Quantum Multicomputers
Rodney Van Meter
Keio University
慶應義塾大学
@QCE 2025 Albuquerque, NM
2025/9/1
2
Happy (U.S.) Labor Day! Dedication:
To:
…and also to my Dad, who passed away in June.
3
A Thought
We live in extraordinary times:
but also:
4
5
6
Networks for Modular Quantum Systems
Rodney Van Meter
Keio University
慶應義塾大学
@QCE 2025 Albuquerque, NM
2025/9/1
7
EQIS 2003, Kyoto
8
What is the Quantum Computer �Architecture “Stack”?
This is not quite an architecture diagram per se, more like a community structure.
Programming languages inadvertently left out, and this was before thinking about education, community, supporting technologies, problem partners, and customers.
rdv Ph.D. thesis, 2006,
https://arxiv.org/abs/quant-ph/0607065
9
AQUA Group @SFC 2024
（now about 40 people）
10
11
Joint work with >110 quantum co-authors since 2003:
Muhammed Ahsan, Hideharu Amano, Yoshinori Aono, Luciano Aparicio, Alán Aspuru-Guzik, Praveen Balaji, Naphan Benchasattabuse, Angela Sara Cacciapuotti, Marcello Caleffi, Joshua C.A. Casapao, Poompong Chaiwongkhot, Areeya Chantasri, Byung-Soo Choi, Hyensoo Choi, Joaquin Chung, Andrew N Cleland, Gordon Cui, Simon Devitt, Clément Durand, David Elkouss, Suguru Endo, Hiroshi Esaki, Paolo Fittipaldi, Austin Fowler, Andrew Greentree, Frédéric Grosshans, Michal Hajdušek, Lajos Hanzo, Charles D Hill, Lloyd CL Hollenberg, Dominic Horsman, Rikizo Ikuta, Kaori Ishizaki, Kohei M. Itoh, Liang Jiang, N Cody Jones, Hideyuki Kawashima, Raj Kettimuthu, Jungsang Kim, Alexander Kolar, Marii Koyama, Wojciech Kozlowski, Noboru Kunihiro, Thaddeus Ladd, Sitong Liu, Sebastien GR Louis, Ananda G. Maity, Nam Mannucci, Takaaki Matsuo, Peter L McMahon, Sara Metwalli, Shigetora Miyashita, Darcy QC Morgan, Yoshihiro Mori, Bill Munro, Shota Nagayama, Ken M. Nakanishi, Yehuda Naveh, Kae Nemoto, Trung Duc Nguyen, Shin Nishio, Ryo Nojima, Yasuhiro Ohkura, Hiroyuki Ohno, Takafumi Oka, Kento Oonishi, Alto Osada, Mark Oskin, Yulu Pan, Natchapol Patamawisut, Poramet Pathumsoot, Tien Trung Pham, Pawan Poudel, Bruno Rijsman, Kazuhiro Saito, Daisuke Sakuma, Bernard Ousmane Sane, Toshihiko Sasaki, Rei Sato, Ryosuke Satoh, Takahiko Satoh, Alireza Seif, Hikaru Shimizu, Naoyuki Shinohara, Ansh Singal, Akihito Soeda, Tomah Sogabe, Kento Samuel Soon, Ashley Stephens, Michihiko Sugawara, Kenji Sugisaki, Sujin Suwanna, Shigeya Suzuki, Amin Taherkhani, Tomoki Tanaka, Theerapat Tansuwannont, Kentaro Teramoto, Andrew Todd, Monet Tokuyama, Joe Touch, Agung Trisetyarso, Tomoki Tsuno, Yosuke Ueno, Shumpei Uno, David S Wang, Stephanie Wehner, James D Whitfield, Robert Wille, Qian Xu, Yoshihisa Yamamoto, Hiroshi Yamauchi, Haoxiong Yan, Hikaru Yokomori, Nobuyuki Yoshioka, Claire Yun, Man-Hong Yung, Allen Zang,
and probably some others
(just the quantum people, not purely classical)
(includes RFC & unpublished arXiv but not roadmap & research reports)
Rodney Van Meter
Professor, Faculty of Environment and Information Studies, Keio University
Cyber-Informatics Program Chair, Graduate School of Media and Governance
Vice Center Chair, Keio Quantum Computing Center
Board Member, WIDE Project
Leader, Advancing Quantum Architecture (AQUA)
Quantum Internet Task Force (QITF)
Co-chair, Quantum Internet Research Group (QIRG)
Editor in Chief, IEEE Transactions on Quantum Engineering
rdv@sfc.wide.ad.jp
12
From Experiment to System
What takes us in that direction?
