Tannu Moinuddin Qureshi JigSaw Boosting Fidelity of NISQ Programs via Measurement Subsetting Operated in Noisy Intermediate Scale Quantum NISQ mode 2 Quantum Computers are Noisy ID: 928611
Download Presentation The PPT/PDF document "1 Poulami Das * Swamit" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
1
Poulami
Das* Swamit Tannu#Moinuddin Qureshi*
*
#
JigSaw
: Boosting Fidelity of NISQ Programs
via Measurement
Subsetting
Slide2Operated in “Noisy Intermediate Scale Quantum (NISQ)” mode
2
Quantum Computers are NoisyNoise determines the overall fidelity of NISQ systemsNear-term quantum computers have high error rates
NISQ Device
11
01
11
10
Output Log
Repeat Trials
Correct
Erroneous
Program
Slide33
The Problem: Measurement Errors
Measurement errors are dominant sources of errors in large programsProgramCNOT q0, q1Measure q0
Measure q1
NISQ Hardware
q0
q1
Map
Error!
Error!
Each trial measures all program qubits
All measurements must be Error-Free
Slide4Measurement crosstalk can increase the effective error-rate
Measurement crosstalk increases with program size
4
Challenge with Measurements at Scale
Crosstalk
Isolated
Multiple
Slide5The Insights
Insight: Measure
fewer qubits and Reconstruct distribution 5Goal: Reduce Measurement Errors
X
Ideal
Are high fidelity Circuits with Partial Measurements alone sufficient?
Slide6The Insights
6
Need for Correlation
X
Ideal
Ideally, we want both correlation and high fidelity
Slide77
Outline
Background and MotivationJigSaw: Insights and Design
Evaluations
Slide8Program gives full correlation; CPM gives high fidelity.
JigSaw
combines both8JigSaw: Design. . .
Original program
Circuits with Partial Measurements
(CPM)
NISQ Device
NISQ Device
Global-Mode
(50% of Trials)
Subset-Mode
(50% of Trials)
Recompile
Recompile
Slide9?
CPM updates the global distribution and improves the fidelity
9Bayesian Reconstruction in JigSaw
Q2Q1
Q0Prob.
0 0 0
0.15
0 0 1
0.05
0 1 00.10 1 1
0.05
1 0 0
0.15
1 0 1
0.2
1 1 0
0.1
1 1 1
0.2
Q
1
Q
0
Prob.
0 0
0.1
0 1
0.1
1 0
0.1
1 1
0.7
Q
2
Q
1
Q
0
Score
0 0 0
0.06
0 0 1
0.01
0 1 0
0.06
0 1 1
0.47
1 0 0
0.06
1 0 1
0.09
1 1 0
0.06
1 1 1
1.87
Q
1
Q
0
Q
2
0
1
Global Mode
Subset Mode
Update Coefficients
00
01
10
11
0.5
0.5
0.2
0.8
0.5
0.5
0.2
0.8
0.06
=
Output
Cannot infer solution
Correct!
Q
2
Q
1
Q
0
Prob.
0 0 0
0.02
0 0 1
0.01
0 1 0
0.02
0 1 1
0.18
1 0 0
0.02
1 0 1
0.03
1 1 0
0.02
1 1 1
0.70
Slide10JigSaw
-M: Multi-Layer
JigSawProgram. . .CPM
JigSaw-M uses heterogeneous-sized CPM
Default subset size: 2
. . .
More Unique CPM
Other subset size: 3
Slide1111
Outline
Background and MotivationJigSaw: Insights and DesignEvaluations
Slide1212
IBMQ hardware: 27 to 65 qubits
Evaluation MethodologyNoise aware compilation, 32K– 256K TrialsFigure-of-Merit 1: Probability of Successful Trial (PST) PST =
A higher PST and Fidelity is desirable
Figure-of-Merit 2: Fidelity
Fidelity =
Slide1313
Results for PST
EDM: Ensemble of Diverse Mappings, Tannu et al., MICRO 2019Average: 3.1x Best-Case: 8.4x
Slide1414
Results for Fidelity
AverageJigSaw: 2.1xJigSaw-M: 2.5x
Slide1515
How many CPM do we need?
Diminishing ReturnsQAOA-12 (p4) on IBMQ-ParisFew unique CPM are sufficient, scale linearly with number of qubits (default)N-qubit program has NC2 possible CPM of subset size 2
Slide1616
Impact of Recompilation
Recompiling each CPM enhances the effectiveness of JigSawAverage: No Recompilation-> 1.9x, With Recompilation-> 2.9x
Slide1717
Scalability Analysis
Complexity is determined by number of unique outcomesJigSaw does updates only for non-zero outcomes (limited by trials)The linear complexity makes JigSaw scalable to large machinesProgram Size (Num. of Qubits)JigSawJigSaw-M
Memory
(GB)
Operations (Billion)
Memory (GB)
Operations (Billion)
100
10.4
4
1.7
500
5
2.1
20
8.4
*Assuming 1 Million trials, and pessimistically each trial yields a unique outcome
Slide1818
Conclusion
Measurement errors limit fidelity of large NISQ programsJigSaw mitigates impact of measurement errors via subsettingJigSaw measures:All program qubits in 50% of trialsSubset of qubits in remaining 50% trials
JigSaw uses Bayesian post-processing to combine the results
Fidelity improves by 2.5x to 8.4x on 3 IBMQ machines
Slide1919
Thank You!