3D Wireless Network‐on‐Chip Architecture with interlayer cooling

3D Wireless Network‐on‐Chip Architecture with interlayer cooling
‐‐Md Shahriar Shamim & Fernando Cueva
5/6/2015
1
Overview
• Introduction
• Multi‐core chip
• Network‐on‐Chip
• 3D NoC
• Thermal Challenges
• Why 3D Wireless?
• Proposed Architecture
• Results
• Conclusion
5/6/2015
2
Multi‐core Chips: A Necessity
• Need for explosive computational power
• Consumer/Entertainment Application
• Scientific Application
• Increasing clock frequency is not possible as it increases power dissipation
Solution: Core level Parallelism,
distribute tasks to multiple cores 5/6/2015
3
Challenges: Interconnection of Cores
• Traditional Interconnect
architectures are not scalable
• Delay limit number of cores
Solution: Scalable interconnect infrastructure for communication
5/6/2015
4
Network‐on‐Chip (NoC)
• Packet based on‐chip network
• Route packets, not wires –Bill Dally, 2000.
• Dedicated infrastructure for data transport
• Decoupling of functionality from communication
• A plug‐and‐play network independent of the cores
High-performance
ARM processor
High-bandwidth
memory interface
High-bandwidth
ARM processor
AHB
B
R
I
D
G
E
Timer
UART
APB
Keypad
PIO
DMA Bus
master
AMBA bus: ARM
NoC infrastructure
Multiple publications in IEEE ISSCC, 2010 from Intel, IBM, AMD, and Sun
Microsystems show that multi-core NoC is a reality
5/6/2015
5
5
Problem with Traditional wire Interconnect
• Limitation of Wireline Interconnect
• Multi‐hop wireline communication
• High Latency and energy dissipation source
destination
-core
-NoC interface
-NoC switch
80% of chip power will be from on-chip interconnects in
the next 5 years – ITRS, 2007
5/6/2015
6
6
Emerging Interconnect Technologies
Goal:
High Bandwidth + Low Energy Dissipation
Three Dimensional
Integration
Wireless
Interconnects
Optical Interconnects
5/6/2015
7
7
3D Integration
• Stacking multiple active layers
• Heterogeneous integration
• Higher connectivity & less hop count High
bandwidth
• Shorter average path length  Lower Power
Challenges
• High power densities
• Thermal issues
• High temperatures
• Hotspots
• Limited ability to extract heat only from top or bottom
layer
•
Pavlidis et al., “3-D topologies for Networks-on-Chip”, IEEE Transactions on Very Large Scale
Integration (TVLSI), 2007.
5/6/2015
8
8
Micro‐channel based Sophisticated Cooling Layer
• Microchannels between active layers circulating with chilled fluids
• What about TSVs?
•
•
•
pumping liquid can cause extreme pressure drops across the cooling layer ‐‐> structural instability.
Complex manufacturing process as the TSVs and micro‐channels will co‐
exist between the cooling layer.
Longer TSVs ‐‐> Higher delay and power dissipation SABRY et al. Energy-Efficient Multiobjective Thermal Control for Liquid-Cooled 3-D Stacked
Architectures. Trans. Comp.-Aided Des. Integ. Cir. Sys. Vol:30, Issue:12, page(s): 1883-1896
5/6/2015
9
What should we do?
Can incorporating another emerging interconnect technology fix these problems?? Maybe Photonic???Wireless?
What about wireless?? It does not need any physical interconnect layout. Hmm, it can alleviate the height limitation of the cooling layer.. 5/6/2015
But Photonic interconnect also requires dedicated physical layout like TSVs..
10
10
Wireless Interconnect
• Use of on‐chip wireless links
• Single Hop Shortcut
• Reduce latency and energy dissipation in communication
• No physical interconnect layout is necessary
• Wireless port/wireless interface (WI) consists of transceiver and antenna
• Antenna Technology:
• Metal zigzag antennas (mm‐wave) are CMOS compatible
J. Lin et al., “Communication Using Antennas Fabricated in Silicon Integrated Circuits,” IEEE Journal of
Solid-State Circuits, vol. 42, no. 8, August 2007, pp. 1678-1687.
5/6/2015
11
Proposed Architecture
• Hierarchical in nature
• Two level Hierarchy
• Bottom layer
• Mesh Connectivity
• Upper layer
• Switches grouped into subnets
• One hub per subnet
• All switches from one subnet connected to the hub from that subnet
• One wireless per subnet
• Wireless interconnected with each other in all‐to‐all fashion
5/6/2015
12
Performance Evaluation of 3D Wireless NoC
CANNEAL
• 4 layers
• 8 wireless routers
• 2 in each layer
• 2 MAC protocols
• Token based
• CDMA
Peak temperature (ºC)
• 64 cores
FFT
LU
RADIX
BODYTRACK
80
70
60
50
40
30
20
10
0
3D‐Mesh‐TSV
CDMA based 3D‐HiWiNoC
Fig. Peak temperature in presence of real
application traffics
• Lower temperatures for several benchmarks
Lower packet energy, lower temperature, comparable bandwidth
5/6/2015
13
Performance Evaluation of 3D Wireless NoC
• What about bandwidth and energy?
BW(3D‐Mesh_TSV)
BW(CDMA based 3D‐HiWiNoC)
Packet Energy(3D‐Mess‐TSV)
Packet Energy(CDMA based 3D‐HiWiNoC)
Normalized Peak Bandwidth 1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
Normalized Packet Energy 1.2
1.2
0
0
CANNEAL
FFT
LU
RADIX
BODYTRACK
Fig. Peak temperature in presence of real
application traffics
Lower packet energy, lower temperature,
But bandwidth is also reducing. Why??
5/6/2015
14
Why 3D wireless NoCs with interlayer cooling suffers from bandwidth degradation?
• Number of active links in wireless architecture is less than 3D wireline mesh in order to accommodate micro‐channel liquid cooling layer.
• In 64 core 4 layer system,16 TSV based links connecting the vertically adjacent switches across the cooling layer is eliminated. • Results in a loss of an aggregate bisection bandwidth of 1.2Tbps.
• Wireless bandwidth is only 16GBps. Limitation???
5/6/2015
15
Conclusion
• Interlayer Wireless Interconnects
• Eliminate TSVs across the cooling layer
• Make cooling layers modular in design
• Improves pressure drops and thermal efficiencies
• Improvement in peak temperature reduction and energy efficient.
• However, suffers from Bandwidth degradation.
• Requires performance evaluation of interlayer communication
5/6/2015
Finally!!!
16
Questions???
5/6/2015
17