---
title: Pulse-Accelerated Qubit Excitations
url: https://www.emergentmind.com/topics/pulse-accelerated-qubit-excitations
type: topic
---

# Pulse-Accelerated Qubit Excitations

Pulse-accelerated qubit excitations are control protocols in which the time required to rotate, invert, transfer, reset, or entangle qubit states is reduced by engineering the drive at the pulse level rather than accepting fixed microwave templates or long gate decompositions. Across superconducting transmons, semiconductor charge and spin qubits, NV-center hybrid registers, and silicon devices, the recurring objective is the same: accumulate the desired excitation rapidly while suppressing leakage, power broadening, off-resonant driving, crosstalk, and decoherence exposure. The subject therefore spans shaped analog envelopes, sub-harmonic and cross-resonant drives, single-flux-quantum pulse trains, and direct optimal-control schedules [2306.10162] [2501.10710] [1508.00821] [1902.02911] [2311.08742].

## 1. Control principles and mathematical structure

A large fraction of the literature formulates pulse-accelerated excitation in the rotating-frame language of a driven two-level or weakly anharmonic multilevel system. In a representative shared-line transmon model, the effective Hamiltonian is written as
\[
\hat{H} = -\frac{1}{2}\sum_j \Delta_j \hat{\sigma}_{j,z} +\frac{1}{2}\sum_j\left[s_x(t)\hat{\sigma}_{j,x}-s_y(t)\hat{\sigma}_{j,y}\right],
\]
with \(\Delta_j=\omega_j-\omega_{\mathrm d}\) and complex envelope \(s(t)=s_x(t)+is_y(t)\). In the simplest gate-calibration setting, the target rotation is enforced through a pulse-area constraint such as
\[
\int_0^T s(t)\,dt = \pi/2,
\]
so acceleration is immediately coupled to the time-bandwidth problem: shortening \(T\) broadens the pulse spectrum and increases overlap with unwanted transitions [2501.10710].

An equivalent pulse-area logic appears in smooth \(\pi\)-pulse design for disordered qubit ensembles. For a driven two-level system with Hamiltonian
\[
H=\frac{\delta}{2}\sigma_z+\frac{f(t)}{2}\sigma_x,
\]
the resonant solution depends on
\[
\varphi(t)= \int_0^t f(t_{1})\, dt_{1},
\qquad
\varphi(\tau)=\pi
\]
for a \(\pi\)-pulse. That work turns acceleration into an optimization problem: choose a smooth envelope \(f(t)\) so that the final inversion error is flat in detuning order by order, yielding \(x(\tau)=O(\delta^N)\) and residual ground-state occupation
\[
n_0(\tau,\delta)\propto \left(\frac{\delta}{\Omega}\right)^{2N},
\]
with the numerical fit
\[
n_0(\delta)=1-n_\uparrow \approx 34 \left( 0.16 \frac{\delta}{\Omega}\right)^{2N}.
\]
The underlying message is that acceleration is not merely a matter of larger instantaneous amplitude; it is a matter of redistributing action in time so that detuning sensitivity cancels rather than accumulates [1508.00821].

Pulse trains add a second general mechanism: coherent accumulation. In a driven tunneling qubit with
\[
\hat{H}(t)=-g(t)\hat{\sigma}_{z}+\Delta\hat{\sigma}_{x},
\qquad
g(t)=\frac{1}{2}\varepsilon_{0}+A(t)\cos(\omega t+\theta),
\]
the resonant \(N\)-pulse dynamics reduce to an effective area built from a Bessel-dressed coupling,
\[
P_{2}(t)=\sin^{2}\left(\Delta\int_{0}^{t}J_{N}\left(\frac{2A(t')}{\omega}\right)dt'\right).
\]
For periodic trains the dynamics acquire quasienergies
\[
E_{N} = \frac{\Delta}{\tau}\int_{-\infty}^{\infty}J_{N}\left(\frac{2A_{k}(t)}{\omega}\right)dt,
\]
so repeated pulses can achieve complete inversion even when a single pulse at the same higher-order resonance is weak. This suggests a broad taxonomy: pulse-accelerated excitation can arise from pulse-area optimization, spectral shaping, or coherent multi-pulse accumulation rather than from any single control trick [1503.01665].

