Ammar Karkar 1 , Kenneth Tong 2 , Raed Al-Dujaily 1 , Alex Yakovlev 1 - - PowerPoint PPT Presentation
Ammar Karkar 1 , Kenneth Tong 2 , Raed Al-Dujaily 1 , Alex Yakovlev 1 - - PowerPoint PPT Presentation
Ammar Karkar 1 , Kenneth Tong 2 , Raed Al-Dujaily 1 , Alex Yakovlev 1 and Terrence Mak 3 (1)Newcastle University, (2) University College London, (3)The Chinese University of Hong Kong Introduction Surface wave Hybrid architecture
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 2
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 3
Scalability is the issue:
- System-on-chip
- Network-on-chip
- Global communication
- Alternative communication fabric
10/12/2012 4
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 5
Why surface wave :
- Lower cost to implement:
Does not require non-CMOS devices to be integrated (e.g. Optical Interconnect) less industrial challenges(e.g. 3D technology)
- Consume less power than wireless RF
- Provide one-to-many communication unlike(RF
waveguide transmission lines)
10/12/2012 6
Surface wave (less power) (one-to-many) Wireless (one-to-many) Waveguide (less power)
Off-chip seems to be the better option for three
reasons:
- lower implementation cost and mitigate area overhead
- used for both inter/intra-chip communication
- propagation speed, close to the speed of light (air as
dielectric material)
10/12/2012 7
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 8
Shared media and limited number of frequency channels wire based local communication:
- scale well with technology
- cheapest implementation cost
Hybrid multilayer Network:
10/12/2012 9
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 10
Proposed surface acts as wave guide of the
propagated signal:
- 𝑊+ 𝑒 = 𝑊+ 0 𝑓−𝛽𝑒 (1)
- 𝑇21 = 𝐹 + 20 log 𝑓−𝛽𝑒 (2)
10/12/2012 11
α ≈ 6.33 E ≈ -23.8
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 12
NoC component Baseline Architecture SWI Hybrid Architecture (proposed) RF-I with transmission line Router (mm2) 1.08533 1.51237 1.51237 RF circuit (mm2)
- 0.408
0.463 Link (mm2)
- 0.17152
active area overhead rate to baseline arch.(% of total die) 2.29% 2.3%
10/12/2012 13
Area overhead consideration for the proposed Interconnect fabric
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 14
10/12/2012 15
6×4 Network average delay vs. PIR for Hybrid and Baseline Architecture 6×4 Network Throughput vs. PIR for Hybrid and Baseline Architecture
10/12/2012 16
- Hybrid-Arch. PIR and throughput improvement over
Baseline Arch. At the edge of Network Saturation
5 10 15 20 25 30 35 40 45 50 Random Transpose Butterfly Shuffle Hotspot1 Hotspot2 Bitreversal Improvem vement ent(% (%) Traffic ic PIR Throughput
10/12/2012 17
Normalized Power consumption vs. Baseline architecture for
Different NoC size, traffic
Introduction Surface wave Hybrid architecture Analysis of link power dissipation Area estimation Preliminary results Conclusion
10/12/2012 18
Scalability issues in global communication Hybrid architecture (metal and SWI for local and
global Communication, respectively)
Significant potential of the proposed architecture
to mitigate these issues with relatively small area penalty
Future work includes developing an optimized
topology on design time or on the fly, as well as, investigating 1-to-M/M-to-1 traffic pattern for this fabric.
10/12/2012 19
10/12/2012 20