13
14
15
16
QKD
initial idea
First laboratory distributed entanglement experiments
1980s
1990s
2000s
2010s
2020s
2030s
First QKD network testbeds:
Boston, Geneva, Tokyo
First entangled networks
Commercial Quantum Internet
Entanglement networks proposed
Fiber and satellite links tested
Demonstrated in the lab
First commercial products
Operational network Beijing-Shanghai
Quantum key distribution
Entangled networks
17
QKD
initial idea
First laboratory distributed entanglement experiments
1980s
1990s
2000s
2010s
2020s
2030s
First QKD network testbeds:
Boston, Geneva, Tokyo
First entangled networks
Entanglement networks proposed
Fiber and satellite links tested
Demonstrated in the lab
First commercial products
Operational network Beijing-Shanghai
Quantum key distribution
Entangled networks
Commercial Quantum Internet
18
First laboratory distributed entanglement experiments
1980s
1990s
2000s
2010s
2020s
2030s
First entangled networks
Entanglement networks proposed
Fiber and satellite links tested
Entangled networks
Quantum Multicomputer/�Data Center Networks
Commercial Quantum Internet
Entanglement
19
Making Node-to-node Entanglement
Link types vary, but a common type interferes photons at the midpoint of the link: Memory-Interference-Memory, or MIM
Making Node-to-node Entanglement
Link types vary, but a common type interferes photons at the midpoint of the link: Memory-Interference-Memory, or MIM
Entanglement Swapping (~Teleportation)
Entanglement distribution in quantum networks is performed by entanglement swapping on photonic qubits.
What has to be satisfied for Entanglement swapping?
➡️ the photons have to be indistinguishable
Properties that determine how distinguishable the photons are
22
Spectral distinguishability
Photon Wave Packet Overlap
Photon synchronization at the BS is crucial: good fidelity = high visibility!
Photon arrival time at the BS determines the interference visibility.
Factors that can lead wave packets to not overlap
23
Bad Overlap
24
Two photons have to arrive at the beam splitter (BS) at the same time.
This will give poor visibility
(makes poor or no entanglement)
Improving Photon Wave Packet Overlap
We use Optical Delay Lines (ODLs) to adjust the timing of the photons
Timings we have to consider
25
From OZ optics manual
Good Overlap (same arrival time)
26
Two photons have to arrive at the beam splitter (BS) at the same time.
This will give good visibility (makes good entanglement)
27
Hong-Ou-Mandel dip (data from our lab)
Bad overlap
(P=50% two detectors will click)
Good overlap (one detector should click)
Making Node-to-node Entanglement
Link types vary, but a common type interferes photons at the midpoint of the link: Memory-Interference-Memory, or MIM
Making Node-to-node Entanglement
Link types vary, but a common type interferes photons at the midpoint of the link: Memory-Interference-Memory, or MIM
Importance of Entanglement
Remote CNOT gates (between distant qubits) consume entanglement.
30
local CNOT gates
Z measurement
X measurement
classical communication
Bell pair
Eisert et al., 2000, https://journals.aps.org/pra/abstract/10.1103/PhysRevA.62.052317
Node 1
Node 2
Importance of Entanglement
Remote CNOT gates (between distant qubits) consume entanglement.
