This paper presents a fully passive 2nd-order noise-shaping successive approximation register (SAR) analog-to-digital converter (ADC) designed specifically for low-power and low-cost Internet of Things (IoT) applications. By optimizing the coefficients, a substantial 24 dB in-band quantization noise suppression is achieved. To further reduce power consumption and the total unit capacitor count, a hybrid switching procedure and optimal logic are utilized. The measurement result shows that this design achieves an effective number of 10.31 bits over a 3.125 MHz signal bandwidth. At a power supply of 1.8 V, the power consumption is measured to be 728 µW with a sampling rate of 50 MS/s. Fabricated in 180-nm CMOS technology, the ADC core occupies an area of 0.117 mm2. The Schrier figure-of-merit (FoM) of 160.13 dB is obtained.
This paper presents a high-linearity millimeter-wave (mmWave) Doherty power amplifier (PA) for 5G FR2 wireless communications. A precise mutual distortion cancellation (PMDC) method for parallel Doherty PA is proposed. The auxiliary path current amplitude is adjusted by the common-source common-gate (CSCG) dual adaptive bias circuit (DADB). Meanwhile, the main path current phase is tuned by the phase nonlinear compensation capacitor (PNCC). These two techniques are combined to enable precise distortion cancellation, thereby optimizing the Doherty PA’s AM-AM and AM-PM distortions. The chip is fabricated in a 40 nm CMOS process. The measurement results show that the proposed Doherty PA achieves a small-signal gain of 19.6 dB, a 3-dB bandwidth from 21 to 29 GHz, a saturated output power (Psat) of 19.1 dBm, an output 1-dB compression point (OP1dB) of 17.9 dBm, a peak power added efficiency (PAEpeak) of 22.7%, a power added efficiency at 6-dB power back-off (PAE -6dB) of 15.1%, and an AM-PM distortion of 0.93° at 26 GHz. When using a 64-QAM 100 MHz OFDM modulated signal as the input, which meets the 5G NR FR2 communication protocol and has an input PAPR of 11.38 dB, the measured output EVM is -25 dB, and ACLR is -30.5 dBc. The proposed PA realizes an average output power (Pavg) of 12.6 dBm and an average power added efficiency (PAEavg) of 11.3%.
This paper presents a fully-integrated dual-mode switched-capacitor (SC) DC-DC converter to supply always-on wake-up timer in biosensor systems. To balance chip area and conversion efficiency, optimal sizes of power switches, flying capacitor (Cfly) and output capacitor (Cout) in SC cell are determined through power loss analysis method. Power-on and self-powered modes (dual modes) under two voltage domains and an on-chip ultra-low-power (ULP) bias circuit are proposed to further improve the conversion efficiency. Furthermore, a hybrid self-biased current (HSBC) scheme is utilized to achieve low output voltage ripple. Implemented with a 40-nm CMOS process, this DC-DC converter can realize the voltage conversion from 1.8 V to 0.4 V. Measurement results show that the peak conversion efficiency reaches 75 % at average load arriving 3 µA, while the voltage ripple is below 12 mV over 10 nA-5 µA load range.
Cost aggregation is a crucial step in the accurate stereo depth estimation process known as semi-global matching. However, this step is challenged by storing large amounts of aggregated data, which is necessary to achieve high matching accuracy under large resolution and large disparity conditions. In this paper, we propose a multi-path optimization aggregation strategy and re-select the complementary combinations of key paths in the forward and backward scanning directions to improve the matching accuracy as much as possible. An error rate of only 5.21% is achieved on the KITTI 2015 dataset. Next, we propose DCT-based truncated compression and selective storage to alleviate the problem of memory increase due to the introduction of reverse critical aggregation paths. Experiments show that the matching error rate increases by only 0.6% on the KITTI 2015 dataset with 53% memory savings. Finally, 1920 × 1080 @62fps @128MHz is achieved on ZCU102 FPGA.
Convolutional neural networks (CNNs) have achieved remarkable success in computer vision tasks, driving the rapid development of hardware accelerators. However, memory efficiency remains a key challenge, as conventional accelerators adopt layer-by-layer processing, leading to frequent external memory accesses (EMAs) of intermediate feature data, which increase energy consumption and latency. While layer fusion has been proposed to enhance inter-layer feature reuse, existing approaches typically rely on fixed data management tailored to specific architectures, introducing on-chip memory overhead and requiring trade-offs with EMAs. Moreover, prevalent residual connections further weaken fusion benefits due to diverse data reuse distances. To address these challenges, we propose layer-centric tile fusion, which integrates residual data loading with feature merging by leveraging receptive field relationships among feature tiles. A reuse distance-aware caching strategy is introduced to support flexible storage for various data types. We also develop a modeling framework to analyze the trade-off between on-chip memory usage and EMA-induced energy-delay product (EDP). Experimental results demonstrate that our method achieves 5.04–43.44% EDP reduction and 20.28–58.33% memory usage reduction compared to state-of-the-art designs on ResNet-18 and SRGAN.
Versatile Video Coding (VVC) has significantly increased encoding efficiency at the expense of numerous complex coding tools, particularly the flexible Quad-Tree plus Multi-type Tree (QTMT) block partition. This paper proposes a deep learning-based algorithm applied in fast QTMT partition for VVC intra coding. Our solution greatly reduces encoding time by early termination of less-likely intra prediction and partitions with negligible BD-BR increase. Firstly, a redesigned U-Net is recommended as the network's fundamental framework. Next, we design a Quality Parameter (QP) fusion network to regulate the effect of QPs on the partition results. Finally, we adopt a refined post-processing strategy to better balance encoding performance and complexity. Experimental results demonstrate that our solution outperforms the state-of-the-art works with a complexity reduction of 44.74% to 68.76% and a BD-BR increase of 0.60% to 2.33%.