CTRL didn’t give away the “secret sauce” of Fire Opal at IEEE Quantum Week, which was held September 18-23, 2022, in Colorado, USA, but they did give out free samples on toothpicks. Unfortunately, Gordon Ramsay doesn’t write for Quantum Zeitgeist, but what follows is a best attempt to describe the experience.
How Fire Opal Works
Like a multiple-course meal, Fire Opal works in multiple stages. It starts by reducing circuit depth as much as possible, as well as minimizing the use of CNOT gates. It then optimizes the mapping of virtual qubits to physical qubits, taking into consideration the error rates of the qubits and their connections. Two more stages before measurement takes places are circuit crosstalk elimination and optimized gate replacement. And, finally, there is a measurement error mitigation stage at the end.
That’s a bit high level, so for more detail check out https://arxiv.org/abs/2209.06864. You can also request a demo from the Q-CTRL Fire Opal team.
One important note is that the Fire Opal team strives to give fair comparisons. When you read about purported performance improvements, those numbers are comparing Fire Opal’s results to that of fully decomposed circuits optimized with Quantinuum’s TKET and/or IBM’s Qiskit at its highest optimization setting. And even with Fire Opal’s advantages, it is not yet at its peak; for one example, Dr. Hush noted that error detection and correction will become possible once mid-circuit measurements are available on hardware. There was also mention of Fire Opal boosting quantum volume, the details of which are also included in the aforementioned paper.
Three Ways to Use Boulder Opal
Fire Opal’s gate optimization is built on top of Q-CTRL’s Boulder Opal product, so Boulder Opal was part of the same presentation. The discussion included the pros and cons of Boulder Opal’s three control options: Optimal Control, which underperforms on real devices but is the fastest and easiest option, Robust Control, which considers imperfect knowledge of the model and produces better results for real world problems, and Learning Control, which sounds like classical machine learning applied to control systems.
Fire Opal includes components that Boulder Opal doesn’t have, such as transpilation, and is intended for developers who want to improve quantum algorithm performance on commercial devices. Boulder Opal is intended for the researchers and engineers who are building such devices. You can request a demo of Boulder Opal, too.
A Holistic Approach
Between the talk of circuits and measurements on the one hand, and the talk of hardware on the other hand, it becomes clear that the “secret sauce” of Fire Opal is employing every possible strategy Q-CTRL possibly can. The end result is that you can give them noisy circuits that generate uniform distributions, and the Fire Opal team will return to you useful results. They’re not quite at the level of noise-free quantum computing simulators, but you can see the same result(s) stand out that a simulator will give you. It’s not quite fault tolerance and the post-NISQ era, but Fire Opal is allowing practical quantum computing today.