The Kookaburra is an important bird in some Australian Aboriginal societies because it was he who woke people up to watch the first sunrise. It is in a nod to this legend that IBM named the quantum processor it should unveil in 2025. A chip of more than 4000 qbits which should “awaken” men to the advent of the first quantum supercomputer.
Read also: IBM keeps its promise and announces Eagle, its 127-qubit processor (2021)
IBM has just updated its quantum roadmap – which is by far the most accurate on the market. The American, who has already won the first quantum computer sales contract in 2021 for the benefit of the Fraunhoffer Institute (Germany) is advancing his pawns at a hellish pace.
After unveiling Eagle last year (127 qbits), the IT giant is not stopping. This year, it will be the Osprey processor (433 qbits). Then Condor in 2023 (1121 qbits), Flamingo in 2024 (1386 qbits) to arrive at Kookaburra in 2025 – with 4158 qbits.
What is interesting in this progression is the detail of the increase in power. IBM is not going to design better chips from scratch every time. No, Flamingo like Kookaburra are quantum “systems” made up of three chips communicating with each other, a bit like our multicore chips (but in quantum).
The 1,386 qbit “Flamingo” quantum system will thus be made up of three Flamingo chips of 462 qbits each capable of exchanging via an integrated communication link.
Same approach for the “Kookaburra” system which will actually be made up of three 1,386 qbit Kookaburra chips.
Clearly, in its roadmap, IBM considers Osprey as the limit of monolithic chip design before switching to a chip aggregation system. Yes, IBM has really found a way to parallelize the power of its quantum chips…
A revolution in supercomputers?
With this parallelization of quantum chips – which requires communication between the chips in real time, the design of multi-chip systems and the quantum parallelization necessary for calculations, which is nothing! – IBM foresee not only an explosion of theoretical power… but above all its exploitation.
Read also: How IBM Should Lead the Revolutionary Quantum Advantage Pathway as Soon as 2023 (2022)
And quantum computing being what it is – with its strengths and weaknesses – we cannot do without CPUs and GPUs. IBM therefore plans, from 2025, to combine all these known (and mastered!) resources with its QPUs (quantum processing units). The object? Produce a new generation of hybrid, quantum supercomputers. Machines whose arrival should, according to IBM, have the same magnitude of impact as when we have “replaced paper maps with GPS satellites”.
Without software, hardware is nothing
The interconnection and parallelization of quantum chips – and it already sounds incredibly difficult – but it is only one of the building blocks of the quantum revolution (and of IBM’s roadmap). The other important part is to be found on the software side. Because for these machines to become popular, you have to be able to use them. IBM’s ambition is to make their programming transparent to non-physicists – in other words, “normal” developers.
This is why IBM has been developing software tools in parallel for years, in particular its Qiskit environment. A free program, which allows you to develop code in a classic way and to see it “translated” automatically into quantum language.
The tool being quite mature, IBM will add multitasking to it from 2023 (the threads) in order to stick to the parallelization of hardware development (of several chips).
Read also: These researchers have designed an incredibly reliable quantum chip that is cheaper to manufacture (2022)
Then comes the elimination of errors, which are inevitable when a system works in the form of probabilities like quantum chips. To arrive, beyond 2026, at quantum systems of more than 10,000 qbits. They will then not only be powerful enough to make a theoretical difference in certain applications, but will above all benefit from libraries and more mature software tools, particularly in the fields of machine learning, natural sciences or optimization problems.
IBM in front of everyone?
Once again, no member of the editorial staff of 01net.com is a physicist specializing in quantum particles and it is very difficult to judge the relevance of IBM’s model (qbits with superconductors) on the sole basis of its publications. But we can discern, in the substance and the form of his communication, that the American has strong backbones.
Besides the fact that it is the only player to publish such a roadmap, it has so far always delivered its chips and software components on time. No one else in the industry seems to have found the martingale of scaling up through parallelization or has such a clear commitment to delivering systems over 10,000 qbits by 2026.
Also see video:
Also see video:
If IBM actually provides its systems with the metronomic regularity to which it has accustomed us in recent years, the American could reiterate its past domination of the world of classical computing in that of quantum…