31
1 Bell pair = 1 remote CNOT gate
Eisert et al., 2000, https://journals.aps.org/pra/abstract/10.1103/PhysRevA.62.052317
=
A very rough look at the necessary quantum computer roadmap
Why data center-scale networks are necessary!
32
Evolving Hardware Environment
33
more application qubits
early qubits
NISQ
higher fidelity
Evolving Hardware Environment
34
more application qubits
NISQ
NISQier
less Noisy ISQ
first FT demos
early qubits
higher fidelity
Evolving Hardware Environment
35
more application qubits
NISQ
NISQier
less Noisy ISQ
first FT demos
early qubits
longer code distances
more FT qubits
Max-SQ�(max single-chip FT system)
higher fidelity
Evolving Hardware Environment
36
more application qubits
fault tolerant quantum multicomputer
NISQ
NISQier
NISQ multicomputer
less Noisy ISQ
first FT demos
early qubits
longer code distances
more FT qubits
Max-SQ�(max single-chip FT system)
interconnects required
higher fidelity
(Selected) Early Multicomputer Designs (not Experiment!)
Duke Ion Trap
Multiple logical qubits/node
Full crossbar
2005
Design by PresentationGO.com
(My apologies if your work or your favorite work doesn’t show up here)
(Selected) Early Multicomputer Designs (not Experiment!)
Duke Ion Trap
Multiple logical qubits/node
Full crossbar
2005
Keio/NII/HP Multicomputer
Top-down architecture
Starting from workload
Multiple logical qubits/node
Various unswitched network topologies
Line topology works well!
2006
Melbourne Ion Trap
One logical qubit/node
Unspecified multiplexer
2006
Harvard Small-Register
Inter-module QEC
Unspecified MEMS switched topology
2007
Design by PresentationGO.com
(Selected) Early Multicomputer Designs (not Experiment!)
Duke Ion Trap
Multiple logical qubits/node
Full crossbar
2005
Keio/NII/HP Multicomputer
Top-down architecture
Starting from workload
Multiple logical qubits/node
Various unswitched network topologies
Line topology works well!
2006
Melbourne Ion Trap
One logical qubit/node
Unspecified multiplexer
2006
Keio/NII/Stanford/Melbourne Dist. QD Surface Code
Single surface spanning multiple chips
Extremely large scale
Extremely challenging
Linear topology
2010
Oxford Fully Dist. Surface Code
Small registers
Single surface spanning chips
Square lattice (4 interfaces/cell)
2013
Harvard Small-Register
Inter-module QEC
Unspecified MEMS switched topology
2007
Duke/Keio Multi-tier
Steane code
Multiple logical qubits/node
Tree structure
2015
Design by PresentationGO.com
(Selected) Early Multicomputer Designs (not Experiment!)
Duke/Keio Multi-tier
Steane code
Multiple logical qubits/node
Tree structure
2015
Design by PresentationGO.com
Chicago/Keio �Cosmic Ray Tolerant
Two-tier QEC
Inter-node used to reconstruct after catastrophic data loss
Small-group all-to-all
2022
(Not much happened in here…)
(Selected) Recent Multicomputer Designs
MIT Repeater Tree
Network nodes have memory
Repeater-like
Several topologies
Fat tree
2023
UT Sydney Inter-node Lattice Surgery
Inter-node lattice surgery
Unspecified topology
2024
Spain/Delft
Traffic Analysis
Best traffic analysis since our early work
All-to-all
2023
ARQUIN
Detailed study of remote gates over superconducting → optical transducer
Ad hoc topologies
2024
Chicago/Keio �Cosmic Ray Tolerant
Two-tier QEC
Inter-node used to reconstruct after catastrophic data loss
Small-group all-to-all
2022
Q-Fly
Surface code
Multiple logical qubits/node
Dragonfly-family structure
2024
Design by PresentationGO.com
Questions
42
Questions
43
Quantum Error Correction
Logical qubit confined to one node
Code block spans multiple nodes
44
Quantum Error Correction
Logical qubit confined to one node
Code block spans multiple nodes
45
lattice surgery: Horsman, Fowler, Devitt, rdv, NJP 2012
46
lattice surgery: Horsman, Fowler, Devitt, rdv, NJP 2012
Logical qubit A in node 1
Logical qubit B in node 2
Remote gate mediated by logical Bell pair or physical Bell pairs?
