IBM has updated its quantum computing roadmap to reflect how quantum processors, central processing units (CPUs) and graphics processing units (GPUs) will be woven into a compute fabric capable of solving computationally complex problems.
“We think we have found what it takes to scale quantum computers into what we’re calling quantum-centric supercomputers,” said Jay Gambetta, IBM fellow and vice-president of quantum computing at IBM.
Building on its existing roadmap, IBM has unveiled a 133-qubit Heron processor, slated for 2023, that has redesigned gates and tunable couplers to improve speed and reliability.
“We’re also preparing for the ability to control multiple Heron processors with the same control hardware, enabling quantum computing with classical communication between each processor,” said Gambetta. “Heron will allow for classical parallelisation between quantum chips.”
IBM has also been working on a chip-to-chip coupler, to run two-qubit gates between qubits on different chips. Gambetta said IBM planned to release a minimum viable product in 2024 demonstrating this technology – a 408-qubit processor, built on the Heron technology, called Crossbill, which is made from three chips joined by this modular coupler.
“Our goal is that users feel as if they’re using just one, bigger processor,” he added.
In 2024, Gambetta said IBM also plans to introduce longer-range quantum communication between chips and create clusters of quantum processors using a long-range coupler for connecting qubit chips through a cryogenic cable of around a metre long.
“We will be releasing a demonstration of this architecture by linking together at least three 462-qubit processors, each called Flamingo, into a 1386-qubit system. We expect that this long-range coupler will be slower and lower-fidelity than our on-chip gates, since it involves a physical cable. Our software needs to be aware of this architecture consideration for our users to best take advantage of this system,” said Gambetta.
Kookaburra will be the next quantum processor. Due in 2025, IBM said Kookaburra will be a 1386-qubit multichip processor with quantum communication link support for quantum parallelisation. As a demonstration, Gambetta said IBM plans to connect three Kookaburra chips into a 4158-qubit system connected by quantum communication.
IBM has also been working on the software side of quantum computing, in a bid to deliver quantum advantage sooner by taking a hybrid approach. In effect, a problem is broken down into a series of smaller quantum and classical programs. An orchestration layer is then used to stitch the data streams together into an overall workflow. IBM dubs the approach Quantum Serverless.
“Quantum Serverless centres around enabling flexible quantum-classical resource combinations without requiring developers to be hardware and infrastructure experts, allocating just those compute resources a developer needs when they need it,” said Gambetta. “In 2023, we plan to integrate Quantum Serverless into our core software stack to enable core functionality such as circuit knitting.”
Discussing the challenges of connecting quantum computer systems together, Katie Pizzolato, director of IBM quantum strategy and applications research, said: “The scalability challenge is that there is a limit on how many qubits you can put on a device and how you link devices together to enable a 4158-qubit system.”
She said that between 300 and 400 qubit systems can be linked together using the short coupling technology IBM is developing. The long coupling needs to be fast enough to ensure that the performance of applications is not severely restricted by the slower connectivity between clusters of 300 to 400 qubit systems.
“The idea is to put as much of the hardware as possible in the same fridge which can hold 1,000 qubits,” added Pizzolato.
IBM’s Quantum System Two, unveiled in November 2021, is the first example of how a system could be built to scale up using a modular design. Given the one-metre constraint on the connectivity between qubit systems, clusters of systems could be arranged in a cylindrical fashion, where each cylinder comprises a fridge with a number of interconnected 300-400-qubit systems.
“By 2025, we will have effectively removed the main boundaries in the way of scaling quantum processors up with modular quantum hardware and the accompanying control electronics and cryogenic infrastructure,” said Pizzolato. “Pushing modularity in both our software and our hardware will be key to achieving scale well ahead of our competitors.”
Just as when blade servers changed the construction, energy and cooling requirements of datacentres, IBM said it was already thinking about what a future hybrid datacentre for classical and quantum computing would look like.
“Our experience tells us that the requirements of a quantum datacentre are very similar to those of classical datacentres, with addressable solutions to accommodate cryogenic equipment,” said Pizzolato.
“Key aspects of datacentre design – such as electrical and cooling water requirements, footprint needs, and the standardisation of infrastructure and system elements – are an integral part of our thought process. We have been able to leverage our deep experience in system and datacentre design to move quickly to design our quantum centres.”