Transcript ㄹㅇㅎㄹㅇㅎㄹㅇㅎㄹㅇㅎㄹㅇㅎㄹ - VADA

MPSoC Clock and Power
Olivier Franza, Intel
• Increased uncertainty with process scaling
– Process, voltage, temperature variations, noise, coupling
• Affects design margin over design, power &
performance loss
– Increased power constraints
– Increasing leakage, power (density, delivery) limitations
• More transistors mean:
– Larger clock distribution networks
– Higher capacitance (more load and parasitics)
• With each new technology:
– Gate delay decreases ~25%
– Wire delay increases ~100%
– Cross-chip communication increases
– Clock needs multiple cycles to cover die
© 조준동, 2006년 가을
1
테크놀로지 스케일링에 따른 저항성분은
증가하고 정전용량은 줄어들지 않는다.
© 조준동, 2006년 가을
2
온칩 버스에서 소모하는 에너지는 전체
에너지의 1/4
© 조준동, 2006년 가을
3
Interconnect Delays & Density
Hannu Tenhunen & Dr. Li-Rong Zheng, Royal Institute of Technology
© 조준동, 2006년 가을
4
Multiple Clocks due to Interconnect limitation
© 조준동, 2006년 가을
5
At reduced performance,
larger resource size
© 조준동, 2006년 가을
6
Multiple clock domains
• Low skew and jitter ALWAYS a must
• Clock modeling requires more accuracy
• Within-die variations, inductance, crosstalk,
electromigration, self-heat, …
• Floor plan modularity
• Think adding/removing cores seamlessly!
• Hierarchical clock partitioning
• Reduce global clock and possibly relax its
requirements
• Generate “locally”-used clock “locally”
• Implement clock domain deskewing techniques
• Bound clock problem into simple, reliable, efficient
domains
© 조준동, 2006년 가을
7
DEC/Compaq Alpha
more complex core to improve performance, more
complex clocks (?), Source: DEC/Compaq – Gronoski & al., JSSC 1998 – Xanthopoulos &
al., ISSCC 2001 – Barroso & al., ISCA 2000
© 조준동, 2006년 가을
8
Clock and Power Convergence
Intel® Itanium® Montecito
• Each core split into 3 clock
domains on variable power
supply
• Each domain controlled by
Digital Frequency Divider (DFD)
generating low-skew variablefrequency clocks; fed by central
PLL and aligned through phase
detectors
• Regional Voltage Detector
(RVD): supply voltage monitor
• Second level clock buffer (SLCB):
digitally controlled delay buffer
for active deskewing
• Regional Active Deskew (RAD):
phase comparators monitoring
and adjusting delay difference
between SLCBs
• Clock Vernier Device (CVD):
digitally controlled delay buffer
© 조준동, 2006년 가을
9
On-Chip Interconnects:
Circuits and Signaling,
Wayne Burleson
• Using Vdd programmability
• High Vdd to devices on critical path
• Low Vdd to devices on non-critical
paths
• VddOff for inactive paths
A – Baseline Fabric
B – Fabric with Vdd Configurable
Interconnect
This work builds on a similar idea for FPGAs described in:
Fei Li, Yan Lin and Lei He. Vdd Programmability to Reduce FPGA Interconnect Power, IEEE/ACM
International
© 조준동,
2006년
Conference on Computer-Aided Design, Nov. 2004
가을
10
From Spaghetti wires to Noc
Marcello Coppola, STMicroelectronics
© 조준동, 2006년 가을
11
Benchmarks,
EE Times,7/2005
• Xpipes, Bologna and Stanford : compared w/
Amba AHB multilayer bus, 21% faster, but
worse latency
• When, Univ. of Kaiserslautern: LPDC decoder:
500Mhz vs 64 Mhz (fixed bus), but 30W vs.
700mW, twice the die size.
• Arteris: better die size, comparable power
consumption, 740Mhz (250Mhz)
• SonicsMX: power-efficient mobile-handset w/
power management
• STNoC, Spidergon: topology w/ degree 2-3
© 조준동, 2006년 가을
12
NoC Applications
http://www.eit.uni-kl.de/wehn
• Turbo-Decoder UMTS compliant, 100Mbit: large
flexibilty w/ 14 parallel units, area = 16.84 mm2
(14mm2 PUs, 2.8mm2 NoC)
• LDPC Decoding,
T. Theocharides, G. Link, N. Chip, T.
Theocharides, G. Link, N. Vijaykrishnan, M. J. Irwin, Int.
