Python orchestration, primitive compilation, deterministic simulation, TTL/photon emulation and traceable outputs.
Quantum Control Software Engineer
Trapped-ion control software from pulse intent to measurable evidence.
Deterministic timing, hardware adapters, photon acquisition, and evidence-gated calibration.
Implemented with ARTIQ/Kasli-SoC; architected for portable FPGA and AWG/DDS backends.
01 · TIMING
Measurement-aware pulse validation
Specify what real captures must prove across pulse width, alignment, phase window and repeatability.
Width · alignment · phase · repeatability 02 · ARCHITECTUREPhysics requirements → maintainable control contracts
Confirm experiment semantics with physicists, then isolate hardware and scale changes behind typed registries and adapters.
Shared requirements · bounded change · evidence 03 · VERIFICATIONSoftware simulation + electrical emulation
Exercise the same contracts through deterministic simulation and TTL/photon hardware before any ion claim.
No-ion failure-path coverage 04 · FULL STACKOpenQASM → hardware → PMT evidence
Preserve circuit identity through primitive compilation, backend execution, photon acquisition and analysis.
Traceable circuit-to-photon ownership 05 · SCALE UPScalable control through bounded modules
Separate the global registry from active real-time working sets, controller shards and evidence-owned hand-offs.
Synthetic software envelope · physical throughput N/RM4i.66xx AWG/DDS adapter, multi-tone carrier control, SSB upconversion and hardware-specific APIs.
No personal Spectrum operation, eleQtron proprietary topology or physical-ion gate-fidelity claim.
05 · Evidence-gated calibration
Close the loop without weakening the evidence.
A scan is not a calibration update. The system must identify stale dependencies, run the right experiment, accept a server-owned fit, stage a bounded candidate and preserve why it was accepted or rejected. Only a separately trusted real-measurement boundary may promote a value into the physical baseline.
A validated ten-node graph drives dependency planning, scoped execution, fitting, staged values and recalibration order.
Both producing chains reach stale = 0. Rabi then returns an empty plan; the MS parity check remains intentionally callable.
Request payloads cannot label themselves “real.” Synthetic candidates remain staged evidence and never enter the baseline.
No commissioned detector, approval provider, physical rollback path or ion-derived calibration is claimed here.
01 · INSPECTEvaluate freshnessTTL, upstream hashes, scope and current tier.
02 · PLANOrder dependenciesBuild a bounded ancestor-first maintain plan.
03 · MEASURERun the scanUse a typed axis, deterministic seed and server-owned run.
04 · ANALYSEFit onceConsume the registered fitter output; do not trust client values.
05 · DECIDEApply acceptanceCheck R², bounds, max delta and failure class.
06 · STAGERecord the candidateRetain run ID, graph identity, source and reason.
07 · REPEATRe-plan until stableStop when producing nodes are fresh or a gate fails.
Frequency first, then π-time.
qubit_freq[q0] makes the drive prerequisite current before rabi_pi_time[q0] fits and stages a bounded per-qubit π-time. A second producing-node plan is empty.
Two Rabi prerequisites, then the pair.
Both qubits must be current before sideband and detuning fits, derived gate time and parity-fringe verification. The parity target is check-only and deliberately runs on every request.
INTERACTIVE SOFTWARE-SIMULATED EVIDENCE
REFERENCE LANE READY
Rabi AutoCal · q0 π-time
25 points · 200 shots · seed 20260806
Axis: block:gate_operation.duration_us
Synthetic truth: tπ = 4.4 µs · contrast = 0.95 · offset = 0.02
INTERACTIVE SOFTWARE-SIMULATED EVIDENCE
PAIR LANE READY
MS gate AutoCal · q0–q1
33 displayed · 32 fitted · 200 shots
Axis: block:gate_operation.phase_turns
Synthetic result: parity amplitude = 0.8573 · phase = −0.0190 rad · R² = 0.9937
Declarative dependency graph
Ten nodes, three scopes, one planner.
Global, per-qubit and per-pair calibrations share one graph contract. Click a node to inspect its procedure and evidence boundary. The graph describes software ordering and acceptance; its reference values are synthetic.
