IBM Speeds Quantum Experiments as It Plans a Larger Nighthawk System

International Business Machines (IBM) says researchers are completing some quantum experiments many times faster on its Nighthawk r2 processor, giving them more opportunities to test calculations that can require millions of repeated measurements. The November 11 announcement puts speed alongside size as a practical measure of progress toward useful quantum computing. [1]

The company reported a 15-fold speedup in E.ON experiments applying quantum machine learning to power-grid optimization and a 12-fold speedup in a magnetic-materials simulation developed with Oak Ridge National Laboratory. Both comparisons use IBM’s earlier Heron quantum hardware as the baseline, measuring gains within the company’s quantum processor family. For researchers, faster sampling can shorten waits and allow more measurements for accuracy. The industrial benefit depends on the complete calculation, including the conventional computing that prepares and interprets it. [1, 2]

IBM also outlined a larger machine: two planned Nighthawk r3 processors, connected into a 1,040-qubit system in 2027. The current r2 results concern work already run on a 120-programmable-qubit processor. The larger connected system remains a roadmap commitment. [1, 2]

A quantum circuit is a sequence of operations on qubits, the physical units that carry quantum information. Each measurement produces a sample, and researchers repeat circuits to estimate quantities from many samples. Nighthawk r2 has a dedicated reset mechanism for each programmable qubit: a controllable connection drains energy into a cold environment, returning the qubit to its starting state. IBM says the idle interval can fall to one microsecond while neighboring qubits retain their quality. Its peak rate is up to 100,000 samples per second, about 25 times Heron’s roughly 4,000. Those rates, described in an August 31 technical post, are hardware measures; individual workloads also depend on circuit length, accuracy requirements and processing time. [3]

IBM says customers and partners have executed more than 34 billion circuits on r2. An Algorithmiq experiment simulating a heterogeneous quantum material runs eight times faster than its earlier Heron version, according to the company. “Useful quantum computing hinges on whether results can be delivered with accuracy and speed, both are fundamental,” said Jay Gambetta, IBM Research director. [1, 2]

IBM reports accurate estimates from circuits containing more than 7,500 operations using probabilistic error amplification. The method deliberately adds controlled amounts of noise, runs the resulting circuits and estimates the result at zero noise. It requires a model of the device’s errors and extra measurements. Faster sampling helps pay for that work. This error-mitigation approach estimates a less noisy answer from repeated runs; fault-tolerant computing protects quantum information through ongoing error correction. IBM targets its first Starling fault-tolerant system for 2029. [2, 4]

Nighthawk r3 is designed for 520 programmable qubits and circuits containing 10,000 operations. Connecting two processors would extend the architecture beyond a single chip. In earlier experiments linking Nighthawk processors, IBM reported average quantum-state transfer fidelity of 99 percent, with a best result of 99.3 percent. Fidelity describes how closely the transferred state matches the intended result. A useful linked machine also requires reliable operations across the connection and coordinated control of the whole system. [1, 5]

The 2027 system will need to sustain accuracy and throughput across a larger assembled machine. For r2, researchers can now investigate how faster sampling changes the time and total computing cost of a scientifically useful result. Those full-workload measurements will show how much of the processor’s speed reaches practical applications. [1–5]

Sources

[1] IBM announcement of Nighthawk results and plans

[2] IBM Quantum Developer Conference technical account

[3] IBM technical explanation of Nighthawk r2

[4] IBM tutorial on probabilistic error amplification

[5] IBM research on processor connections