47
lattice surgery: Horsman, Fowler, Devitt, rdv, NJP 2012
Remote gate mediated by logical Bell pair or physical Bell pairs?
Logical Bell pair requires lots of fault-tolerant memory (7~8 logical qubits/node to buffer one high-fidelity Bell pair)
Physical Bell pairs: transversal logical gate, lattice surgery or stabilizer measurement
Purify(*) then inject(+)?
Inject then purify?
Requires high-fidelity physical memory, physical-layer purification
Requires lots of physical-layer entanglement
(* Purification ~ error detection
+ Inject ~ encode in QEC)
Questions
48
Multicomputer interconnect
49
Traffic usually centrally coordinated, proceeds in regular patterns
Round 1
Multicomputer interconnect
50
Round 2
Traffic usually centrally coordinated, proceeds in regular patterns
Multicomputer interconnect
51
Round 3
Traffic usually centrally coordinated, proceeds in regular patterns
Variable Placement & Management
52
Compilation for Multicomputers
53
Graph state-based:
Liu et al., QCE ‘23 https://arxiv.org/abs/2306.03758 �Vijayan et al., QST 9 025005 (2024)
Litinski, Quantum 3, 128 (2019)
Questions
54
Multicomputer interconnect
55
Direct connection in 2-D fabric. �Most nodes need four network interfaces.
Multicomputer interconnect
56
If each link is MIM, we need many BSAs!
=
Multicomputer interconnect
58
Nodes all share one pool of detectors via switch
N nodes, N/2 BSAs
A
B
C
D
E
F
…
Approach being taken by e.g. �Monroe, Kim & IonQ
Multicomputer interconnect
59
Nodes all share one pool of detectors via switch
N nodes, N/2 BSAs
A
B
C
D
E
F
…
Measure here to create entanglement between the two nodes on the left
Multicomputer interconnect
60
Nodes all share one pool of detectors via switch
N nodes, N/2 BSAs
A
B
C
D
E
F
…
Measure here to create entanglement between the two nodes on the left
Multicomputer interconnect
61
Nodes all share one pool of detectors via switch
N nodes, N/2 BSAs
A
B
C
D
E
F
…
Entanglement created between nodes!
…but it’s a consumable resource, so we have to make lots of it!
62
Koyama, …, rdv, QCE 2024 / https://arxiv.org/abs/2405.09860
All Pairings, not all permutations
63
All permutations (Beneš network) requires 6 switch points, but all pairings requires only 2.
All Pairings, not all permutations
64
All permutations (Beneš network) requires 6 switch points, but all pairings requires only 2.
All Pairings, not all permutations
65
All permutations (Beneš network) requires 6 switch points, but all pairings requires only 2.
66
Koyama, …, rdv, QCE 2024 / https://arxiv.org/abs/2405.09860
67
Classical Dragonfly
Many (classical) data centers and supercomputers use a dragonfly topology.
Adapting this to quantum data centers or multicomputers is nontrivial.
68
Basic Idea of a Group
Current challenge on our agenda is configuring these groups into a larger system,��planning for large-scale multicomputers and data center networks.
69
End nodes�(computation, sensing, etc.)
