In a press release, D-Wave provided a preview of their next generation quantum annealing machine codenamed Pegasus. In the press release, D-Wave highlighted key advances in this new architecture including an increase of qubit connectivity from 6 to 15, lower noise and an increased qubit count to over 5000 qubits.
D-Wave had previously disclosed this architecture at the technical Adiabatic Quantum Computing (AQC) conference in June 2018 and we had previously reported on it in our article describing the event. Some additional details about this next generation machine described at that event, but not mentioned in the press release include:
There are apparently three different versions of this architecture with designators P6 (680 qubits), P12 (3080 qubits), and P16 (5640 qubits). Although we don’t know the specific introduction timeline for these machines when asked, D-Wave responded “We’ll be releasing core elements of the next generation system over this year and into mid-2020”.
The qubit connectivity increase from 6 to 15 is important. For comparison, the IBM 20 qubit design has a maximum connectivity of 6 and the Google 72 qubit Bristlecone device has a maximum connectivity of 4. Embedding a problem into the actual physical connectivity of the machine can be quite a challenging problem. To compensate for the limited connectivity, the D-Wave machines use a technique called chaining while the gate level machines use a technique called SWAP gates. In the AQC presentation, D-Wave’s engineer showed data indicating that the new architecture will decrease chain lengths for typical problems by over 50%. This will result in higher performance and more accurate results. (Note that certain completely different technologies do not have this problem. For example, the IonQ ion trap machine provides All-to-All connectivity between qubits.)
In the AQC presentation, D-Wave also mentioned improved annealing control features including anneal offsets, anneal pause, anneal quench, and reverse anneals. These features were first implemented in the D-Wave 2000Q and will be carried over to the Pegasus machines.
Finally, another feature being implemented in the Pegasus machines is to improve the system overhead to reduce the overall run time. Their goal was to maintain the overall system overhead for the 5640 qubit Pegasus P16 machine at the same levels as the 2048 qubit D-Wave 2000Q.
Although not directly related to the Pegasus engineering developments, D-Wave has continued to make progress in offering their Leap cloud services, improved tools including new capabilities to support hybrid classical/quantum computing, and an expanding customer base that has created over 100 early applications. When D-Wave formally announces customer availability of the Pegasus architecture machines, it should help them continue their momentum.
For additional details you can view D-Wave recent preview press release here, the AQC video mentioned above here, a white paper that describes the topology in more detail here, and a nice GIF picture that demonstrates the metamorphosis from the previous Chimera to the Pegasus topology here.