## 2. Superconducting transmons: short gates, spectral selectivity, and shaped microwave control

In superconducting transmons, pulse acceleration is tightly constrained by weak anharmonicity and crowded spectra. One route is sub-harmonic control. A transmon driven near \(\omega_d\approx \omega_q/3\) can exploit the quartic Josephson nonlinearity to generate an effective three-photon drive. The resulting rotating-frame Hamiltonian is
\[
\hat H^R_{\rm sub}/\hbar = (2\alpha|\eta|^2-3\delta)\, q^\dagger q +\frac{1}{2}\alpha q^\dagger q^\dagger q q +\frac{1}{3}\alpha(\eta^3 q^\dagger+\eta^{*3}q),
\]
so the useful drive rate scales cubically with the applied amplitude. Experimentally, this produced \(\pi/2\) gates as fast as \(35\) ns, with interleaved randomized-benchmarking fidelities \(F(X_\pi)=99.4\%\), \(F(X_{\pi/2})=99.7\%\), \(F(Y_\pi)=99.5\%\), and \(F(Y_{\pi/2})=99.7\%\). The same study reports that a \(1\%\) drift in drive amplitude causes about a \(3\%\) drift in the effective gate amplitude, so acceleration through nonlinear pumping comes with amplified calibration sensitivity [2306.10162].

A second route addresses the selectivity loss that accompanies fast pulses in frequency-multiplexed architectures. The selective-excitation-pulse method shapes the frequency-domain envelope so that
\[
S(\omega_j)=0
\qquad \text{for all non-target qubits } j,
\]
thereby placing spectral nulls at spectator frequencies. In a three-transmon shared-line experiment, \(40\) ns \(X_{\pi/2}\) gates were implemented with \(\sigma/2\pi=25\) MHz. Single-qubit randomized benchmarking yielded average Clifford fidelities \(99.87(1)\%\), \(99.80(2)\%\), and \(99.82(2)\%\) for the three qubits, essentially matching Gaussian-pulse performance, while spectator excitation rates were reduced from \(10\%\) and \(7\%\) to \(0.2\%\) and \(0.01\%\). The result is acceleration without surrendering addressability in the regime where detunings are only tens of MHz [2501.10710].

The same speed-selectivity problem appears in the presence of weakly detuned spectators and microscopic two-level defects. The spectrally balanced pulse, implemented as a dual-DRAG or recursive-DRAG construction,
\[
\Omega^{(2)} = \Omega^{(0)} + \frac{\ddot{\Omega}^{(0)}}{\Delta^2},
\]
places symmetric spectral holes at \(\omega_0\pm \Delta\) without shifting the pulse center frequency. At \(\Delta_0/2\pi=45\) MHz, \(g/2\pi=1\) MHz, and \(t_g=25\) ns, the error per Clifford on the driven qubit improved from \(1.60\times10^{-3}\) to \(1.42\times10^{-3}\), while spectator excitation per Clifford dropped from \((1.6\pm0.2)\times10^{-4}\) to \((0.8\pm0.3)\times10^{-5}\). In the strongly coupled TLS setting, the same method produced order-of-magnitude suppression for \(20\)–\(50\) ns gates over detunings \(20\)–\(85\) MHz, but it ceased to work below \(20\) MHz because the required spectral holes overcut the target band itself [2502.10116].