To other switches
From other switches
NxN optical switch
Cryostat with BSAs
70
71
Intra-group communication
Photons cross two group switches total
72
Inter-group communication
Photons cross three group switches total
73
Inter-group communication
via longer route
Photons cross four group switches total
Configuring the network
74
Configuring the network
75
Group parameters:
= group size (# of end nodes)
Configuring the network
76
Group parameters:
= group size (# of end nodes)
= # of BSAs
Configuring the network
77
Group parameters:
= group size (# of end nodes)
= # of BSAs
= switch radix� (# of output ports)
Configuring the network
78
Group parameters:
= group size (# of end nodes)
= # of BSAs
= switch radix� (# of output ports)
Global parameters:
= # of groups
(total # of end nodes)
(total # of detectors)
79
Inter-group communication
via longer route
Photons cross up to four group switches total
Architecture scales to tens of thousands of nodes
Implementation is limited by the loss of the group switch
Questions
80
Three Pillars: Experiment, Simulation, Theory
Real Testbed
Simulation
81
81
working on both &
going back and forth
@Nagayama lab in Shin-Kawasaki
to confirm in practice
to project large-scale operation
82
Nagayama Moonshot lab
Nov. 2022
83
84
Aug. 2025
85
Four work packages:
86
Node A is emulating a computational node (at the top)
87
Nov. 2024
Group switch
Node A EPPS
Node B EPPS
Pulse Freq. Conversion (Node B, C)
single-�photon
detectors
Node C EPPS
Node A Freq. Conv. on separate table
BSA
optics
88
Workload: Quantum Fourier Transform (QFT)
89
Tested Configurations
90
monolithic
tiling
versus
Execution of the QFT on Q-Fly
91
17 groups
8 nodes/group
1 app qubit/node
136-qubit QFT
(Exp. 6 on next slide)
5 groups
2 nodes/group
13 app qubits/node
130-qubit QFT
(Exp. 1 on next slide)
Execution of the QFT on Q-Fly
92
17 groups
8 nodes/group
1 app qubit/node
136-qubit QFT
(Exp. 6 on next slide)
Execution of the QFT on Q-Fly
93
17 groups
8 nodes/group
1 app qubit/node
136-qubit QFT
(Exp. 6 on next slide)
Q-Fly evaluation
94
tiling
95
Questions
96
Questions
97
Coda:
Wide-Area Entangled Networking & Quantum Communications Education
98
Quantum Internet: Quantum Recursive Network Architecture (QRNA)
99
rdv et al., Prog. Informatics. 2013, https://arxiv.org/abs/1105.1238
Matsuo et al., Phys. Rev. A 2019
rdv et al., QCE ‘22, https://arxiv.org/abs/2112.07092
Satoh et al., QCE ‘22, https://arxiv.org/abs/2112.07093
Teramoto et al., QuNet ‘23, https://dl.acm.org/doi/abs/10.1145/3610251.3610556
Benchasattabuse, Ph.D. dissertation, Keio 2025
Key Architectural Principles
100
RuleSet-based operation
Matsuo et al., Phys. Rev. A 2019
Two-pass connection setup
Flags
Routing
rdv et al.,
Net. Sci. 2013
Architecture
rdv et al.,Prog. Info. 2013
Layered Protocols
rdv et al.,Trans. Networking, 2009
Aparicio et al., SPIE 2012
Multiplexing
Flags
Internetworking
Nagayama et al., Phys. Rev. A 2016
Hijacking of a Repeater
Satoh et al., QST, 2018
Network Coding
Matsuo et al.,Phys. Rev. A 2018
Taherkhani et al., QST 2017
Application Protocols
Quantum Internet Task Force
103
103
Established in May 2019 by researchers from
Hokkaido Univ., The Univ. of Tokyo, Waseda Univ., ICU, NII, Mercari, Inc., Keio Univ., YNU, Osaka Univ., NICT
104
We have created the Quantum Internet Research Group (QIRG) inside the Internet Research Task Force (IRTF). Co-chairs are Van Meter (Keio) and Kozlowski (Netherlands SURF).
522 list members (as of 2023/9/8)
RFC9340
This first document took four years.� (not even standards track)
105
AQUA:�Products you can use!
