digital quantum computing based on superconducting electronics · 2020. 5. 29. · superconducting...
TRANSCRIPT
CO
NF
IDE
NT
IAL
Digital Quantum Computing based on
Superconducting Electronics
O. Mukhanov Seeqc, Inc., Elmsford, NY
NY CREATES Seminar (online)
2020-May-20
Outline of Presentation
MotivationSuperconducting SFQ TechnologyDigital Quantum Computing System Initial resultsConclusions
2
Quantum Computing TodayGoogle 53-qubit “Sycamore” Quantum Supremacy setup
From ICQT conf. July 2019 By Erik Lucero, Google | Wired, Dec. 2018
ENIAC (1946, University of Pennsylvania)
How this installation transformed to modern computers?
Qubit Control: current microwave based approach
Large, expensive electronics racks distant from qubits
Analog qubit control:- Complex- High latency- Noisy- Expensive- Large I/O count- Non-scalable
Microwave equipment
Just Qubit Ctrl; Not including I/0 for readout and tunable coupler*Google quantum supremacy chip: C. Neill et. al., Science 360, 6385, 195-199 (2018)
Qubit Control Wiring Overhead
Rent’s Rule should be used for Quantum Circuits
6
2
10
100
1000
4
20
200
2000
Qubitcount
Using µwaves*
Large I/O Count = scalability problem. Long Wires = latency problem
Additional system integration considerations: Cost: current cost of a cable channel from room
temperature to milliKelvin stage is ~ $5,000 (per Oxford Instruments). Typically ~2-3.5 cables on average per
superconducting qubit (transmon) ~ > $10-11M cable cost for a 1000 qubit
system. Most expensive cables are between 4K to 20 mK stages
Reliability of a few hundred-channel cabling farm of a dilution refrigerator system is likely be not sustainable.
Options for co-located cryogenic classical circuits CryoCMOS Can work in cryogenics, but still dissipates relatively large power Best fit is for semiconductor spin qubits (Microsoft, Intel, EPFL). Google, Microsoft, Intel teams developed mixed-signal circuits for superconducting
and spin qubits (ISSCC’19, IEDM’19, ISSCC’20). General approach: recreating room temperature electronics with cryoCMOS.
Superconductor electronics Advantage: very low-power, very fast electronics based on Single Flux Quantum
(SFQ) logic.
Problem: much lower integration density than CMOS• Recreating microwave electronics (AWG, etc.) with SFQ logic is a complex task
(will require too much hardware). The result may not be competitive.
Is there a competitive SFQ-based approach?
SFQ Pulse: Low Energy Quantized-area pulse
• SFQ pulse width: tSFQ ~ Φ0/2IcR where 2IcR - pulse height
• High Speed: for Nb junctions, ultimate limit tSFQ → 0.4 ps; for complex RSFQ circuits, practical fclock ~ 1/(10 tSFQ)
• Low Power: SFQ pulse energy ~ ¾ Φ0Ic ~ 2 x10-19 Joule (for Ic ~ 100 µA for 4K operation) or ~2x10-20 -21 Joule (for Ic ~ 1-10 µA for 20K operation)
Maximum Clock Frequency for IC can reach ~100s GHz at low power
2IcRtSFQ
V
t
∫ Φ= 0VdtQuantized area Single Flux Quantum (SFQ) pulse generated by overdamped Josephson junction
RSFQ Technology - productized
∫Vdt = Φ0 = h/2e = 2.07 mV·ps
Both Data and Clock are SFQ voltage pulses V(t) with quantized areas
RSFQ - Rapid Single Flux Quantum(invented in mid-80s*, adopted to Hypres since early 90s, became the
main digital superconducting electronics world-wide by mid-90s)
750 GHz digital frequency divider internal memory gate-level pipelining high-throughput low switching power dc bias only local timing amendable for synchronous and asynchronous schemes
@4K, ESFQ= 10-19J@20 mK, ESFQ= 10-21J
* O. A. Mukhanov, V. K. Semenov, and K. K. Likharev, “Ultimate performance of the RSFQ logic circuits," IEEE Trans. Magn., MAG-23, pp. 759-762, Mar. 1987.
The 1st real (commercial-grade) application of digital superconducting electronics
DSP-Decimation FilterADC modulator
30 GS/s X-band Rx chip
Sumitomocryocooler
compressor
vacuum enclosurewith LTS chipand HTS filters mounted inside
output amplifiers
current source
temperaturecontroller
data acquisition and processing board (FPGA)
2005-2007
Low-pass Rx chip
HYPRES’ RSFQ Digital-RF Receiver
circa 2010
O. A. Mukhanov, D. Kirichenko, I. V. Vernik, T. V. Filippov, A. Kirichenko, R. Webber, V. Dotsenko, A. Talalaevskii, J. C. Tang, A. Sahu, P. Shevchenko, R. Miller, S. B. Kaplan, S. Sarwana, and D. Gupta, “Superconductor Digital-RF receiver systems,” IEICE Trans. Electron., vol. E91-C, pp. 306-317, Mar. 2008.
