國立虎尾科技大學 |

A Low Power Time-Domain Computing-In-Memory Micro for Binary Neural Networks.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	A Low Power Time-Domain Computing-In-Memory Micro for Binary Neural Networks./
Author:	He, Yixuan.
Description:	1 online resource (103 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 84-11, Section: B.
Contained By:	Dissertations Abstracts International84-11B.
Subject:	Electrical engineering. -
Online resource:	click for full text (PQDT)
ISBN:	9798379475161

A Low Power Time-Domain Computing-In-Memory Micro for Binary Neural Networks.
He, Yixuan.

A Low Power Time-Domain Computing-In-Memory Micro for Binary Neural Networks. - 1 online resource (103 pages)

Source: Dissertations Abstracts International, Volume: 84-11, Section: B.

Thesis (Ph.D.)--Northeastern University, 2023.

Includes bibliographical references

In order to enable hardware to match complex deep learning algorithms, binary neural networks are developed and considered as hardware-friendly deep learning algorithms due to the fact that it can achieve similar inference accuracy with fewer computing resources compared with traditional convolutional neural networks. As for its VLSI implementations, the computing-in-memory (CIM) technology has been proved to solve the memory-wall bottleneck problem shown in traditional von Neumann machine and can be a perfect choice to implement neural networks with binary data. Therefore, this work proposes two novel computing-in-memory cores that implement XNOR-and-accumulate of binary neural networks with all-digital elements. Both designs use 8T-SRAM cells to perform XNOR operations inside memory array, but they accumulate the related XNOR output values in diverse ways. The time-domain CIM micro does accumulation in time domain with specialized racing structures and delay lines, whereas the all-digital micro uses low-power pathfinding counters in pure digital domain. The circuits are built and simulated in Cadence using Samsung 65nm CMOS technology with 1V power supply. The results show correct functionality and lower power consumption compared to other existing micros. The time-domain micro achieves 2730 GOPS throughput and 431.7 TOPS/W power efficiency. And the all-digital micro has 1638 GOPS throughput and 537.6 TOPS/W power efficiency. With further exploration, the proposed time-domain and all-digital pathfinding approaches can become new favorites in the field of in-memory-computing for deep learning because they have their own advantages in terms of throughput, power efficiency compared to mixed-signal approaches and traditional digital methods.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798379475161Subjects--Topical Terms:

596380
Electrical engineering.
Subjects--Index Terms:

Binary neural networksIndex Terms--Genre/Form:

554714
Electronic books.

A Low Power Time-Domain Computing-In-Memory Micro for Binary Neural Networks.
LDR:03154ntm a22003977 4500 001 1141722
005 20240414211456.5
006 m o d
007 cr mn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798379475161
035 $a (MiAaPQ)AAI30423111
035 $a AAI30423111
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a He, Yixuan. $3 1465672
245 1 2 $a A Low Power Time-Domain Computing-In-Memory Micro for Binary Neural Networks.
264 0 $c 2023
300 $a 1 online resource (103 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 84-11, Section: B.
500 $a Advisor: Kim, Yong-Bin.
502 $a Thesis (Ph.D.)--Northeastern University, 2023.
504 $a Includes bibliographical references
520 $a In order to enable hardware to match complex deep learning algorithms, binary neural networks are developed and considered as hardware-friendly deep learning algorithms due to the fact that it can achieve similar inference accuracy with fewer computing resources compared with traditional convolutional neural networks. As for its VLSI implementations, the computing-in-memory (CIM) technology has been proved to solve the memory-wall bottleneck problem shown in traditional von Neumann machine and can be a perfect choice to implement neural networks with binary data. Therefore, this work proposes two novel computing-in-memory cores that implement XNOR-and-accumulate of binary neural networks with all-digital elements. Both designs use 8T-SRAM cells to perform XNOR operations inside memory array, but they accumulate the related XNOR output values in diverse ways. The time-domain CIM micro does accumulation in time domain with specialized racing structures and delay lines, whereas the all-digital micro uses low-power pathfinding counters in pure digital domain. The circuits are built and simulated in Cadence using Samsung 65nm CMOS technology with 1V power supply. The results show correct functionality and lower power consumption compared to other existing micros. The time-domain micro achieves 2730 GOPS throughput and 431.7 TOPS/W power efficiency. And the all-digital micro has 1638 GOPS throughput and 537.6 TOPS/W power efficiency. With further exploration, the proposed time-domain and all-digital pathfinding approaches can become new favorites in the field of in-memory-computing for deep learning because they have their own advantages in terms of throughput, power efficiency compared to mixed-signal approaches and traditional digital methods.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Electrical engineering. $3 596380
650 4 $a Computer engineering. $3 569006
650 4 $a Information technology. $3 559429
653 $a Binary neural networks
653 $a Computing-in-memory
653 $a Low power VLSI
653 $a Time-domain computing
653 $a Deep learning
655 7 $a Electronic books. $2 local $3 554714
690 $a 0544
690 $a 0489
690 $a 0464
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Northeastern University. $b Electrical and Computer Engineering. $3 1182193
773 0 $t Dissertations Abstracts International $g 84-11B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30423111 $z click for full text (PQDT)