Slow-Light Optical Delay Lines and Buffers
Rod Tucker
92-Terabit/s Throughput Router
May 25, 2004
Cisco Introduces
92-Terabit Router
―Long-awaited
carrier-class router
promises
continuous uptime,
higher throughput
and easier
provisioning for
service providers.‖
Cisco CRS-1
Line rate: 40 Gbit/s, Up to 2,300 ports
Model with 1.2-Tbps capacity: US$ 450,000
Model with 92-Tbps capacity: US$ 30 million
Cisco CRS-1
Up to 40 Gbit/s
Up to 2300 ports
Line cards
Buffer capacity: 10 Gbit/port (~ 250 ms delay)
Buffer capacity: < 1 Kbit/port (~ 1 ms delay)
Optical Packet Switch
Key Bottleneck:
Optical Buffer
DEMUX Packet Header Buffering & MUX
Synchronization Replacement Routing
Optics
Electronics
Controller
Optical Buffer Structures
Variable Delay Line
Delay
Delay Line
Recirculating
Loop
Cross Point
Staggered
Delay Line
Delay Line Delay Line
Cross Point Cross Point
Comparison of Buffer Technologies
Total buffer capacity of 104 Gigabits
~ 103 RAM chips
< US$ 100k in cost
< 1 kW power dissipation
Cisco CRS-1 with 1000 ports,
250 ms buffering per port
Total fibre length = 40 Gm
150 times distance from Earth to Moon!
.
Photonic packet switch with 1000 ports,
250 ms buffering per port
using optical fibre delay lines
Optical Buffering
Appears to be impractical for real applications
However…
Scientists put
a light wave
on hold
Jan. 18, 2001 — Physicists say they can effectively
catch a light pulse in a bottle, hold onto it and release
it, in an operation described as slowing light to a dead
stop. It’s actually the information about the light wave
that’s being captured, the researchers say, and such
techniques could be applied to a future generation of
quantum computers and ultrasecure communication
devices.
Volume 285 Number 1 July 2001 pp.66-73
Frozen Light
Slowing a beam of light to a halt may pave the way for new optical communications technology.
By Lene Vestergaard Hau
FREEZING OF LIGHT
begins with a process
in which a carefully
tuned laser beam
renders an opaque
material transparent to
a second laser beam.
UC Berkeley Press Release
Researchers use semiconductors to set speed limit on light
By Sarah Yang, Media Relations | 28 September 2004
BERKELEY – In a nod to scientific paradox, researchers at the University of California, Berkeley, have
slowed light down in an effort to speed up network communication.
They have shown for the first time that the group velocity of light - the speed at which a laser pulse travels
along a light wave - can be slowed to about 6 miles per second in semiconductors. While that speed is not
exactly the pace of a turtle, it is 31,000 times slower than the 186,000 miles (or 300 million meters) per
second that light normally clocks while traveling through a vacuum.
Summary
• Sabbatical at Berkeley
• Explore the capabilities of slow light buffers
- Demystify
- Cut through the hype
• Fundamental Limitations and Capabilities of Slow Light
- Electromagnetically-induced transparency
- Photonic Crystals and Microresonators
- Applications:
- Optical packet buffering
- Optical signal processing
Optical Buffer
Control
Packet length, tpacket
Bit period, tbit
Data in Data out
Buffer
Hold-off time, THO Storage time, TS
Data Bandwidth
1
B packet
tbit
Basic Variable Delay Line
L
z (t)
Control
Input Output
Delay Line
x
dx
Group velocity: vg ( x, t )
dt
Group Velocity Change
Refractive index n 1 Refractive index n 2
Waveguide 1 Waveguide 2
Input Output
Optical frequency
Speed of light in air
c
Group velocity: v gi
k i dni
ni
Propagation constant d
dn2
v g1 n2
Slow-down factor: S ( d n g 2 Group
dn1 indices
vg2 n g1
n1
d
Group Velocity Change
Waveguide 1 Waveguide 2
Input Output
dn2
vg1 n2
Slow-down factor: S ( d
vg 2 dn1
n1
d
Limited control
Large control via
derivative terms
Group Velocity
Group
Velocity
x
WG1 WG2
Bandwidth
x
Bit Period
x
Bit Length Lin
= Period x Velocity x
Reduced Traffic Speed
100 km/h Speed Limit 10 km/h Speed Limit
Reduced Group Velocity
Optical Pulses in Slow Light Delay Line
(WG1) (WG2) (WG3)
Input Slow Light Output Storage Time
Region Region Region
vg T L vg 2
Group vg1
Velocity
Profile Capacity
vg2 L
x C
0 Lbit
L
Field Delay-Bandwidth
Intensity Product
T B packet
x
Lin Lbit
FIFO Buffer Using Variable Delay Lines
z1 (t) z2 (t) z3 (t) zM-1 (t) zM (t)
Input Output
DL1 DL2 DL3 DLM-1 DLM
vg
vg1
vg2
0 x1 x2 x3 xM x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xM
Packet 1
Po
x