D. Gupta et al., "Modular, Multi-Function Digital-RF Receiver Systems," in IEEE Transactions on Applied Superconductivity, vol. 21, pp. 883-890, 2011.
Circa 2005
Energy-Efficient Successors of RSFQ Logic
Several post-RSFQ logics (ERSFQ, eSFQ , RQL, AQFP) were introduced ERSFQ and eSFQ achieve the fundamental SFQ energy dissipation related to magnetic
flux crossing Josephson junction ESFQ ~ IbiasΦ0 ~ 10-19 Joule Eliminates static dissipation from bias resistors (dominating dissipation) Retains all advantages of conventional RSFQ:
dc-powered, amendable for serial biasing to reduce total dc bias current ballistic interconnects (no extra power for integrate connections) high speed operation (can work at 100s of GHz) largely preserves already developed cell libraries
Icb = Ib
Vb = Φ0⋅ƒclk
Lb
Conventional RSFQ Energy-Efficient RSFQ (ERSFQ)
PS= Ib Vb
PD= Ib Φ0 PD= Ib Φ0
PS= 0
Ic
Ib ~ ¾ Ic
Vb
Rb
Ic
SFQ SFQ
Power Dissipation in ERSFQ/eSFQ
ERSFQcircuit
Ib VbVI
000
)()()( Φ⋅=⋅⋅=⋅⋅= ∫∫ b
T
b
T
IdttVIdttVtIE
Dissipated energy per one clock period:
Total power dissipation:
P = fclk⋅ Ib⋅Φ0
CO
NF
IDE
NT
IAL
13
It is difficult to improve this system
CO
NF
IDE
NT
IAL
14
SE
EQ
C A
PP
RO
AC
H
We want to redesign it with this:
Digital Quantum Computer
cryostat configuration
O. Mukhanov et al., "Scalable Quantum Computing Infrastructure Based on Superconducting Electronics," 2019 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, 2019
Resonant Excitation of Qubits by SFQ pulses
Feasibility proven by Wisconsin and Syracuse University groups
By varying distance between SFQ pulses in the train using control theory, one can achieve higher fidelities as shown by Saarland Univ. team
R. McDermott, M. G. Vavilov, B. L. T. Plourde, F. K. Wilhelm, P. J. Liebermann, O. A. Mukhanov, T. A. Ohki, “Quantum-classical interface based on single flux quantum digital logic,” Quantum Sci. Technol., 2018
Replacing microwaves with digital SFQ pulse train
R. McDermott and M. G. Vavilov, “Accurate Qubit Control with Single Flux Quantum Pulses,” Phys. Rev. Appl. 2, 014007 (2014)
By varying distance between SFQ pulses in the train using control theory, one can achieve higher fidelities as shown by Saarland Univ. team
Recent theoretical advance shows that achieving >99.99% fidelities is doable with low SFQ hardware complexity (55 bit shift register with 25 GHz clock) K. Li, R. McDermott, M. Vavilov, ”Scalable Hardware-Efficient Qubit
Control with Single Flux Quantum Pulse Sequences,” arXiv:1902.02911 [quant-ph]
R. McDermott, M. G. Vavilov, B. L. T. Plourde, F. K. Wilhelm, P. J. Liebermann, O. A. Mukhanov, T. A. Ohki, “Quantum-classical interface based on single flux quantum digital logic,” Quantum Sci. Technol., 2018
Improving Fidelity by using Optimal Control
Pulse Generation Unit (PGU)
Using superconducting SFQuClass circuits
100 µm
100 µm
Single Flux Quantum Classical circuits fit for a proximal location to qubits
SFQuClassTM circuit technology: • Energy-efficient ERSFQ or eSFQ
varieties of RSFQ logic• rescaling JJ critical currents to
<10µA (~10-20 J per switching).• use of high-kinetic inductors.• qubit-aware SFQ circuit layout
SFQ pulser generating SFQ pulse at each cycle of input sinewave clock
Fragment of PGU for storing and generating the optimal SFQ pulse pattern at 25 GHz clock
Fabricated with SeeQC 1000 A/cm2 critical current density process and a unit Josephson junction critical
current scaled down to 10 mA from 125 mA typical for the SFQ circuits designed for operation at 4 K.
qMCM = qubit chip + SFQ controller chip
SFQ Control Chip(made in SeeQC)
2-transmon chip (made in Syracuse U.)
