Deep neural network (DNN) accelerators [1-3] have been proposed to accelerate deep learning algorithms from face recognition emotion in mobile or embedded environments [3]. However, most works only the convolutional layers (CLs) fully-connected (FCLs), and different DNNs, such as those containing recurrent (RLs) (useful for recognition) not supported hardware. A combined CNN-RNN accelerator [1], separately optimizing computation-dominant CLs, memory-dominant RLs FCLs, was reported increase...

An energy-efficient deep neural network (DNN) accelerator, unified processing unit (UNPU), is proposed for mobile learning applications. The UNPU can support both convolutional layers (CLs) and recurrent or fully connected (FCLs) to versatile workload combinations accelerate various In addition, the first DNN accelerator ASIC that variable weight bit precision from 1 16 bit. It enables operate on accuracy-energy optimal point. Moreover, lookup table (LUT)-based bit-serial element (LBPE) in...

Generative adversarial networks (GAN) have a wide range of applications, from image style transfer to synthetic voice generation [1]. GAN applications on mobile devices, such as face-to-Emoji conversion and super-resolution imaging, enable more engaging user interaction. As shown in Fig. 7.4.1, consists 2 competing deep neural (DNN): generator discriminator. The discriminator is trained, while the fixed, distinguish whether generated real or fake. On other hand, trained generate fake images...

Spiking-Neural-Networks (SNNs) have been studied for a long time, and recently shown to achieve the same accuracy as Convolutional-Neural-Networks (CNNs). By using CNN-to-SNN conversion, SNNs become promising candidate ultra-low power Al applications [1]. For example, compared BNNs or XOR-nets, provide lower consumption higher [2]. This is because perform spike-based event-driven operation with high spike sparsity, unlike CNN's frame-driven operation. Fig. 22.5.1 shows that energy of SNN...

In-memory computing (IMC) processors show significant energy and area efficiency for deep neural network (DNN) processing [1–3]. As shown in Fig. 16.5.1, despite promising macro-level throughput, there remain three main challenges to extending gains system performance with a high integration level. First, most previous works had fixed configuration size of IMC macros, when the macro was smaller than DNN layer's dimension, repetitive memory accesses were required IA/OA, consuming >40% power....

An activatable fluorescent probe from indocyanine was developed for the detection of tumor-enriched γ-glutamyltranspeptidase (γGT). The exhibited a dramatic fluorescence enhancement (F/F0 = 10) as well bathochromic shift (>100 nm) upon treatment γGT with low limit 0.15 unit/L and further successfully applied sensitive in mouse model colon cancer.

An energy-efficient floating-point DNN training processor is proposed with heterogenous bfloat16 computing architecture using exponent computing-in-memory (CIM) and mantissa processing engine. Mantissa free calculation enables pipelining of operation for while reducing MAC power by 14.4 %. 6T SRAM bitline charge reusing reduces memory access 46.4 The fabricated in 28 nm CMOS technology occupies 1.62×3.6 mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Recently, transformer-based large language models (LLMs), shown in Fig. 20.5.1, are widely used, and even on-device LLM systems with real-time responses anticipated [1]. Many transformer processors [2–4] enhance energy efficiency by increasing hardware utilization reducing power consumption, but their system consumption response time still not suitable for mobile devices. Since LLMs, such as GPT-2, have many parameters (400-700M), External Memory Access (EMA) consumes 68% of the total power....

This study considers a comparative on pressure sensors for the measurement of sloshing impact pressure. For study, four are used: one piezoresistive sensor, piezoelectric and two integrated circuit (ICP) sensors. installed tank wall ceiling rectangular with narrow breadth. Several types studies carried out, including sensitivity to temperature differences between test medium. The forced regular irregular motions applied partial water filling, signals due measured at different filling...

