Pulse-level benchmarks to independently evaluate the performance of quantum control technologies in terms of quantum-classical integration and feedback latency
As quantum computers scale up in size and utility, the performance of control systems becomes a critical factor in making the best of any quantum processing unit. Existing gate-level benchmarks evaluate the overall performance of quantum computers but do not allow for individual assessment of the controller quality, which instead operates at the pulse level. At Quantum Machines, we proposed pulse-level metrics based on the operational feedback latency of the controller in its quantum-classical computation. These metrics aim at evaluating the quantum-classical integration and feedback capabilities of any quantum control technology, to push towards useful quantum computers.
Feedback Timescales
Feedback is the heart of our benchmarking proposition, as it is critical for a wide range of quantum algorithms. We categorize feedback operations into three types: conditional operation, full control flow, and parametric changes. All of these are considered real-time feedback operations as they use readouts.
Pulse-level benchmarking quantifies the performances of these feedback types across three timescales: quantum real-time (QRT), system real-time (SRT), and near quantum real-time (NRT). The controller latency is compared to the coherence of the quantum system for QRT, to the drift of the setup and device for SRT, and to the duration of the overall quantum circuit, including averaging for NRT.
Feedback latency is defined as the time from the acquisition of the last sample needed for computation to the first sample of the subsequent pulse sequence. These operations are dependent on the computed measurement results; thus, processing time builds up latency, and different computations will result in different latencies. If the computation and response are performed within the coherence time of the quantum processing unit (QPU), then they are said to happen in quantum real-time (QRT). Requiring QRT processing to be time-deterministic (takes a known amount of time) allows the controller to completely determine the timing of procedures, e.g., tracking the evolution of the system and phase changes.
At Quantum Machines, we developed one QRT benchmark for each feedback type, and we here show one example.
QRT benchmark example: Active reset
Active reset serves as an excellent example to illustrate the role of latency in quantum sequences. The standard reset based on thermalization requires waiting for a duration of about ten times the relaxation time of the qubit. This wait can extend to, for instance, 1 ms, to reasonably ensure the qubit’s successful initialization.
Active reset is a protocol of measurement and a conditional reset pulse. The processing required to evaluate the measurement result and produce the conditional drive is what composes the feedback latency. For the active reset to be useful, this time has to be much shorter than the coherence time of the qubit. For example, our OPX controller allows active reset with latencies of the order of 200 ns, speeding up sequences and enabling multi-qubit initialization with high fidelities.
# THERMALIZATION RESET
measure('readout', 'resonator', ..., I, Q)
wait(10 * relax_time) # latency e.g. ~ 1 ms
play('pi', 'qubit')
# ULTRA-FAST ACTIVE RESET
measure('readout', 'resonator', ..., I, Q)
# latency < 250 ns
play('pi', 'qubit', condition = I > I_threshold)
System Real-Time (SRT) Benchmarks
Every QPU has drifting parameters (e.g., qubit frequency, laser intensity, DC voltage bias, etc.). One effective method for compensating and mitigating their detrimental effects is to monitor these changes in real-time and update these parameters during quantum circuits. Such calculations and parametric updates need to occur at a rate significantly faster than the drift time, and as a result, they are described as taking place in system real-time (SRT).
In our ongoing work on benchmarking quantum controllers we developed a system real-time measurable metric that incorporates multi-shot and potentially multi-qubit subprograms that can run together with an application subprogram, ensuring high operational fidelities.
SRT benchmark example: Frequency tracking
A remarkable application for SRT parameters update is frequency tracking, a calibration routine that allows to track and correct the drive frequency used for qubit control. This is done thanks to very fast computation performed on measurement results (e.g., frequency estimation done after a few-points Ramsey). This routine, embedded between shots of an application circuit, allows to run always with spot on drive frequency. In SRT calibrations, lower controller latency directly allows for more advanced corrections and higher bandwidth.
Some classical calculations may be employed in quantum algorithms, such as variational algorithms and other hybrid quantum-classical applications, without directly impacting circuit accuracy. Instead, they influence circuit runtime, which is evaluated in near real-time (NRT) benchmarks. Current NRT benchmarks, like IBM’s CLOPS, often do not differentiate the controller’s role. Presently, NRT quantum computing is constrained by the sluggishness of quantum control – classical communication, which takes milliseconds to seconds, significantly contributing to latency. At Quantum Machines, we are developing NRT benchmarks for individual assessment of quantum controller quality. Simultaneously, in collaboration with NVIDIA, we are working on hardware and software technologies to reduce quantum control and GPU communication latency to as low as 50 milliseconds.”
We use cookies to help personalize content, tailor and measure ads, and provide a safer experience. By continuing to use this website you consent to the use of the cookies in accordance with our Cookie Policy.
