When it comes to large companies with significant investments in quantum computing for a broad user base, no one has the momentum IBM currently does.
It’s not just about qubit counts or reliability or programmability; they’re setting the stage for real applications without bothering with pushing an on-prem hardware business. Everything IBM has quantum-wise is happening in its own cloud. If their current roadmap is any estimate, those capabilities will get much richer, it’s just not going to be a quick process. Then again, realistic estimates of quantum feasibility for a wide enough set of applications wasn’t expected for at least a decade, so perhaps they are farther ahead of the game than anyone could have hoped.
Todays’ roadmap, shared by Jay Gambetta, VP of Quantum at IBM, spares us the detailed qubit improvements or in-depth connectivity or coherency details and instead provided a glimpse at a select few of the enabling technologies we’ll see between now and 2025. If all of these milestones share anything in common, it’s about making quantum computing more accessible to ordinary developers, beginning in particular with HPC—the real showcase area for some of the world’s most challenging computational problems.
The roadmap begins well after IBM’s quantum efforts were underway but it’s an important point along the path because it’s when their quantum circuits moved to their cloud. In 2020 the company focused on demonstrating and prototyping applications. In 2021 we’ll see those applications running 100x faster, Gambetta says. Next year the number of algorithms that can be explored with one of those defining enabling technologies, dynamic circuits, will also grow.
In 2023, IBM expects something truly interesting to happen. Gambetta says that ordinary developers, using languages of their choice, can allow developers to use IBM’s Circuit API (something that’s been around since 2016) to let those developers create an object that defines quantum instructions and push those to their quantum machine. For now, this works for simple circuits, but Gambetta say to achieve this goal they need to be able to run more circuits much faster. That leads us into another enabling technology for this roadmap, dynamic circuits. We’ll get that in a moment.
By the time all of this ends IBM will be able to offer a thousand qubits for users over its cloud, refining along the way not just coherency and connectivity, but how users can tap into the resource most efficiently with maximum use of those devices.
So let’s look at a few of the enabling technologies for this roadmap in a bit more detail.
Later this year we’ll see IBM’s Qiskit Runtime. This will support iterative, efficient running of different circuits and will update future circuits based on previous measurements. Later in 2021 they’ll demo this with the goal of showing major performance increases. Gambetta says that with the current implement all loops run on a user’s computer and go back and forth on the cloud. With the Qiskit Runtime, loops will be executed physically close to the quantum computer, providing a speed-up as large as 100X.
The other enabling advance is in dynamic circuits, or smart circuits, that can provide branching within the circuit. They can change their future state, in other words. Coupled with that is the growth in classical control hardware to make all of this possible. This includes changes with how resets are handled for more efficient zero state preparations in addition to using iterative phase estimation (a core sub-algorithm for quantum applications that allows for quick decisions about the minimum number of resources to use). Over the next couple of years, IBM will be able to predict certain circuits will be used more frequently and can create a library of optimized pre-built ones.
By the time IBM is hitting the thousand qubit mark, the need for massive error correction strategies will also be in place. The challenge of doing this for the execution of millions of circuits is not trivial. “To do this we need advanced control electronics built from the ground up to solve for more complex computation,” Gambetta explains.
One of the more interesting branches of their roadmap is focused on HPC. Gambetta says it will make sense to have pre-built quantum runtimes for classical and quantum resources to support work in optimization, science, finance and ML. He says too that HPC and the natural sciences provide a chance to work with some of the world’s most important problems that can actually be amenable to quantum systems.
“Having HPC tightly bound on quantum resources will allow us to level up to solve more complex computations,” Gambetta says.