Control hardware is the classical machinery between an optimized pulse and the qubit: waveform generators and synthesizers, cables, filters and amplifiers, digitizers, FPGAs that make real-time decisions, and the software stack that schedules and calibrates everything [1]. Its properties — bandwidth, resolution, latency, noise, distortion, channel count — are constraints that every optimal-control problem inherits. As processors grow, the electronics become a scaling problem in their own right.

From waveform to qubit

  • Generating pulses. Pulses are usually synthesized at baseband by arbitrary waveform generators and mixed up to the qubit frequency. RF system-on-chip devices can now synthesize carriers of several GHz directly. One open design, QICK, reached 99.93% average gate fidelity on a transmon [2].
  • Bandwidth and resolution. Sample rates and filters limit how fast pulses can change, and the resolution of the converters limits how precisely they are set. These limits enter optimal control as constraints or through a parametrization in smooth functions; see Pulse parametrization & constraints. Hardware nonlinearities can be included in the optimization itself [3].
  • Distortion. Cables and filters distort pulses before they reach the device. The qubit can measure the distortion, of microwave pulses [4] or of flux pulses [5], and the measured response can be pre-compensated. The qubit can also act as a vector network analyser of its own control line [6]. See System identification for control.

Real-time control and feedback

Error correction, feedback and adaptive protocols need decisions within a fraction of the coherence time. Dedicated hardware and gateware feed qubit measurement results back or forward in real time [7]. FPGA-based signal processing demodulates readout, assigns the qubit state and triggers feedback with a latency of about 110 ns [8]. Below-threshold surface-code memories ran with a real-time decoder [9], and neural-network agents have acted on qubits in real time from FPGAs [10]. Modular, open FPGA systems such as QubiC give full access to the control stack [11]. See Measurement-based feedback and Control for quantum error correction.

Scaling the interface

Every qubit in a dilution refrigerator needs control and readout lines from room temperature. With millions of qubits the wiring cannot scale, so control electronics are moving closer to the qubits.

  • Cryogenic CMOS. Cryo-CMOS circuits are designed for operation at a few kelvin [12]. A 28-nm cryogenic controller for transmons dissipates less than 2 mW at 3 K [13]. Cryogenic CMOS has controlled silicon spin qubits [14].
  • Superconducting logic. Single-flux-quantum circuits could drive qubits directly with bitstreams derived from optimal control, measured by a Josephson photon counter at the millikelvin stage [15].

Such controllers offer fewer degrees of freedom than room-temperature instruments. The 28-nm controller, for example, implements a 16-word instruction set of XY gates [13], which restricts the pulses that optimal control can use.

Software stacks

Above the electronics sits software that compiles circuits into pulse schedules, calibrates them and analyses results. It includes pulse-level languages [16], calibration frameworks [17] and tools that combine control, characterization and optimization [18]. See Pulse-level compilation and Closed-loop & in-situ optimization.

Limitations

  • Pulses optimized without the true transfer function of the signal chain fail on hardware, and that transfer function drifts with temperature and time.
  • Latency adds up from readout, digitization, decision and pulse generation, and limits feedback speed whatever the qubit.
  • Low-power cryogenic electronics trade flexibility and precision for scalability, and their noise can couple into the qubits.