REFERENCE GRAPH · tisl.autocal-graph/v1
GLOBAL
PER QUBIT
PER PAIR
Failure behaviour is part of the product
Reject a plausible answer for the right reason.
AutoCal is valuable only if uncertainty, stale state and authority failures stay visible. The reference implementation makes each failure terminal or retryable by policy instead of quietly averaging it into the next baseline.
A smooth-looking fit can still fail.
R² below the declared threshold rejects the candidate. Re-running identical data is not treated as recovery.
Result: explicit terminal rejectionTransport faults are not physics faults.
A temporary submission failure may retry within the recipe budget; the report retains every attempt and final state.
Result: retry class remains inspectableChanged inputs invalidate downstream values.
TTL expiry or an upstream content-hash change places ancestors back into dependency-first recalibration order.
Result: stale values are refusedMovement is recorded, not hidden.
A 0.585 µs fitted shift crosses the 0.5 µs drift threshold but remains inside the 2.0 µs max delta.
Result: staged with a drift reasonBounds protect the current value.
A 3.0 µs jump from the current staged value exceeds max delta; the proposal is rejected and the store remains unchanged.
Result: failed_acceptanceA request cannot grant itself authority.
Changing a payload field to “real” does not create trusted evidence. The bundled service keeps commit disabled.
Result: baseline stays emptyWhat this portfolio can demonstrate now
- Validated declarative graph, scope instances and dependency-first maintain plans.
- Registered
scan_fit,checkandderiveprocedures with server-owned fit outputs. - Staged and baseline views, append-only decision history, TTL and upstream-hash freshness.
- Bounded retry, loop and run/shot/wall budgets with deterministic reports.
- Rabi frequency → π-time AutoCal with small-drift annotation, max-delta rejection and an empty producing-node second plan.
- Two-qubit Rabi prerequisites → MS sideband → detuning → derived gate time → parity verification in a reproducible cold-start artifact.
What still requires a real laboratory boundary
- A commissioned device adapter, PMT/camera mapping and independently verified measurement provenance.
- A server-side trusted-real provider; client requests must never manufacture this capability.
- Approval, interlock, safe-state and rollback procedures accepted on the installed system.
- Physicist-reviewed recipe ranges, fit models, thresholds and validity windows for the actual apparatus.
- Physical-ion Rabi π-time, MS detuning, DDS amplitude or phase writeback, state preparation and gate-fidelity evidence.
Evidence boundary: every value, trace, threshold and drift scenario on this page is a deterministic synthetic reference used to test the software control loop. It demonstrates orchestration, provenance and rejection behaviour—not the true calibration of an ion, an installed eleQtron system or any gate-fidelity result. The displayed parity amplitude is a software acceptance metric; it is neither Bell-state fidelity nor a calibrated DDS amplitude.
01 · Measurement-aware timing validation
A timing claim is not complete until the scope agrees.
The interactive trace is a literature-informed model of the intended sequence. It defines the signals, timing relationships and acceptance questions that a real capture would need to verify.
Published parameter
MW Qubit 1 · Gate operation
Ref [2] · PRX 15, 021079 (2025)
- A / ΩAmplitude
- Ω1 / 2π = 94.8 kHz
- f / ωFrequency
- fμw ≈ 12.6 GHz
- τDuration
- 313 µs
- φPhase
- φ(t) = 0.749 sin(2π · 94.8 kHz · t)
Notation: A amplitude · Ω/2π Rabi rate · f frequency · ω/2π angular-frequency form · τ duration · φ phase · Δ detuning
Each channel has a fixed color. Pulse height is an ON/OFF envelope unless a paper reports intensity or Rabi frequency; stage widths are visually compressed.Evidence boundary: this is a composite literature model, not one experimental shot and not eleQtron proprietary calibration. Ref [1] reports the cooling sequence at an axial frequency of 117.48(11) kHz; Ref [2] reports the MAGIC gate in a different setup at 98.08 kHz. Unpublished hardware settings are shown as N/R (not reported).
References: [1] Sriarunothai et al., J. Mod. Opt. 65, 560–567 (2018) · [2] Nünnerich et al., Phys. Rev. X 15, 021079 (2025) · [3] eleQtron · MAGIC uses radio-frequency control
Public-evidence system study
The scope is an engineering sidecar—not the shot-loop data plane.