Conference on VLSI Design 2005
– 1024 Bit block size, 1.2Gb/s, R=0.75
– NoC: 5x5 2D mesh, dimension-order routing,
large flexibility
– 160nm CMOS Technology, 1.8V, 500 MHz, 110
mm2, ~30 Watt
© 조준동, 2006년 가을
13
Reliable design,
G. De Micheli
1. Manufacturing imperfections: More likely to
happen as lithography scales down
2. Approximations during design: Uncertainty
about details of design
3. Aging: Oxide breakdown,electromigration
4. Environment-induced Soft-errors (Data
corruption due external radiation exposure),
electro-magnetic interference
5. Operating-mode induced: Extremely-low
voltage supply
© 조준동, 2006년 가을
14
Dealing with variability
• Most variability problems that induce timing
errors
1.
2.
3.
4.
Power supply variation
Wire length estimation
Crosstalk
Soft errors
© 조준동, 2006년 가을
15
Adaptive low-power
transmission scheme
Frédéric Worm, Patrick Thiran, Giovanni De Micheli, and Paolo Ienne.
Self-calibrating Networks-on-Chip.In Proceedings of the IEEE International
Symposium on Circuits and Systems, Kobe, Japan, May 2005.
© 조준동, 2006년 가을
16
Reduced Energy Consumption
© 조준동, 2006년 가을
17
Low-Power Network-on-Chip
for High-Performance SoC Design
Lee, K.; Lee, S.-J.; Yoo, H.-J.
Very Large Scale Integration (VLSI) Systems, IEEE Transactions on
Volume 14, Issue 2, Feb. 2006 Page(s):148 - 160
Digital Object Identifier 10.1109/TVLSI.2005.863753
Se-Joong Lee; Kangmin Lee; Seong-Jun Song; Hoi-Jun Yoo
Circuits and Systems II: Express Briefs, IEEE Transactions on [see also Circuits and Systems II:
Analog and Digital Signal Processing, IEEE Transactions on]
Volume 52, Issue 6, June 2005 Page(s):308 - 312
Digital Object Identifier 10.1109/TCSII.2005.848972
성균관대학교 정보통신공학부
© 조준동
2006년 여름
18
Contents
• Introduction
• NoC Architecture
– Overall Architecture
– Packet Routing Scheme
– Chip-to-Chip Connectivity
– Topology Selection
– Physical Transfer Unit Size
– Hierarchical Circuit/Packet Switching
– Synchronization
– NoC Protocol
• Low-Power Techniques
– Low-Swing Signal
– Mux-Tree Based Round-Robin Scheduler
– Crossbar Partial Activation Technique
– Low-Energy Coding On-Chip Serial Link
• Implementation and Measurement Result
• Conclusion
© 조준동, 2006년 가을
19
Introduction
• System-on-Chip (SoC)
– More than a billion transistors are integrated on a single chip.[1.
1]
– Wire delays have become more critical[1. 2]
– The synchronization problem Heterogeneous NoC architecture
• The clock frequencies increase
• The feature sizes decrease
• NoC Solution
– How to interconnect efficiently
• Focus
– Performance and Scalability
© 조준동, 2006년 가을
20
NoC Architecture
• Overall Architecture
– Essential Part of NoC
•
•
•
•
•
•
•
Network0 Interface (NI)
Up-Sampler (UPS)
Link Wires
FIFO Synchronizer with a queuing buffer (SYNC)
Switch
Overall architecture of the On Chip Network
Down-Sampler (DNS)
Off-chip Gateway (OGW)
© 조준동, 2006년 가을
21
NoC Architecture
• Packet Routing Scheme
– Routing Process
• A packet is transferred to a destination according to route information in the
packet header.
Switch port index, Header Format, Header modification
© 조준동, 2006년 가을
22
NoC Architecture
• Chip-to-Chip Connectivity
– Off-chip Gateway(OGW)
• OGWs provide chip-to-chip packet transaction
Chip-to-Chip connection using OGWs
© 조준동, 2006년 가을
23
NoC Architecture
• Topology Selection
– The first step for NoC architecture design
– There are basic topologies
• Mesh Topology
– Mesh topology widely used and studied for parallel computing architecture.
