• Title/Summary/Keyword: Poly-encoder

Search Result 13, Processing Time 0.018 seconds

Poly-encoder based COVID-19 Question and Answering with Task Adaptation (Poly-encoder기반의 COVID-19 질의 응답 태스크)

  • Lee, Seolhwa;Lim, Heuiseok
    • Annual Conference on Human and Language Technology
    • /
    • 2020.10a
    • /
    • pp.188-191
    • /
    • 2020
  • 본 연구는 COVID-19 질의 응답 태스크를 위한 Poly-encoder 기반의 태스크를 제안하였다. COVID-19 질의 응답 시스템은 사람들에게 최신 정보에 대해 빠르고 신뢰성이 높은 정보를 전달하는 특성을 가져야한다. 검색 기반 질의 응답 시스템은 pairwise 연산을 기반으로 수행되는데, Poly-encoder는 사전 학습된 트랜스포머(transformer)기반의 pairwise 연산 방법론 중 기존 Cross-encoder와 Bi-encoder보다 실사용 및 성능이 뛰어남을 보였다 [1]. 특히, Poly-encoder는 정확도가 높으면서도 빠른 응답속도를 가지며 검색기반의 각종 태스크에서 좋은 성능을 보였다. 따라서 본 연구는 COVID-19를 위한 Poly-encoder기반의 질의 응답 태스크를 위하여 기존 질의 응답 태스크와 페르소나 기반의 질의 응답 태스크로 두 가지 유형의 태스크를 생성하여 모델을 학습하였다. 또한 신뢰성 있는 리소스정보로부터 모델에 최신 정보 반영을 위하여 자동 크롤러를 구축하여 데이터를 수집하였다. 마지막으로 전문가를 통한 데이터셋을 구축하여 질문-응답과 질의어-질문에 대한 모델 검증을 수행하였다.

  • PDF

Design of an 1.8V 6-bit 2GSPS CMOS ADC with an One-Zero Detecting Encoder and Buffered Reference (One-Zero 감지기와 버퍼드 기준 저항열을 가진 1.8V 6-bit 2GSPS CMOS ADC 설계)

  • Park Yu Jin;Hwang Sang Hoon;Song Min Kyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.42 no.6 s.336
    • /
    • pp.1-8
    • /
    • 2005
  • In this paper, CMOS A/D converter with 6bit 2GSPS Nyquist input at 1.8V is designed. In order to obtain the resolution of 6bit and the character of high-speed operation, we present an Interpolation type architecture. In order to overcome the problems of high speed operation, a novel One-zero Detecting Encoder, a circuit to reduce the Reference Fluctuation, an Averaging Resistor and a Track & Hold, a novel Buffered Reference for the improved SNR are proposed. The proposed ADC is based on 0.18um 1-poly 3-metal N-well CMOS technology, and it consumes 145mW at 1.8V power supply and occupies chip area of 977um $\times$ 1040um. Experimental result show that SNDR is 36.25 dB when sampling frequency is 2GHz and INL/DNL is $\pm$0.5LSB at static performance.

Design of an 1.8V 12-bit 10MSPS Folding/Interpolation CMOS Analog-to-Digital Converter (1.8V 12-bit 10MSPS Folding/Interpolation CMOS Analog-to-Digital Converter의 설계)

  • Son, Chan;Kim, Byung-Il;Hwang, Sang-Hoon;Song, Min-Kyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.45 no.11
    • /
    • pp.13-20
    • /
    • 2008
  • In this paper, an 1.8V 12-bit 10MSPS CMOS A/D converter (ADC) is described. The architecture of the proposed ADC is based on a folding and interpolation using an even folding technique. For the purpose of improving SNR, cascaded-folding cascaded-interpolation technique, distributed track and hold are adapted. Further, a digital encoder algorithm is proposed for efficient digital process. The chip has been fabricated with $0.18{\mu}m$ 1-poly 4-metal n-well CMOS technology. The effective chip area is $2000{\mu}m{\times}1100{\mu}m$ and it consumes about 250mW at 1.8V power supply. The measured SNDR is about 46dB at 10MHz sampling frequency.

