2019
DOI: 10.1147/jrd.2019.2940427
|View full text |Cite
|
Sign up to set email alerts
|

Silicon interconnect fabric: A versatile heterogeneous integration platform for AI systems

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1

Citation Types

0
3
0

Year Published

2021
2021
2024
2024

Publication Types

Select...
4
3

Relationship

0
7

Authors

Journals

citations
Cited by 33 publications
(3 citation statements)
references
References 34 publications
0
3
0
Order By: Relevance
“…This new platform is adapted from the silicon interconnect fabric (Si-IF) developed at UCLA [19][20][21][22]. The Si-IF is an advanced packaging technology that owns characteristics, including system-on-silicon-wafer, fine pitch ( 10 μm), short inter-dielet spacing ( 100 μm), metal-to-metal bonding, no solder and underfill, high bandwidth, low energy consumption per bit and low latency.…”
Section: Assembly Methodsmentioning
confidence: 99%
“…This new platform is adapted from the silicon interconnect fabric (Si-IF) developed at UCLA [19][20][21][22]. The Si-IF is an advanced packaging technology that owns characteristics, including system-on-silicon-wafer, fine pitch ( 10 μm), short inter-dielet spacing ( 100 μm), metal-to-metal bonding, no solder and underfill, high bandwidth, low energy consumption per bit and low latency.…”
Section: Assembly Methodsmentioning
confidence: 99%
“…Wit h t he capabi l it y for h ig hthroughput chip-to-chip data transmission, chiplet technology provides the potential for incorporating hardware accelerators for machine learning, AI, or neuromorphic computing. 5,6 Chiplets enable the hardware accelerators to be designed and manufactured using ideal and custom technologies while still maintaining low-latency communication with the central processor. Research in this area is active with universities and industry groups, who report chiplet-based accelerators for machine learning and signal processing, taking advantage of interchip data rates exceeding 1 Tb/s.…”
Section: Microelectronicsmentioning
confidence: 99%
“…[1,2,5,6] Recently, there has been interest in additive "direct write" approaches as an alternative method of nanofabrication with two significant opportunities for their utilization.The first opportunity is in HI: creating dense interconnects between many chiplets with diverse functions and geometries that are integrated on large panels or packages, often organic substrates. [3,28,29] Chiplet integration addresses the needs for energy-efficient computing and the need for fast access to lots of memory. [3,29] The area of a single chip is ultimately limited by the reticle size during lithography, hence the integration of multiple chiplets is necessary.…”
mentioning
confidence: 99%