A cross-industry, objective basis is comparing investments
It would nice to have a fact-based answer to such questions as “when will the quantum industry be profitable?” The DARPA Quantum Benchmarking Initiative (QBI) https://www.darpa.mil/research/programs/quantum-benchmarking-initiative might provide the answer if investors and analysts ask the right question. Clicking on the link leads to pages that look like a dry engineering program. I explain further below, but in summary, after years and a lot of caveats, DARPA might say “we will give you money you to build it.”
The QBI program first solicits concepts for a hypothetical quantum computer that somebody might buy based on its computational output – which DARPA calls “utility scale.” If DARPA accepts a concept, the performer’s next step is to create research and development plan, and then verification and validation. For a moderately attentive outsider (like the author), the proposers’ day video conference was the first time DARPA mentioned that they might pay performers, at least in part, to build the system. After issuing several caveats, the program manager said the last phase did not have a specific dollar limit. DARPA also states “QBI is not a competition between performers; DARPA is interested in pursuing all viable approaches for which there is available funding.”
Let us consider how QBI could help investors, starting next quarter. Companies intending to sell quantum computers ought to have a plan to develop their product, and investors can reasonably expect companies to take non-dilutive capital from the government where available.
For example, Subodh Kulkarni, Rigetti President CEO, said on their Q3 earnings call, “there is a DARPA program for what they call benchmarking initiative that’s a sizable program on the order of $300 million [f]or 7 years. So we certainly hope we can get some of that” https://seekingalpha.com/article/4736486-rigetti-computing-inc-rgti-q3-2024-earnings-call-transcript.
I suggest investors and analysts track “QBI status” on an ongoing basis, such as at earnings calls. Initial QBI concepts are in DARPA’s hands and under evaluation, so investors can ask whether a company submitted and then follow up to categorize companies as follows:
- Company submitted and is awaiting a response.
- Company did not submit because their product is not a utility scale quantum computer.
- Company did not submit because they object to the terms of DARPA funding and have enough money anyway.
- Company did not submit because they not have the necessary technology.
Over time, “QBI status” will evolve to include acceptance/rejection of their proposal and progress toward fielding a utility scale quantum computer against competitors.
In conclusion, the DARPA QBI program will perform industry wide assessments may be useful to investors, even though that was not DARPA’s original purpose. There are other industry wide assessments, such as the Quantum Economic Development Consortium’s (QED-C’s) quantum benchmarks, which could be a topic for a future document.
Speed, reliability, and the number of qubits
The Rigetti CEO stated that their transmon “gate speeds consistently achieve an active duration of 60 to 80 nanoseconds, which is four orders of magnitude faster than other modalities such as ion traps and pure atoms.” https://seekingalpha.com/article/4736486-rigetti-computing-inc-rgti-q3-2024-earnings-call-transcript. Transmons’ higher speed should give a +1 score to transmon vendors, but there is a reverse effect that should give ion trap vendors a +1 score.
Fact check of the statement “gate speeds … 60 to 80 nanoseconds … is four orders of magnitude faster.” According to Google’s AI (need to check) Rigetti specs: 1Q= 50 ns, 2Q=213 ns, T1=20.3 μs, T2=10.9 μs, 1Q/T2=218. IonQ specs: 1Q=135 μs, 2Q=600μs, T1=10-100 μs, T2=1 s, 1Q/T2=7400. The 104 ratio compares 1 qubit ops with 2 qubit ops. Also, IonQ’s decoherence time is proportionately much longer.
The figure below (from https://research.google/blog/making-quantum-error-correction-work/ — don’t forget to deal with copyright) is from Google’s recent Willow announcement and shows three levels of quantum error correction that Google apparently believes to be relevant to transmon quantum computers. Less reliable physical qubits will require more powerful quantum error correction, corresponding to rightward movement on the diagram. However, rightward movement requires more physical qubits per error-corrected qubit and reduces the number of error-corrected qubits available to software. All other factors being equal, more reliable physical qubits might not require as much rightward movement and allow a user to solve a larger problem.
[diagram here]
A quantum computer will be far more expensive than classical computer with the same number of bits/qubits or gates, so quantum computers are only of interest if unique quantum properties allow faster solution to a specific problem. Quantum speedup is the term for the relative advantage, and quantum speedup grows rapidly with problem size. So, the transmons’ speed advantage (which is real, even if it is not 104) just means quantum computer based on transmons will show an advantage with fewer qubits than an ion trap quantum computer. In this case, “qubits” means error-corrected qubits.
Ion traps have slower gates than transmons by about 104 but information decays at a slower rate by about 106. This means the less-powerful 5×5 quantum error correction code in the middle of the figure above might be sufficient for ion traps whereas transmons might require the 7×7 code. For a given number of physical qubits, this would make about twice as many (97/49) qubits available to ion trap software compared to transmon software.
The description above gives the basic idea of how to trade qubit speed against qubit reliability. In fact, IonQ is pursuing an approach based on smaller code words but not based on the square grids above.
Technical efforts in the community are investigating many alternative codes, control systems, and how to optimize transmons and ion trap qubits to be most effective when used in quantum error correcting codes.