• Star Topology
– Star topology has not been popularly used because it has a limitation of scalability
– Optimal to SoC design( PUs may be placed irregularly to minimize chip area)
Mesh area  (5 N  4 N ) ASYNC
 (25 N  36 N ) A1SW  4( N  N ) ALNK
Star area  NASYNC  N 2 A1SW  N N ART
the number of PUs = N
© 조준동, 2006년 가을
24
NoC Architecture
• Topology Selection
Energy consumption according to a number of PUs
© 조준동, 2006년 가을
25
NoC Architecture
• Topology Selection
Network area according to a number of PUs
– The network area cost
including the
area of…
•
•
•
•
Switches
Multiplexers
Demultiplexers
links
© 조준동, 2006년 가을
26
NoC Architecture
• Physical Transfer Unit(PHIT) Size
– A packet is divided and transmitted through the core network.
Serialization ratio (SERR) = packet size / phit size
Energy and Area of OCN according to SERR
© 조준동, 2006년 가을
27
NoC Architecture
• Hierarchical Circuit and Packet Switching
– Local Intracluster Network
→ Circuit Switching
• The Circuit switching does not need packet buffers
• Area and Power consumption can be reduced
– Global Intercluster Network
→ Packet Switching
• The global intercluster traffic shares the bandwidth of the
switch-to-switch
link
• The throughput of the shared and limited link is more important
rather than
the latency
© 조준동, 2006년 가을
28
NoC Architecture
• Synchronization
– Heterogeneous Multiprocessing System(multi timing reference)
Synchronization structure in the NoC
© 조준동, 2006년 가을
29
NoC Architecture
• NoC Protocol
Packet Format, Burst READ/WRITE Transactions
© 조준동, 2006년 가을
30
Low-Power Techniques
• Low-Swing Signaling
– The global link consumes higher power than local
link does
Low-swing signaling and its transceiver circuits
– Low-signaling
can alleviate its energy
consumption significantly[2. 13]
© 조준동, 2006년 가을
31
Low-Power Techniques
• Low-Swing Signaling
(a)Energy consumption (b)Energy and delay product
© 조준동, 2006년 가을
32
Low-Power Techniques
• Mux-Tree Based Round-Robin Scheduler
Mux-tree-based round-robin scheduler
© 조준동, 2006년 가을
33
Low-Power Techniques
• Crossbar Partial Activation Techniques
(CPAT)[2.
10]
Schematic Diagram of Crossbar
© 조준동, 2006년 가을
34
Low-Power Techniques
• Low-Energy Coding on On-Chip Serial Link
transitions in parallel with serial communications
© 조준동, 2006년 가을
35
Low-Power Techniques
• Low-Energy Coding on On-Chip Serial Link
– Serialized low-energy transmission (SILENT)
technique[2. 18]
• To minimize the transmission energy on the serial wire
t)
B (by
[i ] using
 b (t ) [i ]  b (t 1)the
[i ], data
for i correlation
0  (n  1) properties
– The
is expressed
as follows:
B (t ) [n encoding
 1: 0] : n  bit algorithm
original data word
at time t
b (t ) [n  1: 0] : n  bit encoded data word at time t
Encoder/Decoder
© 조준동, 2006년 가을
36
Low-Power Techniques
• Low-Energy Coding on On-Chip Serial Link
– Serialized low-energy transmission (SILENT)
Average power consumption on18]
serial communications
technique[2.
© 조준동, 2006년 가을
37
Implementation and
Measurement Results
• Implemented multimedia SoC
– proposed NoC architecture, protocol and lowBlock
diagram of atechniques
prototype SoC
power
© 조준동, 2006년 가을
38
Implementation and
Measurement Results
• Implemented multimedia SoC
– proposed NoC architecture, protocol and lowpower techniques
Measured packet signals
On-chip network power consumption
© 조준동, 2006년 가을
39
Conclusion
• A low-power NoC is designed and
implemented for high-performance SoC
application
• Heterogeneous IPs are interconnected in a
hierarchical star topology
• Various power-efficient techniques were
suggested and implemented
© 조준동, 2006년 가을
40
Reference
•
•
•
•
•
•
•
•
•
•
[1. 1] R.Woo et al., “A 210 mW graphics LSI implementing full 3 D pipeline with 264
Mtexels/s texturing for mobile multimedia applications,” in ISSCC Tech. Dig., 2003, pp. 44–
45.