This article presents generative adversarial network processing unit (GANPU), an energy-efficient multiple deep neural (DNN) training processor for GANs. It enables on-device of GANs on performance- and battery-limited mobile devices, without sending user-specific data to servers, fully evading privacy concerns. Training require a massive amount computation, therefore, it is difficult accelerate in resource-constrained platform. Besides, networks layers show dramatically changing operational...

Seoul National University has conducted a considerable number of six degree-of-freedom irregular small-scale sloshing model tests 1/70–1/25 scales, particularly focusing on the tanks liquefied natural gas (LNG) carriers. An experimental database been created to provide information load severity, which are obtained from lot post-processed results. In this paper, summary is described. The artificial neural network trained based predict severity. Various attributes that affect results...

An energy-efficient neuromorphic computing-in-memory (CIM) processor is proposed with four key features: 1) Most significant bit (MSB) Word Skipping to reduce the BL activity; 2) Early Stopping enable lower 3) Mixed-mode firing for multi-macro aggregation; 4) Voltage Folding extend dynamic range. The CIM achieves state-of-the-art energy efficiency of 62.1 TOPS/W (I=4b, W=8b) and 310.4 W=1b).

A highly energy-efficient neuromorphic computing-in-memory (Neuro-CIM) processor is proposed for ultralow-power deep learning applications. Neuro-CIM can support spiking neural network (SNN) to eliminate the power and area overhead of previous CIM processor. The sign extended bits gating reduces bitline (BL) voltage switching rate due negative small-magnitude weights allowing 38% reduction at 8-b weight condition 25% 4-b condition. In addition, replaces high-precision analog-to-digital...

A low power face recognition (FR) convolutional neural network (CNN) processor is proposed with high efficiency to achieve always-on FR in mobile devices. Three key features enable a power-efficient CNN. First, tile-based clustering (THC) for reducing the computation overhead of hierarchical clustering. It generates an average 37.2% duplicated input entire network. Second, latency core proposed. supports approximated method that removes distance updates and increases pipeline utilization by...

An energy efficient Deep-Neural-Network (DNN) learning processor is proposed for on-chip and iterative weight pruning (WP). This work has three key features: 1) stochastic coarse-fine reduced computation workload by 99.7% compared with previous WP algorithm while maintaining high sparsity, 2) adaptive input/output/weight skipping (AIOWS) achieved 30.1× higher throughput than DNN [1] not only the inference but also learning, 3) memory shared unit removed access WP. As a result, this shows...

This article presents DynaPlasia, a reconfigurable eDRAM-based in- memory computing (IMC) processor with novel triple-mode cell. It enables higher system-level performance and efficiency in resource-limited environment. DynaPlasia proposes five key features that can enhance the energy area of IMC accelerator: 1) dynamic core architecture (DRECA), which dynamically reconfigures effective macro size according to DNN workloads; 2) cell is reconfigured as PE, unit DAC, optimize system resource...

This paper considers scale effects on three-dimensional (3D) sloshing flows. A series of model tests were conducted for three differently scaled tanks. The tanks considered in this study 1:70, 1:50, and 1:30 membrane type based a 138,000 m3 liquid natural gas carrier model. carried out harmonic sway roll motions different filling depths with various excitation frequencies. pressure measuring points the same, as if they up to actual size. main parameters investigated peak rise time sampled...

This article presents a low-power, low-distortion, and compact mixed-signal sinusoidal current generator (CG) IC for bio-impedance (Bio-Z) sensing applications. By utilizing the digital <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\Delta \Sigma $ </tex-math></inline-formula> modulation to bridge digitally synthesized sinewave data analog-domain voltage output, implementation of low-distortion lookup...

This article presents MetaVRain, a low-power neural 3-D rendering processor for metaverse realization on mobile devices. The MetaVRain mainly focused solving high operational intensity problem that appeared during the radiance fields (NeRFs)-based rendering. It imitates brain-inspired visual perception processes and constructs new NeRF acceleration architecture, bundle-frame-familiarity (BuFF). built-in core (VPC) realizes BuFF architecture by accelerating three stages: 1) spatial attention...