In the standard model of quantum computing, we treat qubits as abstract bits and operations as discrete logic gates like CNOT or Hadamard. However, physical hardware—whether superconducting circuits or trapped ions—does not "see" gates. It sees microwave pulses and laser beams.
The traditional "gate-level" abstraction creates a translation tax: a single logical operation is decomposed into a fixed sequence of hardware-native pulses. In the Noisy Intermediate-Scale Quantum (NISQ) era, where qubits lose their state (decohere) in microseconds, this tax is often too high. A growing body of research suggests that by bypassing gates and optimizing the raw physical pulses directly, we can execute algorithms significantly faster and with much higher fidelity.
The Race Against Decoherence
The fundamental intuition behind pulse-level control is simple: decoherence is a function of time. Every nanosecond a qubit spends idling or undergoing a slow operation is a nanosecond where environmental noise can corrupt the quantum state.
Charles D. Hill demonstrated that gate error rates are minimized by increasing the speed of operations relative to the environment's decoherence time. Similarly, Shang‐Yu Huang showed that sub-nanosecond gate operations can achieve fidelities high enough for fault-tolerant computing by simply "outrunning" the noise.
When we use standard gate abstractions, we often use more time than necessary. For example, a complex rotation might be compiled into three standard gates, each consisting of multiple pulses. Pulse-level optimization, such as the NAPA framework developed by Zhiding Liang and Jinglei Cheng, collapses these sequences into a single, continuous control signal. This approach has been shown to reduce circuit latency by an average of 86%, effectively shielding the quantum information by completing the computation before the environment can intervene.
Expanding the Parameter Space
Beyond speed, pulse-level control offers a much richer "language" for quantum programming. In a gate-based variational circuit, the only tunable parameters are the rotation angles. In a pulse-based circuit, researchers can tune the amplitude, frequency, phase, and duration of the microwave signals.
This shift, known as Variational Quantum Pulse (VQP) learning, allows algorithms to access a higher-dimensional parameter space. Zhiding Liang and Hanrui Wang have shown that this direct optimization allows the system to find "shortcuts" to the desired quantum state that are impossible to express using discrete gates. For instance, the SpacePulse framework combines classical subspace partitioning with raw pulse control to reduce circuit durations by over 80% while maintaining chemical accuracy in molecular simulations.
Hardware-Awareness and In-Situ Calibration
Standard gates are "one-size-fits-all" instructions, but quantum hardware is notoriously idiosyncratic. A CNOT gate on an IBM processor might perform differently on Monday than it does on Tuesday due to fluctuations in cryogenic temperatures or magnetic fields.
Pulse-level control allows for hardware-aware optimization. The QuPAD framework, introduced by Zhirui Hu and Weiwen Jiang, replaces fixed-duration gates with parameterized cross-resonance pulses. By "stretching" or "squishing" these pulses in real-time based on the current noise profile of the machine, the system can dodge daily instabilities. This in-situ calibration ensures that the pulses are always optimized for the specific, momentary state of the hardware, a feat that is difficult to achieve when restricted to pre-calibrated gate sets.
Efficient Paths on the Bloch Sphere
The geometric interpretation of quantum states provides another perspective on why this works. A quantum operation is a path on the surface of a Bloch sphere. Standard gates often take "scenic routes"—long, symmetric paths required to maintain the gate abstraction.
Researchers like Xin Yang have experimentally demonstrated "short-path" non-adiabatic geometric gates. By engineering a trajectory that closes the loop on the Bloch sphere more efficiently, they cut the evolution path length in half. This "one-step" implementation, as explored by M. -R. Yun, allows for complex operations like the fSim gate to be executed in parallel rather than sequence, cutting operation times to as low as 12 nanoseconds.
Bridging the Gap
While the benefits are clear, the challenge lies in the complexity of the control. Compiling high-level code directly to pulses is computationally intensive. New tools like QASMTrans by Aaron C. Hoyt and Ang Li are bridging this gap, providing end-to-end compilation that generates pulse schedules directly from high-level QASM code.
By treating the quantum computer as an analog device to be steered rather than a digital processor to be switched, this research direction is turning the "noise problem" into a control problem—one that can be solved with better pulse engineering rather than just better hardware.