106
We’re Not Just Pretty Faces
Upcoming:
107
Learning, Teaching & Explaining
108
109
Keio Online Course for Beginners
Q-Leap
110
Quantum Communications undergrad textbook �now available!
111
Network Designs & Specifications
112
113
Green: available now
I encourage you to read “Timing Regimes”
Blue: coming soon
Yellow: out of scope
Pink: where should we publish???
Software!
114
QuISP: Quantum Internet Simulation Package
Open-source simulator for quantum network built on top of OMNeT++.
Ready to use – create your network topology file and simulation file parameters�(e.g., number of memories, noise channels, simple traffic modeling)
Customization needs C++ and recompilation.�(e.g., changing scheduling policy)
Available via docker (full �functionality) or wasm browser�(limited functionality).�
QoL for beginners in progress…
115
https://github.com/sfc-aqua/quisp
PnPQ: Controlling Devices in the Lab
Open-source Python library for controlling common testbed equipment like waveplates, switches, polarization controllers, and optical delay lines
(we run on a set of RasPis)
116
https://github.com/moonshot-nagayama-pj/PnPQ
TDC Toolkit: Towards Measurement & BSA Nodes
Open-source Rust library for controlling high-precision time taggers used in photon coincidence counting
Will allow network nodes to implement real-time processing of time-tag data
117
https://github.com/moonshot-nagayama-pj/tdc_toolkit
Future Work/Open Questions
118
Future Work/Open Questions
119
Join Us!
120
121
122
123
AQUA
From Experiment to System
Go #QuantumNative
https://aqua.sfc.wide.ad.jp/redirect/qce-2025-keynote
124
Multicomputer History References
125
Additional Slides �(for Q&A or your browsing pleasure)
126
Building and Flying the “Quantum Airliner”: Whole Stack Quantum Computer Development
127
all developed by the Wright Brothers themselves (using wind tunnels, measurements, complex kites)
128
(They were also closed-source and big on defending patents, whereas rival Samuel Langley was what we would now call an open science, open source advocate. The Wright Brothers “won”, but then impeded the growth of the aviation industry in the U.S.; the center of aviation moved to Europe for two decades.)
129
130
https://modernairliners.com/boeing-787-dreamliner/boeing-787-dreamliner-assembly/
From Experiment to System
What takes us in that direction?
131
Hardware Components & Subsystems (BOM)
132
Software Tools We Need
133
Software Tools We Need
134
Software Tools We Need
135
Software Tools We Need
136
writing the application (quantum kernel + classical application code)
writing the experiment
} running the experiment
(mostly vendor owned)
} the future of QSE
how much can we leverage
existing classical tools?
}
}
From Experiment Toward Software Engineering
I showed Dave Farber (grandfather of the Internet) some Qiskit code, complete with job management and data post-processing. He responded (paraphrasing),
This isn’t an application, it’s an experiment.
So how do we get to applications that people can use?
137
None of this is possible without theory
Note that theory comes in many flavors, and architecture is different from e.g. device theory, computational complexity theory, algorithmic ideas, etc.
138
139
140
Entanglement is sufficient but not known to be necessary to scale computation?
141
142
Aug. 2025
143
Nov. 2024
144
145
146
Hong-Ou-Mandel dip (data from our lab)
Bad overlap
Good overlap
Execution of the QFT
147
5 groups
2 nodes/group
13 app qubits/node
130-qubit QFT
(Exp. 1 on next slide)
Execution of the QFT
148
17 groups
8 nodes/group
1 app qubit/node
136-qubit QFT
(Exp. 6 on next slide)
Execution of the QFT on Q-Fly
149
5 groups
2 nodes/group
13 app qubits/node
130-qubit QFT
(Exp. 1 on next slide)
150
151
Execution of the QFT on Q-Fly
152
17 groups
8 nodes/group
1 app qubit/node
136-qubit QFT
(Exp. 6 on next slide)
Execution of the QFT on Q-Fly
153
154
155