[1. 2] AMBA™ AXI Protocol Specification (2003). [Online]. Available: http://www.arm.com
[1. 3] M. Sgroi et al., “Addressing the system-on-a-chip interconnection woes through
communication-based design, ”in Proc. Design Automation Conf., 2001, pp. 667–672.
[1. 4] L. Benini et al., “Powering networks on chips,” in Proc. Int. Symp.System Synthesis,
2001, pp. 33–38.
[1. 5] P. Guerrier et al., “A generic architecture for on-chip packet-switched
interconnections,” in Proc. Design Automation Test Eur. Conf. Exhib.,2000, pp. 250–256.
[1. 6] S. Kumar et al., “A network on chip architecture and design methodology,”in Proc.
Ann Symp. VLSI, 2002, pp. 117–124.
[1. 7] W. J. Dally et al., “Route packets, not wires: On-chip interconnection networks,” in
Proc. f Design Automation Conf., 2001, pp. 684–689.
[1. 8] H. Zhang et al., “A 1-V heterogeneous reconfigurable DSP IC for wireless baseband
digital signal processing,” J. Solid-State Circuits, vol. 35,no. 11, pp. 1697–2000, Nov. 2000.
[1. 9] M. Taylor et al., “A 16-issue multiple-program-counter microprocessor with point-topoint scalar operand network,” in ISSCC Tech. Dig., 2003,pp. 170–171.
[1. 10] S. Lee et al., “An 800 MHz star-connected on-chip network for application to
systems on a chip,” in ISSCC Tech. Dig., 2003, pp. 468–469.
© 조준동, 2006년 가을
41
Reference
•
•
•
•
•
•
•
•
•
•
•
[1. 11] K. Lee et al., “A 51 mW 1.6 GHz on-chip network for low power heterogeneous SoC
platform,” in ISSCC Tech. Dig., 2004, pp. 152–153.
[1. 12] A. Laffely et al., “Adaptive system on a chip (ASOC): A backbone for power-aware
signal processing cores,” in Proc. IEEE Int. Conf. Image Processing, Barcelona, Spain, Sep.
2003.
[1. 13] E. Rijpkema et al., “Tradeoffs in the design of a router with both guaranteed and
best-effort services for networks on chip,” in Proc. Design Automation Test in Eur., Mar.
2003, pp. 350–355.
[1. 14] D. Bertozzi et al., “Xpipes:Anetwork-on-chip architecture for gigascale system-onchip,” IEEE Circuits Syst. Mag., vol. 4, no. 2, pp. 18–31, Feb. 2004.
[1. 15] S. Kimura et al., “An on-chip high speed serial communication method based on
independent ring oscillators,” in ISSCC Tech. Dig., 2003, pp. 390–391.
[1. 16] H. J. Siegel et al., Interconnection Networks for Large-Scale Parallel Processing:
Theory and Case Studies. New York: McGraw-Hill, 1990.
[1. 17] K. Lee et al., “SILENT: Serialized low energy transmission coding for on-chip
interconnection networks,” in Proc. Int. Conf. Computer-Aided Design, 2004, pp. 448–451.
[1. 18] W. J. Dally et al., Digital Systems Engineering. Cambridge, U.K.: Cambridge Univ.
Press, 1998, ch. 10.
[1. 19] M. Cooperman et al., “CMOS gigabit-per-second switching,” J. Solid- State Circuits,
vol. 28, no. 6, pp. 631–639, Jun. 1993.
[1. 20] T. T. Ye et al., “Analysis of power consumption on switch fabrics in network routers,”
in Proc. Design Automation Conf., 2002, pp. 524–529.
[1. 21] K. Lee et al., “A distributed on-chip crossbar switch scheduler for on-chip networks,”
in Proc. Custom Integrated Circuits Conf., May 2003, pp. 671–674.
