4 December 2009 in IEEE Transactions on Advanced Packaging
Fabrication methods of planar printed circuits on fabrics are introduced and their electrical characteristics are measured and analyzed. Wet patterning method like screen printing as well as dry process of sputtering are used to fabricate the patterned film electrodes on various types of fabrics. The minimum width of the patterns is 0.2 mm for screen printing and 0.1 mm for gold sputtering, and the typical sheet resistance is 134 m ¿/¿. Fabrication methods of capacitors of 1 pF-1 nF and inductors of 500 nH-1 ¿H at 10 MHz on the fabrics are also introduced. Bonding and packaging of silicon chip directly on the fabric circuit board are proposed and their mechanical properties are investigated. The ac impedance of the transmission line is measured as 201-215 ¿ with variation, and the time-domain reflectometry profile shows that the -3 dB frequency of the printed transmission line of 15 cm on the fabric is 80 MHz. A complete system composed of a fabric capacitor sensor input, a controller system-on-a-chip, and an LED array display is implemented on the fabric and its operation is demonstrated successfully.
13 December 2017 in IEEE Journal of Solid-State Circuits
A Low-power convolutional neural network (CNN)-based face recognition system is proposed for the user authentication in smart devices. The system consists of two chips: an always-on CIS-based face detector (FD) and a low-power CNN processor. For always-on FD, analog-digital Hybrid Haar-like FD is proposed to improve the energy efficiency of FD by 39%. For low-power CNN processing, the CNN processor with 1024 MAC units and 8192-bit-wide local distributed memory operates at near threshold voltage, 0.46 V with 5-MHz clock frequency. In addition, the separable filter approximation is adopted for the workload reduction of CNN, and transpose-read SRAM using 7T SRAM cell is proposed to reduce the activity factor of the data read operation. Implemented in 65-nm CMOS technology, the 3.30 x 3.36 mm² CIS chip and the 4 x 4 mm² CNN processor consume 0.62 mW to evaluate one face at 1 fps and achieved 97% accuracy in LFW dataset.
23 March 2021 in IEEE Journal of Solid-State Circuits
This article presents HNPU, which is an energy-efficient deep neural network (DNN) training processor by adopting algorithm-hardware co-design. The HNPU supports stochastic dynamic fixed-point representation and layer-wise adaptive precision searching unit for low-bit-precision training. It additionally utilizes slice-level reconfigurability and sparsity to maximize its efficiency both in DNN inference and training. Adaptive bandwidth reconfigurable accumulation network enables reconfigurable DNN allocation and maintains its high core utilization even in various bit-precision conditions. Fabricated in a 28-nm process, the HNPU accomplished at least $5.9\times $ higher energy efficiency and $2.5\times $ higher area efficiency in actual DNN training compared with the previous state-of-the-art on-chip learning processors.
9 September 2020 in IEEE Transactions on Circuits and Systems I: Regular Papers
Deep neural network (DNN) has been widely studied due to its high performance and usability for various applications such as image classification, detection, segmentation, translation, and action recognition. Thanks to the universal applications and high performance of DNN algorithm, DNN is adopted for various AI platforms, including edge/mobile devices as well as cloud servers. However, high-performance DNN requires a large amount of computation and memory access, making it challenging to implement DNN operation on edge/mobile. There have been several ways to solve these problems, including algorithms as well as hardware for DNN. Algorithms that help accelerate DNN in hardware enable much more efficient operation of high-performance AI. This article aims to provide an overview of the recent hardware and algorithm co-design schemes enabling efficient processing of DNNs. Specifically, it will provide algorithm optimization methods for DNN structure, neurons, synapses, and data types. This paper also introduces optimization methods for hardware architectures, PE array, data-path control, and microarchitecture of PE. And we will also show examples of DNN algorithm and hardware co-designed ASICs.
5 August 2020 in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
In this paper, we propose an energy-efficient convolutional neural network (CNN) based super-resolution (SR) processor, super-resolution neural processing unit (SRNPU), for mobile applications. Traditionally, it is hard to realize real-time CNN-based SR on resource-limited platforms like mobile devices due to its massive amount of computation workload and communication bandwidth with external memory. The SRNPU can support the tile-based selective super-resolution (TSSR) which dynamically selects the proper sized CNN in a tile-by-tile manner. The TSSR reduces the computational workload of CNN-based SR by 31.1 % while maintaining image restoration performance. Moreover, a proposed selective caching based convolutional layer fusion (SC2LF) can reduce 78.8 % of external memory bandwidth with 93.2 % smaller on-chip memory footprint compared with previous layer fusion methods, by only caching short reuse distance intermediate feature maps. Additionally, reconfigurable cyclic ring architecture in the SRNPU enables maintaining high PE utilization by amortizing the reloading process caused by SC2LF operation under various convolutional layer configurations. The SRNPU is fabricated in 65 nm CMOS technology and occupies 4 × 4 mm2 die area. The SRNPU has a peak power efficiency of 1.9 TOPS/W at 0.75 V, 50 MHz. The SRNPU achieves 31.8 fps ×2 scale Full-HD generation and 88.3 fps ×4 scale Full-HD generation with higher restoration performance and power efficiency than previous SR hardware implementations. To the best of our knowledge, the SRNPU is the first ASIC implementation of the CNN-based SR algorithm which supports real-time Full-HD up-scaling.
17 August 2022 in IEEE Journal of Solid-State Circuits
This article presents the first keyword spotting (KWS) IC that uses a ring-oscillator-based time-domain processing technique for its analog feature extractor (FEx). Its extensive usage of time-encoding schemes allows the analog audio signal to be processed in a fully time-domain manner except for the voltage-to-time conversion stage of the analog front end. Benefiting from fundamental building blocks based on digital logic gates, it offers better technology scalability compared to conventional voltage-domain designs. Fabricated in a 65-nm CMOS process, the prototyped KWS IC occupies 2.03 mm2 and dissipates 23- $\mu$ W power consumption, including analog FEx and digital neural network classifier. The 16-channel time-domain FEx achieves a 54.89-dB dynamic range for 16-ms frame shift size while consuming 9.3 $\mu$ W. The measurement result verifies that the proposed IC performs a 12-class KWS task on the Google Speech Command dataset (GSCD) with $>$ 86% accuracy and 12.4-ms latency.