Assembled at SeeQC multichip module (qMCM) mounted in dilution fridge at Syracuse U.Wisconsin-Syracuse-SeeQC Project
728_ABMCM16_SFQfon8_RabiWidth_vs_SFQpower
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
RabiWidth ( s)
8
9
10
11
12
13
14
15
16
17
18
SFQ
Pow
er (d
B)
Qubit shows clear thresholding behavior in Rabi oscillation experiments when the SFQ circuit dc current bias and trigger power are varied proving the SFQ qubit control action
724_ABMCM16_SFQfon8_RabiWidth_vs_switchbias
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
RabiWidth ( s)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Switc
h Bi
as (V
)
Courtesy of Britton PlourdeMeasured at Syracuse University
Digital SFQ Control of Qubit
Triggering the SFQ driver near ω01/8 results in clear Rabi chevrons – direct evidence of qubit coherent behavior
Courtesy of Britton Plourde
Measured at Syracuse University
Initial Testing: Digital SFQ Control
SFQ**
2
10
100
1000
4
20
200
2000
3
3
31
76
Not including I/0 for readout and tunable coupler – only qubit control *Google quantum supremacy chip: C. Neill et. al., Science 360, 6385, 195-199 (2018)
Qubitcount
Getting close to microprocessor-like Rent’s rule exponentRent’s Rule for Quantum Circuits
22
µwaves***Wires from 4K to 20mK classical SFQ chip (classical-to-classical non-coaxial cabling only)
No wires to Quantum Chip (capacitive and magnetic coupling only)
Seeqc (Digital) Conventional (Analog) Google (analog) Microsoft (mixed signal) Intel (mixed signal)
Technology SFQ and SFQuClass CMOS (at room temp) CryoCMOS (3K)[ISSCC2019]
CryoCMOS (20mK)[IEDM2019]
CryoCMOS (3K)[ISSCC2020]
Ctrl Signal Digital pulses pattern Analog microwave waveforms
Analog microwave waveforms
Analog waveforms Analog waveforms
Readout Digital to mK Analog to RT Analog to RT Analog to RT Analog to RT
Qubit type Superconducting qubits Superconducting qubits Superconducting qubits Spin-qubits, Majorana qubit (goal) Spin qubits, transmon (potential)
Clock speed 20 – 40 GHz 0.3 – 2 GHz 0.3 GHz 0.1GHz 1 GHz
Power (for Ctrl) 0.0002 mW/qubit 20 mW/qubit 2 mW (potential) 0.2 mW 1.7 mW/qubit (analog part) + 300 mW (digital part)
Heat from cables(limit for available DRs)
Limits to 1M-100M qubits
Limits to 100-150 qubits (360 qubits optimistically)
? 0.1M (without accounting for readout)
?
Interference Low High High Moderate Moderate
Latency(interconnect delay)Latency (readout/control cycle)
Picosecond scaleTens nanoseconds
Subnanosecond scale (or ~100,000x longer)
Milliseconds (or ~100,000x longer).Readout is still conventional
Milliseconds (or ~100,000x longer)Readout is still conventional
Milliseconds (or ~100,000x longer)Readout is still conventional
Rapid Loop Possible Not Possible Not Possible Not Possible Not Possible
Cost $100/qubit for up to 50 qubit processor <$10/qubit for > 50 qubit processor
$40,000/qubit (liner scale)
$4,000/qubit (liner scale)(guess)
$1,000/qubit (guess) $3,000/qubit(guess)
Qubit Readout/Control: Analog vs Digital
Large, expensive electronics racks distant from qubits
Small, low-cost cryogenic chips in close proximity to qubits
5 mm
Analog qubit control:- Complex- High latency- Noisy- Expensive- Large I/O count- Non-scalable
Digital qubit control:- Naturally integrated- Low latency- Less noise- Low cost- Reduced I/O scaling law- Scalable
Not to scale
Conventional: Microwave based
New: Digital SFQ based
SFQ digital circuits SFQ pulse train
Microwave equipment
R. McDermott et al., “Quantum-classical interface based on single flux quantum digital logic,” Quantum Sci. Technol., 3 (2), 024004, 2018O. Mukhanov et al., "Scalable Quantum Computing Infrastructure Based on Superconducting Electronics," 2019 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, 2019
CO
NF
IDE
NT
IAL
25
SFQuClass Digital Quantum Management (DQM) system-on-a-chip to scale a quantum computer
Proximally co-located and integrated with qubit chips in a cryo-cooled environment to maximize digital control:
o Controls and reads out qubits with digital pulses instead of microwaves
o Increases system speed and reduces latency
o Dramatic cost reduction
o Reduces environmentally induced error
o Reduces energy dissipation and heat/noise brought in via analog systems
o Reduces system complexity
o Designed to work with all superconducting qubits
CO
NF
IDE
NT
IAL
26
Hybrid classical-quantum computers
We are building the industry’s first truly hybrid quantum-classical computing system with an integrated SFQ digital quantum management co-processor supported by superconductive classical logic
o Co-located in the same cryogenic system as quantum and SFQuClass DQM layers
o Provide 10-40 GHz speed classical processing resources in support of quantum-classical algorithms and error correction
o Manages digital readout and control of quantum layer
o Seeqc classical computing layer enables industry leading fast hybrid quantum-classical software
CO
NF
IDE
NT
IAL
27
Application-specific quantum computers
We are building a computing platform that is
commercially scalable for targeted problem-specific
applications
o Reduce system resources required to solve high-value problems
o Increase quantum computer performance for given number of qubits
o Quantum hardware and software are tightly co-designed
ConclusionsSFQ-enabled Digital Quantum Computer Drastic Reduction of I/O count (similar Rent’s rule as
for modern CMOS microprocessors) Close proximity = low latency Advantage for algorithms and for Quantum Error
Correction
Cryogenically co-located classical processors/memoryQuantum (and quantum-classical) algorithms matched to
the hardware
more information at seeqc.com