Using a combination of tweaked algorithms, improved control systems and a new quantum service called Qiskit Runtime, IBM researchers have managed to resolve a quantum problem 120 times faster than the previous time they gave it a go.
Back in 2017, Big Blue announced that, equipped with a seven-qubit quantum processor, its researchers had successfully simulated the behavior of a small molecule called lithium hydride (LiH). At the time, the operation took 45 days. Now, four years later, the IBM Quantum team has announced that the same problem was solved in only nine hours.
The simulation was run entirely on the cloud, through IBM's Qiskit platform – an open-source library of tools that lets developers around the world create quantum programs and run them on prototype quantum devices that IBM makes available over the cloud.
SEE: Building the bionic brain (free PDF) (TechRepublic)
The speed-up that was observed was largely made possible thanks to a new quantum service, Qiskit Runtime, which was key to reducing latencies during the simulation.
IBM teased Qiskit Runtime earlier this year as part of the company's software roadmap for quantum computing, and at the time estimated that the new service would lead to a 100-time speed-up in workloads. With a reported 120-time speed-up, therefore, it seems that Big Blue has exceeded its own objectives.
Quantum programs: blending classical processors and quantum circuits
Classical computing remains a fundamental part of Qiskit, and of any quantum operation carried out over the cloud. A quantum program can effectively be broken down into two parts: using classical hardware, like a laptop, developers send queries over the cloud to the quantum hardware – in this case, to IBM's quantum computation center in Poughkeepsie, New York.
"The quantum method isn't just a quantum circuit that you execute," Blake Johnson, quantum platform lead at IBM Quantum, tells ZDNet. "There is an interaction between a classical computing resource that makes queries to the quantum hardware, then interprets those results to make new queries. That conversation is not a one-off thing – it's happening over and over again, and you need it to be fast."
With every request that is sent, a few tens of thousands of quantum circuits are executed. To simulate the small LiH molecule, for example, 4.1 billion circuits were executed, which corresponds to millions of queries going back and forth between the classical resource and the quantum one.
When this conversation happens in the cloud, over an internet connection, between a user's laptop and IBM's US-based quantum processors, latency can quickly become a significant hurdle.
Case in point: while solving a problem as complex as molecular simulation in 45 days is a start, it isn't enough to achieve the quantum strides that scientists are getting excited about.
"We currently have a system that isn't architected intrinsically around the fact that real workloads have these quantum-classical loops," says Johnson.
Based on this observation, IBM's quantum team set out to build Qiskit Runtime – a system that is built to natively accelerate the execution of a quantum program by removing some of the friction associated with the back-and-forth that is on-going between the quantum and the classical world.
Qiskit Runtime creates a containerized execution environment located beside the quantum hardware. Rather than sending many queries from their device to the cloud-based quantum computer, developers can therefore send entire programs to the Runtime environment, where the IBM hybrid cloud uploads and executes the work for them.
In other words, the loops that happen between the classical and the quantum environment are contained within Runtime – which itself is near to the quantum processor. This effectively slashes the latencies that emerge from communicating between a user's computer and the quantum processor.
"The classical part, which generates queries to the quantum hardware, can now be run in a container platform that is co-located with the quantum hardware," explains Johnson. "The program executing there can ask a question to the quantum hardware and get a response back very quickly. It is a very low-cost interaction, so those loops are now suddenly much faster."
The multiple faces of quantum computing
Improving the accuracy and scale of quantum calculations is no easy task.
Until now, explains Johnson, much of the research effort has focused on improving the quality of the quantum circuit. In practice, this has meant developing software that helps correct errors and add fault tolerance to the quantum hardware.
Qiskit Runtime, in this sense, marks a change in thinking: instead of working on the quality of quantum hardware, says Johnson, the system increases the overall program's capacity.
It remains true that the 120-times speed-up would not have been possible without additional tweaks to the hardware performance.
Algorithmic improvements, for example, reduced the number of iterations of the model that were required to receive a final answer by two to 10 times; while better processor performance meant that each iteration of the algorithm required less circuit runs.
At the same time, upgrades to the system software and control systems reduced the amount of time per circuit execution for each iteration.
"The quality is a critical ingredient that also makes the whole system run faster," says Johnson. "It is the harmonious improvement of quality and capacity working together that makes the system faster."
Now that the speed-up has been demonstrated in simulating the LiH molecule, Johnson is hoping to see developers use the improved technology to experiment with quantum applications in a variety of different fields beyond chemistry.
In another demonstration, for example, IBM's quantum team used Qiskit Runtime to run a machine-learning program for a classification task. The new system was able to execute the workload and find the optimal model to label a set of data in a timescale that Johnson described as "meaningful".
Qiskit Runtime will initially be released in beta, for a select number of users from IBM's Q Network, and will come with a fixed set-up of programs that are configurable. IBM expects that the system will be available to every user of the company's quantum services in the third quarter of 2021.
Combined with the 127-qubit quantum processor, called the IBM Quantum Eagle, which is slated for later this year, Big Blue hopes that the speed-up enabled by Runtime will mean that a lot of tasks that were once thought impractical on quantum computers will now be achievable.
The system certainly sets IBM on track to meet the objectives laid out in the company's quantum software roadmap, which projects that there will be frictionless quantum computing in a number of applications by 2025.