University of Siegen apparatus descriptions place deterministic sequencing, detector acquisition and state discrimination inside the experiment loop. Oscilloscopes observe selected electrical nodes through probes or splitters for commissioning, validation and debugging.
01 · SchedulePython or host software compiles the experiment; a real-time sequencer owns TTL triggers and deterministic I/O.
02 · DriveAOM/EOM channels prepare and detect; AWG/DDS plus microwave conversion deliver coherent control.
03 · Fluoresce369 nm emission passes collection optics, spatial filtering and a narrow-band optical filter.
04 · AcquireEMCCD, PMT or SPAD data enters a frame grabber, counter, TDC, ADC or FPGA for classification and feedback.
- Laser lockPhotodiode, cavity transmission and error signal.
- RF / microwaveIF envelope, trigger, mixer or crystal-detector output.
- Clock / triggerTTL fan-out, camera-ready and AWG trigger alignment, skew and jitter.
- Detector front endPulse amplitude, noise, threshold, saturation and signal integrity.
Software boundary
Python/SCPI can configure the scope, acquire waveforms and emit commissioning, calibration or monitoring artifacts. The real-time controller still owns shot scheduling, while detector interfaces and analysis software own structured photon counts, frames, timestamps and bright/dark decisions.
Not claimed: eleQtron's current oscilloscope vendor, bandwidth, sample rate, channel count or installed topology are not public.System-study references: [4] Piltz, 2016 · EMCCD/PMT triggering and ADwin sequencing · [5] Huber, 2024 · AWG hand-off and image feedback · [6] eleQtron · Python control software and programmatic device control · [7] DLR QSea II · SPAD and control-electronics integration direction
Requested window, sample depth and scope readback must agree. A plausible trace from a shortened pre-trigger record is rejected.
All channels must share one record length, and the expected dark window must prove that the capture contains the intended shot phase.
Repeated captures must agree numerically. A conflicting capture stays a disagreement; it is never averaged into a green result.
02 · Extensibility and maintainability
Scale qubits by adding contracts—not branches.
The orchestration core should not care whether control stays in one zone, crosses a shuttling boundary or reaches another module through a photonic link. Hardware-specific behavior belongs behind registries and adapters.
Core mechanisms implemented Shuttling & photonic links: architecture targetsOne source of truth
Process, dataset and backend registries define what exists. A new lane cannot silently omit its timing probe, readback contract or safety gate.
Hardware owns its I/O
The implemented control contract keeps orchestration separate from device I/O. Spectrum M4i.66xx/DDS is modelled as an eleQtron-facing adapter target; shuttling and photonic links enter through the same seam—not an if qubits > N rewrite.
Agents can change it safely
Code maps identify ownership; contract tests detect registry drift; fail-closed admission blocks incomplete capability or stale runtime evidence before execution.
CURRENT CONTROL MODELSingle control regionRegistry-driven qubit, pair, PMT and calibration contracts.
EXTENSION TARGETMulti-zone shuttlingTransport plan becomes a scheduled adapter with explicit timing and hand-off evidence.
EXTENSION TARGETPhotonic module linkRemote entanglement becomes another typed execution/readout lane with heralding provenance.
Agent maintenance loop
- 1 · Read contractLocate registry, dataset owner and evidence boundary.
- 2 · Add adapterKeep device-specific I/O outside orchestration.
- 3 · Run guardsCatch drift, stale mappings and unsupported paths.
- 4 · Emit evidenceRetain provenance, limitation and terminal outcome.
03 · Timing verification system
Test the same contract through two independent realities.
A fast deterministic simulator checks logic and data contracts. An electrically live lane runs the production hardware-timed control core with real TTL/DDS envelopes and a controlled photon source—still without ions.
Verification lanes implementedDeterministic software simulation
Compile the exact primitive OpenQASM, execute synthetic counts and verify repeatable evidence without lab network access.