Spectral cleanliness can also be studied at the level of line shapes rather than gates. On IBM transmons, analytic models for rectangular, Gaussian, \(\mathrm{sech}\), \(\mathrm{sech}^2\), and exponential pulses showed that pulse shape strongly changes the transition profile \(P(\Delta)\). The mean absolute error of analytic fits improved by factors of \(4\) to \(8\) relative to Lorentzian fitting, and the uncertainty in the extracted qubit resonance frequency was reduced by a factor of \(4\) [2301.10004]. A complementary analysis of smooth finite-duration pulses that start and end linearly in time showed substantially weaker power broadening and much faster-decaying wings than rectangular pulses: the line wings scale as \(P\propto \Delta^{-4}\) rather than \(\Delta^{-2}\), and for a \(\pi\)-area pulse the first satellite drops from \(11.64\%\) for a rectangular pulse to \(1.48\%\) for a sine pulse, with the first sideband shifted from \(\Delta\approx \pm 8.42/T\) to \(\Delta\approx \pm 11.1/T\) [2404.12236]. In this part of the literature, acceleration means not maximal temporal compression alone, but efficient finite-time excitation with sharply reduced spectral contamination.

## 3. Digital pulse trains, active reset, and hardware-efficient excitation pathways

A different branch of the field replaces analog microwave envelopes with digitally timed impulses. In the single-flux-quantum approach, each pulse has width about \(2\) ps and quantized area
\[
\int_{-\infty}^{\infty} V_{\rm SFQ}(t)\,dt = \Phi_0 = \frac{h}{2e}.
\]
In a three-level transmon model, a single SFQ pulse acts as a small \(y\)-rotation
\[
U^{(3)}_{\rm SFQ} = \exp\!\left(-\frac{i\delta\theta\,\hat{\Sigma}_y}{2}\right),
\]
and useful control is obtained by choosing a binary clocked bitstream \(\mathbf S\) whose pulses are synchronized to free evolution. The SCALLOPS framework uses a \(25\) GHz global clock, subsequences of \(35\)–\(55\) bits, and repeated streaming of compact registers. It reports \(>99.99\%\) single-qubit gate fidelity for \(21\) distinct qubit frequencies, typically in under \(12\) ns, with transient \(|2\rangle\) population reaching \(\sim10^{-2}\) mid-sequence but returning below \(10^{-4}\) at the end of each subsequence repetition [1902.02911].

Pulse acceleration can also be directed toward state removal rather than gate synthesis. In fixed-frequency transmons, a pulsed reset protocol deliberately uses higher excitation as a resource:
\[
\ket e \xrightarrow{X_\pi^{e\to f}} \ket f
\quad\text{then}\quad
\ket{f0}\xrightarrow{\text{drive}} \ket{g1}\xrightarrow{\kappa_r}\ket{g0}.
\]
The first pulse transfers \(\ket e\) to \(\ket f\) in \(75\) ns; the second pulse activates \(\ket{f0}\leftrightarrow\ket{g1}\) transfer in \(120\) ns. The pulse sequence itself lasts \(210\) ns, and the full reset including resonator emptying takes \(2.21\ \mu\)s. Best residual populations of \(1.7\pm0.1\%\) were reported, and at low trigger rate the qubit population after reset was \(0.43^{+0.16}_{-0.20}\%\), lower than the thermal equilibrium value \(1.5\pm0.3\%\). At a \(10\) kHz trigger rate, the residual population was \(1.2\%\) with pulsed reset versus \(11.0\%\) without it [1802.08980]. This directly corrects the common misconception that higher transmon levels are only deleterious leakage channels; in reset engineering, controlled access to \(\ket f\) is the acceleration mechanism.

