Honeywell Quantum Solutions announced the release of its next generation quantum computer. The System Model H1 uses trapped-ion technology built on Honeywell’s advanced QCCD architecture. The new quantum computer will be available under a newly created subscription service. The system can be directly accessed via a cloud API, as well as through Microsoft Azure Quantum.
As part of its announcement, Honeywell also released its long-term technology roadmap that leads to a fault-tolerant quantum computer sometime in the 2030s.
The System Model H1
Like its predecessor, the Model H1 uses ytterbium ions for computations and barium ions for cooling. In early 2020, Honeywell released a paper with detailed information on its quantum charged-coupled device (QCCD) architecture. It is an advanced trapped-ion architecture that allows for arbitrary movement of ions and parallel gate operations across multiple zones. QCCD has been the subject of many research papers. When I spoke with Tony Uttley, president of Honeywell Quantum Solutions, in March 2020, he indicated that QCCD was the architecture that would support the future generations of Honeywell quantum processors.
It has high fidelity operations, low crosstalk, and features mid-circuit measurements and conditional feed-back. The QCCD architecture can also support the operations necessary for implementing quantum error correction. Error correction is critical to the future of all quantum computing systems. It offers a viable path towards building large quantum computers.
Presently the Model H1 has ten fully connected qubits, up from 6 qubits in the Model H0. It has a two-qubit gate fidelity of at least 99.5%. Two qubit gate error rates are very important. It’s what happens when a two qubit gate is applied to a quantum circuit and if it produces the right or wrong result. In other words, the Model H1 two qubit gates produces the correct results an average of 99 ½ times out of every 100 operations.
According to Honeywell’s roadmap, the Model H1 can accommodate up to 40 ion qubits. More qubits mean more processing power. Because of the architectural flexibility and potential qubit capacity of its current QCCD configuration, Honeywell will be able to upgrade the H1 generation with up to 40 qubits, higher fidelities and unique feature modifications.
Quantum volume of System Model H1
Quantum volume is a hardware-agnostic performance measurement for gate-based quantum computers. IBM developed it in 2017. QV considers such things as the number of qubits, connectivity of qubits, gate fidelity, cross talk, circuit compiler efficiency, and more. In other words, it takes more than just adding qubits to increase the quantum volume: the larger the QV, the more powerful the machine.
With 10 qubits, Model H1 scientists measured its quantum volume to be 128. That is the highest measured quantum volume in the industry. The measurement was made by running appropriate algorithms on quantum volume circuits. The results of those measurements are on Honeywell’s website. It’s the accepted norm for quantum scientists to publish scientific proof of new quantum hardware, software, theories, and new records such as quantum volume.
IonQ recently announced that it had managed to successfully load 32 ions in its quantum computer. That is a commendable achievement that advances the entirety of quantum computing. However, it also announced it had calculated, rather than measured, a staggering new quantum volume of 4 million.
Calculating, instead of properly measuring quantum volume, can produce vastly different results. For instance, if the quantum volume of Honeywell’s Model H1 was calculated by the same method used by IonQ there would be a big difference. Instead a measured QV of 128, it would have a QV of 1024. Let’s go a little further. Assume Honeywell fills the Model H1 to 4o ion qubits. Of course, we have no way to measure that right now. But we can calculate the QV as did IonQ. With 40 qubits, its calculated quantum volume would be about 17,000,000,000,000.
With the exception of the Honeywell Model H1, none of the generations show the number of qubits or timeframes for deployment. I expect a large increase in qubits with the announcement of each new generation. According to Uttley, Model H2 is currently being worked on in a vacuum chamber that sits next to the newly announced Model H1. For the remaining generations, it will take a number of years from design to deployment of a beta for a limited number of customers. Each generation is based on some QCCD configuration.
The roadmap moves from a trap with linear topology to increasingly complex grid topologies in H3, H4, and H5 grids. A linear topology trap is relatively easy to build and its ions are easy to manipulate. Shuttling and performing operations on ions in the grid topology is more complex. Shuttling operations require removal of ions from a trap, shuttling them along paths, turning at junctions, and merging them.
I asked Tony Uttley if they plan to use optical interconnects in Model H5. He explained that it doesn’t appear optical coupling will be necessary. They are using ion trap tiling instead which is like quilting and faster than photonic coupling. By quilting traps together, it creates an integrated surface for shuttling.
With Model H1 complete and Model H2 in process, Uttley explained why he is confident the remainder of his roadmap is doable. “We have already demonstrated that we can turn corners in the Model H3 because we have built a precursor trap that does that. Why do we think we can build a Model H4 with integrated optics? Because we’ve already designed and built and tested our own photonic devices that allow us to get laser sources to where they need to be in an integrated circuit. Honeywell does what it says it’s going to do. And we only talk about things that we have demonstrated.”
Demand is high
Uttley explained how demand for service has changed. “At this stage of quantum computing, researchers don’t use just 15 to 30 minutes of quantum computing like they did a few years ago. If they have something meaningful, they need tens of hours to run algorithms. Hybrid algorithms like VQE or QAOA use intensive classical quantum code processing. Our Model H0 allows people to test our capabilities. Everyone who subscribes to Model H1 today were previous users of Model H0. Demand for Model H1 service is high.”
- There are many advantages to integrated photonics. Not only from a scaling standpoint, but also for performance improvements. One advantage is elimination of vibration. Honeywell has already built prototypes plus there are several recent integrations using all colors. I was surprised that Honeywell didn’t incorporate integrated photonics in Model H3 instead of waiting until H4 to begin using it. Honeywell will have it tested and ready to deploy early but doesn’t feel it is needed until H4 which also prepares it for H5.
- Model H1 indicates a maximum of 40 qubits. Other work suggests a higher capacity is possible, perhaps as many as 50 qubits or more.
- There is a big plus with subscription service. Honeywell is offering customers direct access to the researchers who develop and operate the systems.
- The availability of error correction within 5 to 10 years is critical to not only Honeywell’s long term hardware designs, but to all quantum computing companies as well.
- Mid-circuit measurement is a feature currently unique to Honeywell. Qubit measurement destroys a qubits quantum state. QCCD allows a qubit to be measured and then reset to its original state in the middle of a circuit instead of at the end. In addition to acting as a conditional if statement, it has been also demonstrated to reduce the number of required qubits. In one example it reduced the number of qubits from 6 to 2 to run Bernstein-Vazirani.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.