© 조준동, 2006년 가을
42
Reference
•
•
•
•
•
•
•
•
•
•
•
•
[2. 1] International Technology Roadmap for Semiconductors [Online]. Available:
http://public.itrs.net
[2. 2] W. Dally et al., “Route packets, not wires: On-chip interconnection networks,”in Proc.
Des. Autom. Conf., Jun. 2001, pp. 684–689.
[2. 3] L. Benini et al., “Networks on chips: A new SoC paradigm,” IEEE Computer, vol. 36,
no. 1, pp. 70–78, Jan. 2002.
[2. 4] D. Bertozzi et al., “Xpipes: A network-on-chip architecture for gigascale system-onchip,” IEEE Circuits Syst. Mag., vol. 4, no. 2, pp. 18–31, 2004.
[2. 5] E. Rijpkema et al., “Trade offs in the design of a router with both guaranteed and
best-effort services for networks on chip,” in Proc. Des., Autom. Test Europe Conf., Mar.
2003, pp. 350–355.
[2. 6] V. Nollet et al., “Operating-system controlled network on chip,” in Proc. Des. Autom.
Conf., Jun. 2004, pp. 256–259.
[2. 7] J.-S. Kim et al., “On-chip network based embedded core testing,” in Proc. IEEE Int.
SoC Conf., Sep. 2004, pp. 223–226.
[2. 8] S.-J. Lee et al., “An 800 MHz star-connected on-chip network for application to
systems on a chip,” in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, Feb. 2003, pp.
468–469.
[2. 9] M. Taylor et al., “A 16-issue multiple-program-counter microprocessor with point-topoint scalar operand network,” in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, Feb.
2003, pp. 170–171.
[2. 10] K. Lee et al., “A 51 mW 1.6 GHz on-chip network for low-power heterogeneous SoC
platform,” in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, Feb. 2004, pp. 152–153.
[2. 11] H. Wang et al., “A technology-aware and energy-oriented topology exploration for
on-chip networks,” in Proc. Des., Autom. Test Europe Conf., Mar. 2005, pp. 1238–1243.
[2. 12] BONE: Network-on-Chip Protocol [Online]. Available: http://ssl.kaist.ac.kr/ocn
© 조준동, 2006년 가을
43
Reference
•
•
•
•
•
•
•
•
•
[2. 13] R. Ho et al., “Efficient on-chip global interconnects,” in IEEE Symp. VLSI Circuits
Dig. Tech. Papers, Jun. 2003, pp. 271–274.
[2. 14] C. Svensson, “Optimum voltage swing on on-chip and off-chip interconnect,”IEEE J.
Solid-State Circuits, vol. 36, no. 7, pp. 1108–1112, Jul. 2001.
[2. 15] P. Gupta et al., “Design and implementing a fast crossbar scheduler,”IEEE Micro, vol.
19, no. 1, pp. 20–28, Jan./Feb. 1999.
[2. 16] E. Shin et al., “Round-robin arbiter design and generation,” in Proc. IEEE Int. Symp.
Syst. Synthesis, Oct. 2002, pp. 243–248.
[2. 17] P. Landman et al., “Architectural power analysis: The dual bit type method,”IEEE
Trans. Very Large Scale Integr. (VLSI) Syst., vol. 3, no. 2, pp. 173–187, Jun. 1995.
[2. 18] K. Lee et al., “SILENT: Serialized low-energy transmission coding for on-chip
interconnection networks,” in IEEE Int. Conf. Comput.-Aided Des. Dig. Tech. Papers, Nov.
2004, pp. 448–451.
[2. 19] R. Woo et al., “A 210-mW graphics lsi implementing full 3-D pipeline with 264
Mtexels/s texturing for mobile multimedia applications,” IEEE J. Solid-State Circuits, vol.
39, no. 2, pp. 358–367, Feb. 2004.
[2. 20] C. Kretzschmar et al., “Why transition coding for power minimization of on-chip
buses does not work,” in Proc. Des. Autom. Test Europe Conf. (DATE), Feb. 2004, pp. 512–
517.
[2. 21] M. R. Stan et al., “Bus-invert coding for low-power I/O,” IEEE Trans. Very Large
Scale Integr. (VLSI) Syst., vol. 3, no. 1, pp. 49–58, Mar. 1995.
© 조준동, 2006년 가을
44