IBM Nighthawk r2’s 25× Speedup Comes From Resetting Qubits Faster

IBM’s Nighthawk r2 processor increases circuit throughput by attacking a practical bottleneck: waiting for qubits to return to their ground state before the next circuit run. Rather than simply adding more programmable qubits or accelerating every gate, IBM added independent high-speed reset capability to each qubit. The result, according to IBM, is more than 100,000 maximum circuits per second on ibm_phoenix, up to 25 times the throughput of its Heron fleet. 16

The key change: active, per-qubit reset

A quantum processor must start a new circuit in a known initial state, generally written as (|0\rangle). If a qubit is still excited after a prior run, the system has to wait for it to relax naturally or reset it through another method. That turnaround time can dominate workloads that execute many short circuits or many repeated shots.

Nighthawk r2 introduces an independent dissipative-reset element for each of its 120 programmable qubits. IBM describes the mechanism as actively cooling a qubit back to its ground state; reporting on the device describes a high-dynamic-range coupler that draws energy from the programmable qubit on demand. The reported effective relaxation time falls from roughly 200 microseconds to about 25 nanoseconds when reset is active. 36

The system-level consequence is a much shorter delay between circuit executions. IBM reduced the default repetition delay from 250 µs to 1 µs, allowing the next run to begin far sooner. 1

What the 25× figure means: It is a throughput measure for circuit execution, especially meaningful where reset and repetition delay are substantial parts of total runtime. It does not mean every quantum gate operates 25 times faster.

What is on the processor?

Nighthawk r2 is a 120-programmable-qubit processor built in a square-lattice layout, with each qubit connected to up to four neighbors. Its physical elements include:

• 120 programmable qubits that carry out quantum operations;
• 218 dedicated couplers that provide two-qubit connectivity; and
• 120 independent reset elements, one associated with each programmable qubit.

That is 458 physical quantum elements in total. The reset elements are supporting hardware, not additional computational qubits. 1218

This design reflects a trade-off that is increasingly important in quantum engineering: a processor can improve useful work not only by increasing qubit count, but also by improving connectivity, fidelity, control, initialization, and the rate at which the system can complete repeated workloads.

Why reset quality matters as much as reset speed

Fast reset would be less useful if it left qubits poorly initialized or disrupted nearby operations. IBM says Nighthawk r2’s active reset reduces initialization error by around 25× while maintaining Heron-class gate fidelity. It also says the reset mechanism is designed to operate without materially disturbing neighboring qubits. 6

That combination matters because initialization is part of the full computational error budget. A faster experiment that begins from a less reliable state is not necessarily a better experiment. IBM’s claim is that Nighthawk r2 improves turnaround while preserving the gate-quality baseline of Heron. 16

Why more circuits per second can be more useful than more qubits

Circuit throughput is most consequential for workloads that require repeated execution: sampling, estimating observables, calibration, variational methods, and other experiments that gather statistics over many shots. In those cases, long reset waits can leave the processor idle for a significant share of the experiment.

IBM reports that Nighthawk r2 has demonstrated accurate observable estimation on circuits containing more than 7,500 gates. That is a circuit-depth result, while the 100,000-plus figure is a throughput metric; together they are intended to show capability across both workload complexity and execution rate. 6

The distinction is important. A high maximum-circuits-per-second figure alone does not establish that a machine can reliably execute arbitrary long or complex algorithms. Likewise, a deep-circuit demonstration does not by itself describe how quickly a device can produce repeated results. Nighthawk r2 is designed to improve both dimensions without increasing the programmable-qubit count beyond 120. 617

Mid-circuit reset and quantum-error-correction research

The reset capability is also available for dynamic, mid-circuit workflows. In principle, that enables a qubit to be measured, reset, and reused during one larger computation rather than remaining unavailable after measurement.

This is especially relevant for quantum-error-correction experiments, where auxiliary qubits may need to be repeatedly measured and refreshed while carrying out checks on data qubits. Fast, independent reset can reduce the overhead of those cycles and makes the hardware more suitable for exploring such protocols. IBM identifies Nighthawk as a platform for scaling workloads ahead of large-scale fault-tolerant computing. 618

Where Nighthawk r2 fits in IBM’s roadmap

Nighthawk r2 is available on the IBM Quantum Platform as ibm_phoenix for eligible plans. IBM presents the processor as a step toward more useful work at a fixed 120-qubit capacity, emphasizing a combination of scale, quality, and speed rather than qubit count alone. 117

It is not, however, a fault-tolerant quantum computer. IBM’s planned Starling system is targeted for 2029 and is described as an error-corrected modular system intended to run 100 million gates on 200 logical qubits. That is a roadmap objective, not a capability delivered by Nighthawk r2 today. 1729

The bottom line

Nighthawk r2’s headline speedup comes from eliminating much of the dead time between circuit runs. By pairing every programmable qubit with independent dissipative-reset hardware, IBM says it can reset qubits faster and more completely, lower the default repetition delay to 1 µs, and reach more than 100,000 circuit executions per second. 16

For quantum users, the practical significance is not “25× faster quantum gates.” It is the potential to collect repeated circuit results much faster while retaining comparable gate fidelity—an improvement that is particularly relevant to shot-heavy workloads, dynamic circuits, and error-correction research. 16