• JSTS:Journal of Semiconductor Technology and Science
• /
• v.14 no.6
• /
• pp.760-767
• /
• 2014

This paper describes a high-radix crossbar switch design with low latency and power dissipation for Network-on-Chip (NoC) applications. The reduction in latency and power is achieved by employing a folded-clos topology, implementing the switch organized as three stages of low-radix switches connected in cascade. In addition, to facilitate the uniform placement of wires among the sub-switch stages, this paper proposes a Mux-Matrix-Mux structure, which implements the first and third switch stages as multiplexer-based crossbars and the second stage as a matrix-type crossbar. The proposed 256-radix, 8-bit crossbar switch designed in a 65nm CMOS has the simulated power dissipation of 1.92-W and worst-case propagation delay of 0.991-ns while operating at 1.2-V supply and 500-MHz frequency. Compared with the state-of-the-art designs in literature, the proposed crossbar switch achieves the best energy-delay-area efficiency of $0.73-fJ/cycle{\cdot}ns{\cdot}{\lambda}^2$.

• ETRI Journal
• /
• v.38 no.1
• /
• pp.100-111
• /
• 2016

This paper presents a low-complexity channel-adaptive reconfigurable $4{\times}4$ QR-decomposition and M-algorithm-based maximum likelihood detection (QRM-MLD) multiple-input and multiple-output (MIMO) detector. Two novel design approaches for low-power QRM-MLD hardware are proposed in this work. First, an approximate survivor metric (ASM) generation technique is presented to achieve considerable computational complexity reduction with minor BER degradation. A reconfigurable QRM-MLD MIMO detector (where the M-value represents the number of survival branches in a stage) for dynamically adapting to time-varying channels is also proposed in this work. The proposed reconfigurable QRM-MLD MIMO detector is implemented using a Samsung 65 nm CMOS process. The experimental results show that our ASM-based QRM-MLD MIMO detector shows a maximum throughput of 288 Mbps with a normalized power efficiency of 10.18 Mbps/mW in the case of $4{\times}4$ MIMO with 64-QAM. Under time-varying channel conditions, the proposed reconfigurable MIMO detector also achieves average power savings of up to 35% while maintaining a required BER performance.

• Journal of Semiconductor Engineering
• /
• v.2 no.3
• /
• pp.136-141
• /
• 2021

This paper analyzes the tree expansion for multiple-input multiple-output (MIMO) detection in the viewpoint of hardware implementation. The tree expansion is to calculate path metrics of child nodes performed in every visit to a node while traversing the detection tree. Accordingly, the tree-expansion unit (TEU), which is responsible for such a task, has been an essential component in a MIMO detector. Despite the paramount importance, the analyses on the TEUs in the literature are not thorough enough. Accordingly, we further investigate the hardware complexity of the TEUs to suggest a guideline for selection. In this paper, we focus on a pair of major ways to implement the TEU: 1) a full parallel realization; 2) a transformation of the formulae followed by common subexpression elimination (CSE). For a logical comparison, the numbers of multipliers and adders are first enumerated. To evaluate them in a more practical manner, the TEUs are implemented in a 65-nm CMOS process, and their propagation delays, gate counts, and power consumptions were measured explicitly. Considering the target specification of a MIMO system and the implementation results comprehensively, one can choose which architecture to adopt in realizing a detector.

• Journal of Sensor Science and Technology
• /
• v.28 no.4
• /
• pp.240-245
• /
• 2019

Phase interpolation is widely adopted in frequency synthesizers and clock-and-data recovery systems to produce an intermediate phase from two existing phases. The intermediate phase is typically generated by combining two input phases with different weights. Unfortunately, this results in non-uniform phase steps. Alternatively, the intermediate phase can be generated by successive approximation, where the interpolated phase at each approximation stage is obtained using the same weight for the two intermediate phases. As a proof of concept, this study presents a 2-GHz 8-bit successive approximation digital-to-phase converter that is designed using 65-nm CMOS technology. The converter receives an 8-phase clock signal as input, and the most significant bit (MSB) section selects four phases to create two sinusoidal waveforms using a harmonic rejection filter. The remaining least significant bit (LSB) section applies the successive approximation to generate the required intermediate phase. Monte-Carlo simulations show that the proposed converter exhibits 0.46-LSB integral nonlinearity and 0.31-LSB differential nonlinearity with a power consumption of 3.12 mW from a 1.2-V supply voltage.

• Journal of Semiconductor Engineering
• /
• v.2 no.2
• /
• pp.130-135
• /
• 2021

Recently, deep neural networks (DNNs) are actively used for action control so that an autonomous system, such as the robot, can perform human-like behaviors and operations. Unlike recognition tasks, the real-time operation is essential in action control, and it is too slow to use remote learning on a server communicating through a network. New learning techniques, such as reinforcement learning (RL), are needed to determine and select the correct robot behavior locally. In this paper, we propose an energy-efficient DNN processor with a LUT-based processing engine and near-zero skipper. A CNN-based facial emotion recognition and an RNN-based emotional dialogue generation model is integrated for natural HRI system and tested with the proposed processor. It supports 1b to 16b variable weight bit precision with and 57.6% and 28.5% lower energy consumption than conventional MAC arithmetic units for 1b and 16b weight precision. Also, the near-zero skipper reduces 36% of MAC operation and consumes 28% lower energy consumption for facial emotion recognition tasks. Implemented in 65nm CMOS process, the proposed processor occupies 1784×1784 um2 areas and dissipates 0.28 mW and 34.4 mW at 1fps and 30fps facial emotion recognition tasks.