Design of a 3.3V 8-bit 200MSPS CMOS Folding/Interpolation ADC (3.3V 8-bit 200MSPS CMOS Folding/Interpolation ADC의 설계)

  • Na, Yu-Sam;Song, Min-Gyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.38 no.3
    • /
    • pp.198-204
    • /
    • 2001
  • In this paper, a 3V 8-bit 200MSPS CMOS folding / interpolation A/D Converter is proposed. It employs an efficient architecture whose FR(Folding Rate) is 8, NFB(Number of Folding Block) is 4, and IR (Interpolating Rate) is 8. For the purpose of improved SNDR by to be low input frequency, distributed track and hold circuits are included. In order to obtain a high speed and low power operation, further, a novel dynamic latch and digital encoder based on a novel delay error correction are proposed. The chip has been fabricated with a 0.35${\mu}{\textrm}{m}$ 2-poly 3-metal n-well CMOS technology. The effective chip area is 1070${\mu}{\textrm}{m}$$\times$650${\mu}{\textrm}{m}$ and it dissipates about 230mW at 3.3V power supply. The INL is within $\pm$1LSB and DNL is within $\pm$1LSB, respectively. The SNDR is about 43㏈, when the input frequency is 10MHz at 200MHz clock frequency.

  • PDF

Design of a $54{\times}54$-bit Multiplier Based on a Improved Conditional Sum Adder (개선된 조건 합 가산기를 이용한 $54{\times}54$-bit 곱셈기의 설계)

  • Lee, Young-Chul;Song, Min-Kyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.37 no.1
    • /
    • pp.67-74
    • /
    • 2000
  • In this paper, a $54{\times}54$-bit multiplier based on a improved conditional sum adder is proposed. To reduce the multiplication time, high compression-rate compressors without Booth's Encoding, and a 108-bit conditional sum adder with separated carry generation block, are developed. Furthermore, a design technique based on pass-transistor logic is utilized for optimize the multiplication time and the power consumption by about 5% compared to that of conventional one. With $0.65{\mu}m$, single-poly, triple-metal CMOS process, its chip size is $6.60{\times}6.69\;mm^2$ and the multiplication time is 135.ns at a 3.3V power supply.

  • PDF

Design of the 1.8V 6-bit 2GSPS CMOS ADC for the DVD PRML (DVD PRML을 위한 1.8V 6-bit 2GSPS CMOS ADC 설계)

  • Park Yu-Jin;Song Min-kyu
    • Proceedings of the IEEK Conference
    • /
    • 2004.06b
    • /
    • pp.537-540
    • /
    • 2004
  • In this paper, CMOS A/D converter with 6bit 2GSPS Nyquist input at 1.8V is designed. In order to obtain the resolution of 6bit and the character of high-speed operation. we present an Interpolation type architecture. In order to overcome the problems of high speed operation further a novel encoder, a circuit for the Reference Fluctuation, an Averaging Resistor and a Track & Hold for the improved SNR are proposed. The proposed Interpolation ADC consists of Track & Holt four resistive ladders with 64 taps, 32 comparators and digital blocks. The proposed ADC is based on 0.18um 1-poly 3-metal N-well CMOS technology, and it consumes 145mW at 1.8V power supply.

  • PDF

Design of a 3.3V 8-bit 200MSPS CMOS folding/interpolation ADC (3.3V 8-bit 200MSPS CMOS folding/interpolation ADC의 설계)

  • Song, Min Gyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.38 no.3
    • /
    • pp.44-44
    • /
    • 2001
  • 본 논문에서는 CMOS로 구현된 3.3V 8-bit 200MSPS의 Folding / Interpolation 구조의 A/D 변환기를 제안한다. 회로에 사용된 구조는 FR(Folding Rate)이 8, NFB(Number of Folding Block)가 4, Interpolation rate 이 8이며, 분산 Track and Hold 구조를 회로를 사용하여 Sampling시 입력주파수를 Hold하여 높은 SNDR을 얻을 수 있었다. 고속동작과 저 전력 기능을 위하여 향상된 래치와 디지털 Encoder를 제안하였고 지연시간 보정을 위한 회로도 제안하였다. 제안된 ADC는 0.35㎛, 2-Poly, 3-Metal, n-well CMOS 공정을 사용하여 제작되었으며, 유효 칩 면적은 1070㎛×650㎛ 이고, 3.3V전압에서 230mW의 전력소모를 나타내었다. 입력 주파수 10MHz, 샘플링 주파수 200MHz에서의 INL과 DNL은 ±1LSB 이내로 측정되었으며, SNDR은 43㏈로 측정되었다.

