2025-11-02 - AQSE - Qiskit Fall Fest - IIT-Guwahati.pdf

Automated Quantum Software Engineering - Keynote address @ IIT Guwahati for Qiskit Fall Fest

1. Aritra Sarkar
2025.11.02
Keynote @ Qiskit Fall Fest - IIT Guwahati
Automating Quantum Computing Software
Application
System

2. B.Tech Avionics
(specialisation: Space Robotics)
Indian Institute of Space Science and Technology
2009-2013
Scientist
Satellite Onboard Control and Digitals Subsystem
Indian Space Research Organisation
2013-2016
Postdoctoral Researcher
Quantum Machine Learning group
Quantum Computing division, QuTech
2022-2024
M.Sc. Computer Engineering cum laude + Ph.D.
Quantum Computer Architecture group
Department of Quantum & Computer Engineering
2016-2018 + 2018-2022
QCA
QIT
XAI
Senior Researcher
Quantum Technology Division
Fujitsu Research India Private Limited
2024-present

3. a Device that uses
the Laws of Quantum Information for Computational Advantage
1 2+ = classical computers, small quantum computers
32+ = quantum computational model
3 1+ = accelerators (e.g., GPU)
1
2 3

4. Computer
Science
Quantum
Physics
❑from understanding to control
❑from passive to active
❑sensors, communication,
simulation, computation
❑from science to technology
❑new ways of computing
❑reversible
❑resource complexity
❑natural quantum simulators
Computer
Engineering
❑need for computing ever
increasing
❑₹ → faster computers → smaller
transistors → quantum effects →
accelerators and ASICs → ₹

5. “properties” of both classical waves (when we are not looking) and classical particles (when we look)
Superposition – Interference – Measurement

6. Superposition – Entanglement – Magic – Interference – Measurement
INTERFERENCE
sour
savory
bitter
sweet
tastefood
if:
then:
mix them up
food wt. const.
sour
savory
bitter
sweet
object properties
SUPERPOSITION
ENTANGLEMENT
MEASUREMENT

7. ●Representing quantum information:
•Physically: energy levels, polarization, spins
•Conceptually: 2-level qubits, qudits
●Mathematical model:
•Linear combination of states
•Weighted by a complex number (amplitude)
•Modulus of amplitude = probability of observing
•The probabilities add up to 1
•Each trial give a different outcome (need the bias)
•Each qubit doubles the number of states
●Why such a weird model?
•because... that agrees with the experiments!
0 (and/or) 1
2
n
states
1 state
when observed
q = z
0 + z
1
q = 1 q = 1
p(0)=|z
0| p(1)=|z
1|
Kurzgesagt – In a Nutshell

8. Gate-based
QC
Universal
TM
Universal
QTM
…
algorithm a
implementation
C2
a
CX
a= QX
a*[QX:CX]
n
physical
process
problem size n
Adiabatic
QC
Cellular
Automata
…
λ
Calculus
implementation
C1
a
C2
a= C1
a*n
[C1:C2]
implementation
Q1
a
implementation
Q2
a
Q2
a= Q1
a*n
[Q1:Q2]
quantum

10. 1.Same Turing degree in arithmetic hierarchy as CC, no hypercomputation
1.Must beat SotA HPC QC sims (e.g., tensor networks-based simulators)
2.QC > CC only in resources exp/poly
1.BQP\BPP few; BQP\BPP + Practical use cases: fewer
2.Advantage in other resources, e.g., space complexity (memory efficiency), generalization, representation capacity
3.Hard to infer quantum advantage from code structure (e.g., embarrassingly parallel => GPU)
2.Many industrial problems are NP-hard
1.NP-complete is most likely outside BQP, thus, no exponential advantage
2.NP problems can be solved with quantum search; poly speedup over CC if no better heuristics known
3.QC models are poly equivalent, CC models are poly equivalent
1.Poly advantage w.r.t. CC in 1 QC model might get negated in another QC model
2.Poly advantage w.r.t. CC in 1 QC model might be dequantized in another CC model
4.Asymptotic complexity might cross over to quantum advantage at impractical problem size
https://arxiv.org/abs/2212.00619
Automated Quantum Software Engineering: why? what? how?

11. NISQ
FTQC
QEC
Classical Simulation Limit
number of qubits
error rate
NISQ: Noisy Intermediate-Scale Quantum
FTQC: Fault Tolerant Quantum Computing

12. Software 1.0 - explicit instructions to the computer written by a programmer
Software 2.0 - specify desirable behavior of a program and search
Software 3.0 - prompt-based, vibe coding, agentic, etc.