• International journal of advanced smart convergence
• /
• v.10 no.1
• /
• pp.12-23
• /
• 2021

This paper proposes a low power radio frequency receiver front-end where, in a single stage, single-balanced mixer and voltage-controlled oscillator are stacked on top of low noise amplifier and re-use the dc current to reduce the power consumption. In the proposed topology, the voltage-controlled oscillator itself plays the dual role of oscillator and mixer by exploiting a series inductor-capacitor network. Using a 65 nm complementary metal oxide semiconductor technology, the proposed radio frequency front-end is designed and simulated. Oscillating at around 2.4 GHz frequency band, the voltage-controlled oscillator of the proposed radio frequency front-end achieves the phase noise of -72 dBc/Hz, -93 dBc/Hz, and -113 dBc/Hz at 10KHz, 100KHz, and 1 MHz offset frequency, respectively. The simulated voltage conversion gain is about 25 dB. The double-side band noise figure is -14.2 dB, -8.8 dB, and -7.3 dB at 100 KHz, 1 MHz and 10 MHz offset. The radio frequency front-end consumes only 96 ㎼ dc power from a 1-V supply.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.3
• /
• pp.75-81
• /
• 2015

This paper proposed the new eight-parallel MDC FFT processor using the eight-parallel MDC architecture and the efficient scheduling scheme. The proposed FFT processor supports the 256-point FFT based on the modified radix-$2^6$ FFT algorithm. The proposed scheduling scheme can reduce the number of complex multipliers from eight to six without increasing delay buffers and computation cycles. Moreover, the proposed FFT processor can be used in OFDM systems required high throughput and low hardware complexity. The proposed FFT processor has been designed and implemented with a 90nm CMOS technology. The experimental result shows that the area of the proposed FFT processor is $0.27mm^2$. Furthermore, the proposed eight-parallel MDC FFT processor can achieve the throughput rate up to 2.7 GSample/s at 388MHz.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.17 no.3
• /
• pp.411-424
• /
• 2017

An all-digital delay-locked loop (DLL) for a mobile memory interface, which runs at 0.11-2.5 GHz with a phase-shift capability of $180^{\circ}$, has two internal DLLs: a global DLL which uses a time-to-digital converter to assist fast locking, and shuts down after locking to save power; and a local DLL which uses a phase detector with an adaptive phase sampling window (WPD) to reduce jitter accumulation. The WPD in the local DLL adjusts the width of its sampling window adaptively to control the loop bandwidth, thus reducing jitter induced by UP/DN dithering, input clock jitter, and supply/ground noise. Implemented in a 65 nm CMOS process, the DLL operates over 0.11-2.5 GHz. It locks within 6 clock cycles at 0.11 GHz, and within 17 clock cycles at 2.5 GHz. At 2.5 GHz, the integrated jitter is $954fs_{rms}$, and the long-term jitter is $2.33ps_{rms}/23.10ps_{pp}$. The ratio of the RMS jitter at the output to that at the input is about 1.17 at 2.5 GHz, when the sampling window of the WPD is being adjusted adaptively. The DLL consumes 1.77 mW/GHz and occupies $0.075mm^2$.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.16 no.6
• /
• pp.860-866
• /
• 2016

An injection-locked ring phase-locked loop (ILRPLL) using a charge-stored complementary switch (CSCS) injection technique is described in this paper. The ILRPLL exhibits a wider lock range compared to other conventional ILRPLLs, owing to the improvement of the injection effect by the proposed CSCS. A frequency calibration loop and a device mismatch calibration loop force the frequency error to be zero to minimize jitter and reference spur. The prototype chip fabricated in 65-nm CMOS technology achieves a $285-fs_{rms}$ integrated jitter at GHz from the reference clock of 52 MHz while consuming 7.16 mW. The figure-of-merit of the ILRPLL is -242.4 dB.

• ETRI Journal
• /
• v.37 no.6
• /
• pp.1143-1153
• /
• 2015

In this paper, we propose an efficient architecture for a real-time image stitching engine for vision SoCs found in motor vehicles. To enlarge the obstacle-detection distance and area for safety, we adopt panoramic images from multiple telegraphic cameras. We propose a stitching method based on a fixed homography that is educed from the initial frame of a video sequence and is used to warp all input images without regeneration. Because the fixed homography is generated only once at the initial state, we can calculate it using SW to reduce HW costs. The proposed warping HW engine is based on a linear transform of the pixel positions of warped images and can reduce the computational complexity by 90% or more as compared to a conventional method. A dual-core SW/HW image stitching engine is applied to stitching input frames in parallel to improve the performance by 70% or more as compared to a single-core engine operation. In addition, a dual-core structure is used to detect a failure in state machines using rock-step logic to satisfy the ISO26262 standard. The dual-core SW/HW image stitching engine is fabricated in SoC with 254,968 gate counts using Global Foundry's 65 nm CMOS process. The single-core engine can make panoramic images from three YCbCr 4:2:0 formatted VGA images at 44 frames per second and frequency of 200 MHz without an LCD display.