FIFO Buffer Using Variable Delay Lines
Call to Read Packet 1
vg1
vg2
x
x1 xM
Packet 1
4 3 2
Po
x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xN xM
Packet 1
4 3 2
Po
x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xN xM
Packet 1
5 4 3 2
Po
x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xN+1 xM
Packet 1
5 4 3 2
Po
x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xN+1 xM
Packet 1
6 5 4 3 2
Po
x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xM
Packet 1
6 5 4 3 2
Po
x
FIFO Buffer Using Variable Delay Lines
vg1
vg2
x
x1 xM
7 6 5 4 3 2
Po
x
RAM Using Staggered Delay Lines
Crosspoints enable packets
to be ―written‖ to and/or ―read‖
from any delay line
Single-packet
delay lines
and/or FIFO’s
Slowdown Factor
vg n1 n2
Group vg1
dn1 dn2
Velocity
Profile d d
vg2
x
L
Slowdown Factor Large
dn2
n2
S ( d
dn1
n1
d
Electromagnetically-Induced Transparency (EIT)
Resonance effect in a three-level system
Signal In Signal Out
Quantum well material
with three energy levels
Pump
Hole in
attenuation
characteristic
Attenuation
Optical Frequency
Electromagnetically-Induced Transparency (EIT)
Background
Attenuation
a(
i.e. Laplace
o
transform
Kramers
dn Kronig
becomes large n
d
D
EIT in Atomic Gas
Lene Hau,
PhysicsWeb
September
2001
Vacuum
500 nK
EIT in Atomic Gas
Information bandwidth < 1 MHz
Waveguide Losses
Fibre: ~0.2 dB/ km
In Out
44 km for 1 neper of loss
8.7 dB
Semiconductor WG: ~0.5 dB/ cm
In Out
12 cm for 1 neper of loss
Analysis of Slow-Light Material
EIT Ideal
Attenuation (a)
Background
Attenuation
a(0
o
Residual, aR Kramers n
Kronig
n navg
nmin
o
min o ma
D
x
D
Bandwidth
Fundamental Limitations of Slow Light
Delay-Bandwidth Delay-Bandwidth
Class Minimum Bit Size
Product Product/Neper Loss
L( navg - nmin ) navg - nmin 0
Ideal
0 a R 0 ( navg - nmin )
ln( 2) ln( 2) Da (0 ) 4 L
EIT LDa (0 )
4 4 aR ln( 2) Da (0 )
a n
navg
Da(0
aR nmin
o
EIT Ideal
Some Numbers
Delay- Capacity/neper Da ( 0 )
Class Bandwidth Bandwidth Bit Size @ 0.5 dB/cm
Product and 40 Gbit/s
235 kbits
Ideal 1.4x104 #,* 740 nm # ~2x106 #
(all bit rates)
300 kHz (Atomic Gas) 2^ 110 mm ^ - ~106
EIT/PO
2 GHz (QW Material) 10 mm ^ 39 bits ~2x106
<1 ^
Fiber/ 8.7 Mbits
2 #, *,** 0.5 cm #,**
crosspoint @ 0.2 dB/km
CMOS/RAM Very Large ~100 nm2
#Fundamental limit
^Example only – not a fundamental limit
* For a 1-cm long device
** 40 Gb/s data
Photonic Crystal Waveguide
Active Microresonator Delay Line
na
Each Resonator
contains EIT material nmax
navg
n min
min p max
D
ka ka ka ka
Resonators tc tc tc
eff
Active Microresonator Delay Line – Bloch Wave
Analysis
Passband
D
n
navg
FSR
p -1 1 p 2
Some Numbers
Delay- Capacity/neper Da ( 0 )
Class Bandwidth Bandwidth Bit Size @ 0.5 dB/cm
Product and 40 Gbit/s
235 kbits
Ideal 1.4x104 #,* 740 nm # ~2x106 #
(all bit rates)
-
300 kHz (Atomic Gas) 2^ 110 mm ^ ~106
EIT/PO
<1 ^ ^ 39 bits
2 GHz (QW Material) 10 mm ~2x106
PC 10 GHz 168 ^ 20 mm ^ 39 bits
Fiber/ 8.7 Mbits
2 #, *,** 0.5 cm #,**
crosspoint @ 0.2 dB/km
CMOS/RAM Very Large ~100 nm2
#Fundamental limit
^Example only – not a fundamental limit
* For a 1-cm long device
** 40 Gb/s data
Two Classes of Slow Light Delay Line
Class A
- Group velocity profile does not change while data stored
- Data enters and leaves slow-light regions across discontinuities
- All previous examples
Class B
- Group velocity can change while data is stored
- No experimental verification
Characteristics of Class B Delay Line
n
q E (t dt ) Eo e j ( s )(t dt )
navg
d (max - min )
s ( - 0)
max - min
nmin
o
min o max
Operation of Class B Delay Line
Refractive
Index
Slope t
Group
Velocity
t
Bandwidth
t
Bit Period
t
Bit Length
Lin
= Period x Velocity
t
FIFO Buffer using Class B Delay Lines
z1 (t) z2 (t) z3 (t) zM (t)
Input Output
DL1 DL2 DL3 DLM-1 DLM
vgc vgc
vg1 vg1
Field vg2 vg2
vg t t
Intensity
vg1 x
Bp
vg2 Bg1
x
Bg2
t
RAM Using Class B Recirculating Loop
vg
t
Locate empty cell Load Retrieve cell Write
Class B B
Class
Class B
vg
t
Read input Load input into loop
Improved RAM
vg
t
Locate empty cell Load Retrieve cell Write
Class B B
Class
Class A Class B Class A
vg vg
vg
x x
t
Read input Load input into loop
Conclusions
• Fundamental Limitations and Capabilities of Slow Light
- EIT/PO
- Microresonators
• Importance of the Delay-Bandwidth Product
• Limitations on Minimum Bit Size (Storage Density)
• New approaches to buffer design
• No magic solution to the optical buffer problem