## 4. Semiconductor, spin, and hybrid-qubit implementations

Semiconductor spin systems provide some of the clearest examples of genuinely nanosecond electrical pulse acceleration. The resonant exchange qubit uses three-electron exchange both as the static splitting and as the oscillatory transverse drive. In the logical qubit subspace the effective Hamiltonian is
\[
\mathcal{H}(\varepsilon) = -\frac{J_z}{2}\sigma_z - \frac{J_x}{2}\sigma_x,
\]
with \(J_z=\frac12(l+r)\) and \(J_x=\frac{\sqrt3}{2}(r-l)\). At the sweet spot \(\varepsilon=0\), \(dJ_z/d\varepsilon=0\) while \(dJ_x/d\varepsilon\neq 0\), so rf gate-voltage bursts produce resonant control without first-order longitudinal sensitivity to detuning noise. Experimentally this yielded a \(\pi/2\)-gate time of \(2.5\) ns and a coherence time of \(19\ \mu\)s under multipulse echo [1304.3413].

A related idea in disordered ensembles is synchronized excitation under optimized smooth \(\pi\)-pulses. Using the finite sine-series ansatz
\[
f(t)=\Omega\sum_{n=1}^{N}k_{2n-1}\sin\!\big((n-\tfrac12)\Omega t\big),
\]
the pulse is engineered so that the first \(N-1\) coefficients in the detuning expansion of the residual ground-state amplitude vanish. For \(N=3\), the residual error scales as \((\delta/\Omega)^6\); for \(N=4\), the paper reports \(n_\uparrow \approx 1-10^{-5}\) for qubit frequencies spanning roughly \(\epsilon\in \omega \pm \Omega\). This is pulse acceleration in the sense of replacing slow adiabatic or multi-pulse refocusing strategies with a single short robust inversion event [1508.00821].

Single-electron charge qubits in semiconductor double quantum dots illustrate a different constraint: finite rise and fall times. In the effective logical basis
\[
\hat{H}_\mathrm{eff}(t) = -\frac{1}{2} \,\epsilon(t) \,\sigma_{z} + \frac{1}{2} \,\Delta \,\sigma_{x},
\]
the fastest high-fidelity control is not obtained by ideal square pulses but by numerically corrected multipulse sequences at \(\epsilon=\pm\Delta\), which realize rotations about two orthogonal tilted axes \(\vec z'\) and \(\vec x'\). Full-wavefunction simulations report final-state fidelities \(>99.99\%\) for optimized pulse sequences, gate errors below \(10^{-4}\) over the Bloch sphere for \(\pi\)-rotations, and up to \(250\%\) gain in fidelity for an \(R_x\) rotation by \(\theta=\frac{3}{2}\pi\) in a quasistatic-noise model [2303.04823]. Here acceleration is constrained by finite pulse rise time and higher-orbital excitation, so speed comes from composite structure rather than from impulsive detuning excursions.

Hybrid spin registers in NV centers show still another pattern: replace slow transition-selective conditional pulses with hard resonant rotations plus free precession under the internal hyperfine Hamiltonian. For the electron-\(^{14}\mathrm N\) system, a \((\pi/2)_x-\tau-(\pi/2)_y\) sequence with \(\tau=1/(2|A|)\) implements CNOT-like behavior in about \(300\) ns, compared with about \(1.8\ \mu\)s for the conventional transition-selective version. In a three-spin register containing a passive \(^{13}\mathrm C\), the new transfer sequence increased the observed \(^{14}\mathrm N\) Rabi-oscillation amplitude by \(21\%\), and for weakly coupled electron-\(^{13}\mathrm C\) logic the method achieved fidelities \(0.99\) and \(0.95\), whereas a selective-pulse realization would have required a \(25\ \mu\)s pulse, much longer than the measured \(T_2^*\approx 10\ \mu\)s [1806.08408]. The acceleration mechanism is thus interaction-assisted free precession, not merely stronger driving.

