Body-Bias control implemented in two designs. The first one is based on a microprocessor hosting bias control hardware and the compensation software, using a C-code look-up table to adjust bias according to temperature via periodic interrupt handler. The second one is based on a full ASIC logic flow and applies a combination of linear laws to achieve voltage and temperature compensation. In both cases, the bias laws are derived from the bias response exposed in Sect. 3.2 . The comparison of both methods is made for the reader to take his decision on which approach best suits his circuit need.