- Same primitive circuit and execution contract
- Deterministic replay and evidence hash
- Freshness, all-dark and capability failures are injectable
- Proves software behavior—not RTIO or wiring
Electrically live control path
The production control core runs on the test master; a scope observes physical 369/935 envelopes while controlled TTL photon pulses exercise the counter/readout chain.
- Real FPGA-backed scheduling and deterministic timeline
- Scope-visible DDS/TTL timing envelopes
- Controlled photon counts through the acquisition boundary
- Proves electrical/dataflow integration—not ion physics
Fail-closed evidence ladder: software-sim results cannot enter calibration; electrically live results cannot claim state preparation or gate fidelity; physical-ion verified capability remains explicitly separate.
04 · Circuit-to-photon full stack
Own every boundary from algorithm to evidence.
The system does not stop at compilation or pulse generation. It carries circuit identity through execution, photon acquisition, analysis and a claim that includes its own limitation.
OpenQASM → control contract → counts implemented Spectrum M4i.66xx AWG/DDS: integration targetThe diagram distinguishes implemented software contracts from the publicly documented eleQtron microwave backend, which is presented as an integration target rather than personal hands-on evidence. Horizontal scrolling is available on narrow screens.
PUBLIC ELEQTRON QUBIT-CONTROL PATH · REF [4]
Python contract to microwave carrier.
Python / control services
↓
Spectrum M4i.66xx AWG + M4i.66xx-DDS
↓
SSB mixer + microwave LO
↓
fμw ≈ 12.64 GHz → 171Yb+ MAGIC processor
Spectrum’s official eleQtron case study identifies the M4i.66xx series as the microwave qubit-control AWG family. The control-software and FPGA boundaries are consistent with eleQtron’s public engineering roles: Python APIs/data interfaces above hardware-specific I/O, and FPGA signal generation plus real-time feedback below.
fkcarrier frequency
φkcarrier phase
Akcarrier amplitude
dfk/dtfrequency slope
dAk/dtamplitude slope
16 bit · up to 1.25 GS/sPCIe AWG family; one, two or four synchronous channels.
k = 1…20 carriers/outputEach carrier exposes fk, φk, Ak, dfk/dt and dAk/dt.
Δtcmd ≈ 6.4 nsfμw ≈ 12.64 GHz after SSB mixing; Δfion ≈ 3–5 MHz.
References: [4] Spectrum Instrumentation · DDS technology enables microwave ion control for quantum computing · [5] eleQtron · Senior Software Engineer—Python control software · [6] eleQtron · FPGA Engineer—Quantum Computing
Claim boundary: the public source confirms the M4i.66xx series, not an exact installed submodel. M4i.6631 appears in the article image caption only. No claim is made about eleQtron’s laser, camera, trap-DAC or laboratory-wide sequencer stack.
Selected research · application bridge
Translate application questions into executable quantum workflows.
Collaborations across quantum machine learning, generative chemistry and quantum chemistry gave me a working vocabulary on both sides of the interface: domain objectives and benchmarks on one side; circuits, simulators, hardware constraints and defensible evidence on the other. Google Scholar profile ↗
01 · ASKApplication objective & success metric
02 · MODELPhysics assumptions & algorithm choice
03 · EXECUTECircuit, backend & resource constraints
04 · EXPLAINEvidence, benchmark & limitations
Unentangled quantum reinforcement learning agents in the OpenAI Gym
Application → experimentTurned standard reinforcement-learning tasks and metrics into a single-qubit variational workflow, classical post-processing and execution on real IBM quantum machines.
Read the arXiv preprintExploring the Advantages of Quantum Generative Adversarial Networks in Generative Chemistry
Domain objective → hybrid modelConnected small-molecule generation goals to hybrid quantum-classical GAN components, then compared physicochemical, goal-directed and validity trade-offs.
Read the ACS paperQuantum simulation of preferred tautomeric state prediction
Scientific workflow → resource constraintsMapped a drug-discovery question through active-space selection, qubit-efficient encoding and VQE while retaining benchmark and hardware-resource limits.
Publisher-stated contribution: performed noiseless and noisy quantum simulations with Yu Shee.
Read the npj paperThese publications evidence cross-domain collaboration in quantum applications. They complement—but do not replace—the separately labelled software and electrical evidence for the trapped-ion control system above.