## 5. Pulse-level optimal control in chemistry, entanglement preparation, and compilation

Pulse acceleration has become especially visible in algorithmic contexts where gate decomposition overhead is dominant. In silicon spin-qubit quantum chemistry, a modular GRAPE-based strategy directly compiles single- and double-qubit excitation operators,
\[
A_{ik}(\theta) = \exp\!\left[\theta\left(Q_i^\dagger Q_k - Q_k^\dagger Q_i\right)\right],
\qquad
A_{ijkl}(\theta) = \exp\!\left[\theta\left(Q_i^\dagger Q_j^\dagger Q_k Q_l - Q_l^\dagger Q_k^\dagger Q_j Q_i\right)\right],
\]
into exchange-only control pulses. The reported minimal pulse times are \(\le 289\) ns for single-qubit excitations and \(\le 927\) ns for double-qubit excitations, compared with \(446\)–\(3212\) ns and \(2549\)–\(14189\) ns for gate-based implementations, giving acceleration factors of \(1.5\)–\(11.1\) and \(2.7\)–\(15.3\), respectively [2606.17357]. The significance is architectural: pulse acceleration is applied to reusable chemistry primitives rather than to a full ansatz, so the speedup remains local and modular as system size grows.

Direct superconducting pulse optimization for molecular ground states pushes the timescale much further. In a compact VQ-SCI encoding of \(\mathrm H_2\), a freestyle pulse of \(0.22\) ns was reported to reach chemical accuracy on real hardware, compared with \(71\) ns for a gate-model \(R_y\) baseline on the same one-qubit encoding. For \(\mathrm{LiH}\), a \(208\) ns real-hardware freestyle pulse on three qubits achieved \(\Delta_{\mathrm{FCI}}=4.96\) mHa, compared with a \(1250\) ns gate-model baseline on the same VQ-SCI encoding [2403.17789]. This is one of the few places where the literature explicitly associates pulse acceleration with a quantum-speed-limit analysis: for the \(\mathrm H_2\) pulse the experimentally reached rotating-frame quantum speed limit was computed as \(0.222222\) ns, equal to the active one-bin pulse duration.

Resource-oriented pulse control extends the same logic to entanglement preparation. In transmon systems with drift-plus-control Hamiltonian
\[
H(t)=H_{\text{sys}}+\sum_i \mu_i(t)H_i,
\]
differential-evolution optimization was used to target entanglement resources directly rather than canonical state vectors. For two qubits the reported negativity was \(\mathcal N(\rho_{01})\approx0.499\); for three-qubit GHZ-class preparation the three-tangle reached \(\tau_3\approx0.999\); and for W-class preparation the squared concurrences were \(\mathcal C^2(\rho_{01})\approx0.444\), \(\mathcal C^2(\rho_{02})\approx0.442\), and \(\mathcal C^2(\rho_{12})\approx0.442\). The Gaussian-square pulse schedules were shorter than compiled gate implementations in all three cases: \(1379.0\,dt\) versus \(2912\,dt\) for Bell-type preparation, \(3750.0\,dt\) versus \(5315\,dt\) for GHZ, and \(6132.0\,dt\) versus \(8224\,dt\) for W [2602.20275]. The common principle is that acceleration improves when the control objective is the resource actually needed, not an overconstrained circuit realization.

Compiler-driven pulse acceleration on IBM hardware takes a more calibration-centric form. The sQueeze framework introduces live-calibrated basis gates \(R_x(\theta)\) and \(R_{zx}(\theta)\) at the pulse level, with the \(R_x\) amplitude obtained from
\[
A_1 \sin^2(\omega A + \phi) + \delta = \sin^2\Bigl(\frac{\theta}{2}\Bigr),
\qquad
A = \frac{\sin^{-1}\Biggl(\sqrt{\frac{\sin^2\Bigl(\frac{\theta}{2}\Bigr)-\delta}{A_1}\Biggr)-\phi}{\omega}.
\]
Across IBM devices, \(R_x(\theta)\) gates were \(52.7\%\) more accurate on average than native Qiskit decompositions, \(R_{zx}(\theta)\) were \(22.6\%\) more accurate on average, and speed-ups reached \(4.1\times\) for single-qubit operations and \(3.1\times\) for two-qubit gates. Benchmark algorithms saw up to \(39.6\%\) improvement in fidelity [2311.08742]. In this part of the literature, pulse acceleration is inseparable from live recalibration and compiler integration.