13. How are quantum algorithm designed?
Approach 1: quantum complexity idealists
•Given: BPP ⊂ BQP; set of universal quantum gates
•Find: new quantum algorithms for specific mathematical properties
•E.g.: a super-polynomial speedup in determining the zeta function of a genus curve over a finite field
•Focus: asymptotic speedup w.r.t. best classical approach based on ideal qubits, gates, connectivity, etc.
•Automate: quantum information theory

14. How are quantum algorithm designed?
Approach2: quantum transformation wizards
•Given: an industrial computational use case
•Find: tweak an existing quantum algorithm
•E.g.: a pipeline for satellite image processing using one quantum convolution layer on a neural network
architecture
•Focus: PoC quantum kernel (maybe QC simulation) embedded within an existing application framework
•Automate: quantize problem formulation (qubo/sat to parametric circuits), data embedding and training
(classical optimizer in PQC)

15. How are quantum algorithm designed?
Approach 3: quantum advantage torchbearers
•Given: constraints of a specific QPU
•Find: demonstration of a PoC for a promising use case
•E.g.: the protein-folding problem on a tetrahedral lattice using a hybrid classical-quantum algorithm with
pulse-level optimized control on an IBM Eagle 127-qubit QPU
•Focus: extracting as much computation power as possible on NISQ hardware using hardware-software co-
design
•Automate: quantum control (pulse-shaping), low-level compilation (circuit optimization, scheduling,
mapping,…)

16. 1.5yr
1.5yr
1.5yr

17. Quantum
Computing
Stack
Quantum
Micro-
architecture

18. Kurzgesagt – In a Nutshell

19. ~2018
A Microarchitecture for a Superconducting Quantum Processor
10.1109/MM.2018.032271060

23. mathematical
abstractions of
quantum algorithms
QISA
Pulses
physical manipulation of
quantum hardware
QASM
Hamiltonians
Super-operators
Decoherence time
Unitary gates
Noise models
Complexity classes
compiler
microarchitecture
analog
control
Heisenberg interface
classical finite descriptions

24. Quantum-accelerated Genome Sequence Reconstruction

25. Quantum Universal Reinforcement Learning

26. Yet Another Quantum Quantizer (YAQQ)
Are all QC alike?
Theorist: Yes!
Church-Turing-Deutsch principle:
•Universal quantum computation set can simulate any
physical dynamics
•QTM later generalized to the circuit model and k-local
gate sets ({H,T,CX}, {D(θ)}, {Toffoli,H})
Solovay-Kitaev theorem:
•Iterated shrinking lemma: find better approx. of U by
recursively inc. decomposed circuit length
•Universal QGS approximate the Hilbert space with
??????(????????????
3.97
Τ
1
??????) gates in ??????(????????????
2.71
Τ
1
??????) compiler time
•Can be generalized to n-qubits/????????????(??????) at exp. scaling
Experimentalist: No!
DiVincenzo criteria:
•Scalable (bounds circuit max. width), well-
characterized qubits
•Long decoherence time (bounds circuit max. depth)
•Universal native gate set
•Initialization to fiducial state
•Measure individual qubits
Qubit plane architecture:
•Connectivity bounds information interaction rate
between qubits. Needs additional gates for routing
(taking some share of max. depth)
•Not all qubit noise are uniform. Noise drift.

27. Cryogenic-CMOS for Quantum Computing
Energy-efficient Quantum Instruction Set Architecture (EQISA)

≡ Quantum Circuit Description Complexity
≡ bits required to erase/uncompute a qubit to a known state
≡ least no. of control bits to prepare the current state from a known state
≡ coding and compressing scheme for quantum circuits
•e.g., CISC for low-power embedded systems
•trade off 4K-RT compressed ins. bw with ASIC/FPGA proc. @ 2.46pJ/b, 40Gb/s
Coding + Concept Discovery
Ver-0: Binary coded Native Gate Set
Ver-1: Huffman coded Native Gate Set
Ver-2: Huffman coded SK Basis Approx.
Ver-3: Huffman coded SK Basis Approx. + cutoff

28. Energy-efficient Quantum Optimal Control (EO-GRAPE & EO-DRLPE)

Quantum Speed Limits for U
•Discrete: Σ (# gates * gate-time)
•Continuous: Energy ≡ Geodesics on Riemannian manifold ≡ Lagrangian mechanics
Trade off fidelity and energy for pulse-level gate synthesis using
Energy-optimized Gradient Ascent Pulse Engineering (EO-GRAPE)
8% ↓ energy
1% ↓ fidelity (< P
thresh)
U = H
w
fe = [0.8, 0.2]

29. Quantum Information Theoretic Compilation

30. Generative AI for Quantum Computer Engineering

31. eXplainable Quantum Machine Learning

32. …others

❑A resource-efficient variational quantum algorithm for mRNA codon optimization
❑Efficient parameterised compilation for hybrid quantum programming
❑Efficient decomposition of unitary matrices in quantum circuit compilers
❑LEGO_HQEC: A Software Tool for Analyzing Holographic Quantum Codes
❑Near-term spin-qubit architecture design via multipartite maximally entangled states
❑ArtA: Automating Design Space Exploration of Spin-Qubit Architectures
❑CutQAS: Topology-aware quantum circuit cutting via reinforcement learning
❑A Scalable Quantum Gate‐Based Implementation for Causal Hypothesis Testing
❑Visualizing quantum circuit probability: estimating quantum state complexity for quantum program synthesis
+ current research at Fujitsu

33. ✓The best model of computing allowed by current laws of physics.
✓Business advantage likely decades away. Scientific use cases nearer.
✓Interdisciplinary expertise required in both theory and engineering.
✓Rapidly evolving, need to stay up to date.
✓Get started with quantum programming via hackathons and summer schools.
✓AI and QC synergy will be crucial for both fields.
✓Beware of the hype! But don’t be ignorant of the enormous potential.
Takeaways