TY - GEN
T1 - 3D System Integration in a Package for Artificial Intelligence
AU - Kuroda, Tadahiro
N1 - Funding Information:
The author is grateful to M. Motomura for discussion. This work was supported by JST ACCEL Grant Number JPMJAC1502, Japan.
Publisher Copyright:
© 2019 IEEE.
PY - 2019/3
Y1 - 2019/3
N2 - The connection problem of the early large-scale computers, the tyranny of numbers, was solved by the invention of the von Neumann computer, integrated circuit and solderless connection (Fig.1). However, the advance that has been achieved in all these technologies has made them the victim of their own success, with Moore's Law scaling and connector miniaturization each reaching its limits, and von Neumann bottleneck exacerbating. Furthermore, the adoption of neural network and deep learning for Artificial Intelligence (AI) has once again brought to the forefront the challenges of scale and wiring in building hardwired computers. The solutions can be found in near-field coupling integration technologies - ThruChip Interface (TCI) [1]- [26] and Transmission Line Coupler (TLC) [27]-[36]. TCI, a magnetic coupling technology, enables stacking DRAMs with an SoC to alleviate the von Neumann bottleneck. The same technology also enables stacking SRAMs with a neural network SoC to solve the challenges of scale and wiring in an AI computer. On the other hand, TLC, an electromagnetic coupling technology, enables a contactless connector that overcomes the scaling limits of its electro-mechanical counterpart.
AB - The connection problem of the early large-scale computers, the tyranny of numbers, was solved by the invention of the von Neumann computer, integrated circuit and solderless connection (Fig.1). However, the advance that has been achieved in all these technologies has made them the victim of their own success, with Moore's Law scaling and connector miniaturization each reaching its limits, and von Neumann bottleneck exacerbating. Furthermore, the adoption of neural network and deep learning for Artificial Intelligence (AI) has once again brought to the forefront the challenges of scale and wiring in building hardwired computers. The solutions can be found in near-field coupling integration technologies - ThruChip Interface (TCI) [1]- [26] and Transmission Line Coupler (TLC) [27]-[36]. TCI, a magnetic coupling technology, enables stacking DRAMs with an SoC to alleviate the von Neumann bottleneck. The same technology also enables stacking SRAMs with a neural network SoC to solve the challenges of scale and wiring in an AI computer. On the other hand, TLC, an electromagnetic coupling technology, enables a contactless connector that overcomes the scaling limits of its electro-mechanical counterpart.
UR - http://www.scopus.com/inward/record.url?scp=85067803988&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85067803988&partnerID=8YFLogxK
U2 - 10.1109/EDTM.2019.8731232
DO - 10.1109/EDTM.2019.8731232
M3 - Conference contribution
AN - SCOPUS:85067803988
T3 - 2019 Electron Devices Technology and Manufacturing Conference, EDTM 2019
SP - 80
EP - 81
BT - 2019 Electron Devices Technology and Manufacturing Conference, EDTM 2019
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2019 Electron Devices Technology and Manufacturing Conference, EDTM 2019
Y2 - 12 March 2019 through 15 March 2019
ER -