The computational bottleneck of pulse optimization has itself become a target. Qalibrate reformulates pulse synthesis for three-electron exchange-only spin qubits through a Liouville-space propagator and a parallel Magnus-expansion construction,
\[
\frac{d}{dt}|\rho(t)\rangle\rangle = \mathcal{L}(t)|\rho(t)\rangle\rangle,
\qquad
U(t)=\mathcal{T}\exp\left(\int_0^t \mathcal{L}(\tau)\,d\tau\right),
\]
and reports up to \(34\times\) speedup in propagator generation for three quantum dots relative to an ODE-based baseline. The study did not yet produce full pulse sequences for three or more dots because autodiff memory overhead remained limiting, but it established accelerated simulation and calibration as a prerequisite for accelerated pulse design [2511.13330].

## 6. Tradeoffs, misconceptions, and open directions

The literature is unusually consistent on one point: pulse acceleration is fundamentally a tradeoff, not a free gain. Shorter pulses broaden bandwidth, and this broadening creates unwanted excitation unless the spectrum is deliberately shaped. That logic is explicit in frequency-multiplexed transmon control, where pulse durations of a few tens of nanoseconds imply spectral widths of tens of MHz, making a \(26\) MHz spectator detuning especially problematic for a \(40\) ns pulse [2501.10710]. It is equally explicit in the spectrally balanced-pulse work, which shows that the method fails when \(\Delta_0/2\pi<20\) MHz because the notches begin to erase the target band itself [2502.10116]. A plausible implication is that future acceleration strategies will continue to converge on spectrum-aware control rather than on amplitude escalation alone.

A second recurring theme is calibration fragility. Sub-harmonic transmon control amplifies amplitude noise because the effective Rabi rate scales cubically, so a \(1\%\) drive-amplitude drift becomes a \(3\%\) gate-amplitude drift [2306.10162]. SFQ control is sensitive to static frequency mismatch through
\[
1-F \approx \frac{(\delta\omega T_g)^2}{6},
\]
with a \(10^{-4}\) infidelity threshold reached near \(\delta\omega/2\pi\sim 300\) kHz for \(T_g\approx 12\) ns [1902.02911]. Qalibrate, in turn, emphasizes that pulse synthesis remains limited by the freshness and accuracy of device characterization even after the simulator itself has been accelerated [2511.13330]. Acceleration therefore shifts the dominant error budget toward calibration, transfer-function fidelity, and drift management.

Several common misconceptions are directly contradicted by published results. One is that higher levels are only harmful. In active transmon reset, the \(\ket f\) state is used deliberately as a staging level for fast population removal [1802.08980]. Another is that a smoother pulse must always mean a slower useful pulse. The finite-duration smooth-pulse analysis shows that one can retain a high filling ratio while achieving much weaker power broadening and far smaller sidebands than a rectangular pulse [2404.12236]. A third is that pulse-level optimization must reproduce a canonical target state to be meaningful. Resource-oriented transmon protocols demonstrate that direct optimization of negativity, three-tangle, or W-type concurrence can be both shorter and better aligned with the actual application objective than exact-state synthesis [2602.20275].

Open directions are also sharply delimited by the existing record. Several works connect shorter duration to reduced decoherence exposure but stop short of a full open-system algorithmic benchmark, as in the silicon chemistry study that argues for improved noise resilience primarily through runtime reduction rather than through Lindblad-level VQE performance [2606.17357]. Other studies remain numerical or backend-simulated rather than fully experimental, or focus on nearest-neighbor and local interactions while leaving nonlocal routing overhead to standard constructions. The field therefore appears to be moving toward a synthesis of three ingredients: hardware-tailored pulse design, online or fast offline calibration, and objective functions that reflect the actual resource being prepared rather than an unnecessarily rigid gate transcript.

Source: https://www.emergentmind.com/topics/pulse-accelerated-qubit-excitations