Design of a high performance 32*32-bit multiplier based on novel compound mode logic and sign select booth encoder (새로운 복합 모드 로직과 사인 선택 Booth 인코더를 이용한 고성능 32*32-bit 곱셈기의 설계)

  • Song, Min Gyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.38 no.3
    • /
    • pp.51-51
    • /
    • 2001
  • 본 논문에서는 CMOS 로직과 pass-transistor logic(PTL)의 장점만을 가진 새로운 복합모드로직(Compound Mode Logic)을 제안하였다. 제안된 로직은 VLSI설계에서 중요하게 부각되고 있는 저전력, 고속 동작이 가능하며 실제로 전가산기를 설계하여 측정 한 결과 복합모드 로직의 power-delay 곱은 일반적인 CMOS로직에 비해 약 22% 개선되었다 제안한 복합모드 로직을 이용하여 고성능 32×32-bit 곱셈기를 설계 제작하였다. 본 논문의 곱셈기는 개선된 사인선택(Sign Select) Booth 인코더, 4-2 및 9-2 압축기로 구성된 데이터 압축 블록, 그리고 carry 생성 블록을 분리한 64-bit 조건 합 가산기로 구성되어 있다. 0.6um 1-poly 3-metal CMOS 공정을 이용하여 제작된 32×32-bit 곱셈기는 28,732개의 트랜지스터와 1.59×l.68 ㎜2의 면적을 가졌다. 측정 결과 32×32-bit 곱셈기의 곱셈시간은 9.8㎱ 이었으며, 3.3V 전원 전압에서 186㎽의 전력 소모를 하였다.

Design of fast 16-bit multiplier with $0.35\mu m $ CMOS technology (fullcustom $0.35\mu m $ CMOS 공정을 이용한 16*16 bit 고속 승산기의 설계)

  • 박현규;신현철;김종진
    • Proceedings of the Korea Institute of Convergence Signal Processing
    • /
    • 2000.12a
    • /
    • pp.229-232
    • /
    • 2000
  • 각종 범용 컴퓨터 및 디지탈 신호처리에서 중요한 역할을 하는 16비트 정수형, 2의 보수 형태의 곱셈연산을 수행하기 위한 고속 승산기구조를 설계하고 시뮬레이션 하였다. 부분곱을 합하는 부분은 일반적으로 전체 곱셈기 처리 지연시간의 절반정도를 차지하므로 이 부분의 설계방법이 곱셈기의 궁극적인 속도향상에 직접적인 영향을 미친다. 부분곱의 개수를 줄이기 위하여 Booth encoder를 사용하였고, partial product(부분곱)의 덧셈시간을 줄이기 위하여 4:2 CSA(can save adder)와 3:2 CSA로 CSA tree를 구성 하였으며, 최종결과는 carry look- ahead tree로 얻어진다. Hyundai CMOS 0.35$\mu\textrm{m}$ 1-poly 4-metal 공정으로 layout하여 설계하였으며, 곱셈시간은 2.7ns(tipical case)이하로 측정되었다.

  • PDF

Design of a High Performance 32$\times$32-bit Multiplier Based on Novel Compound Mode Logic and Sign Select Booth Encoder (새로운 복합모드로직과 사인선택 Booth 인코더를 이용한 고성능 32$\times$32-bit 곱셈기의 설계)

  • Kim, Jin-Hwa;Song, Min-Gyu
    • Journal of the Institute of Electronics Engineers of Korea SD
    • /
    • v.38 no.3
    • /
    • pp.205-210
    • /
    • 2001
  • In this paper, a novel compound mode logic based on the advantage of both CMOS logic and pass-transistor logic(PTL) is proposed. From the experimental results, the power-delay products of the compound mode logic is about 22% lower than that of the conventional CMOS logic, when we design a full adder. With the proposed logic, a high performance 32$\times$32-bit multiplier has been fabricated with 0.6um CMOS technology. It is composed of an improved sign select Booth encoder, an efficient data compressor based on the compound mode logic, and a 64-bit conditional sum adder with separated carry generation block. The Proposed 32$\times$32-bit multiplier is composed of 28,732 transistors with an active area of 1.59$\times$1.68 mm2 except for the testing circuits. From the measured results, the multiplication time of the 32$\times$32-bit multiplier is 9.8㎱ at a 3.3V power supply, and it consumes about 186㎽ at 100MHz.

  • PDF